From chenbc20 at mails.tsinghua.edu.cn Sun Jan 15 18:03:21 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about duplicated variables in statnet Message-ID: <41b025d3.23811.185b8511d80.Coremail.chenbc20@mails.tsinghua.edu.cn> Statnet Developer: Good morning. Thanks for your time and kindness to help look into this help request. I'm now facing a problem about duplicated variables, while I import data into statnet, R is giving me errors about duplicated variables and parallel edges. However from the excel which I input, we can see that there is no such problem. Since that Multiple=T console shouldn't allow density and further analysis ahead, so I have to figure out how to solve this question. I also sent the R console and raw data table screenshot to help understand the question, and hope to hear back soon! thank you very much for your time and hope you have a good day. Chen Bochuan 2023.1.16 -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.27.png Type: image/png Size: 110881 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.42.png Type: image/png Size: 631720 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.56.png Type: image/png Size: 625377 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.56.33.png Type: image/png Size: 32258 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.56.58.png Type: image/png Size: 96606 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.57.16.png Type: image/png Size: 40071 bytes Desc: not available URL: From chenbc20 at mails.tsinghua.edu.cn Sun Jan 15 18:12:41 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about duplicated variables in statnet Message-ID: <66a99d09.23845.185b859a86c.Coremail.chenbc20@mails.tsinghua.edu.cn> Statnet Developer: Good morning. Thanks for your time and kindness to help look into this help request. I'm now facing a problem about duplicated variables, while I import data into statnet, R is giving me errors about duplicated variables and parallel edges. However from the excel which I input, we can see that there is no such problem. Since that Multiple=T console shouldn't allow density and further analysis ahead, so I have to figure out how to solve this question. I also sent the R console and raw data table screenshot to help understand the question, and hope to hear back soon! thank you very much for your time and hope you have a good day. Chen Bochuan 2023.1.16 -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.27.png Type: image/png Size: 110881 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.42.png Type: image/png Size: 631720 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.55.56.png Type: image/png Size: 625377 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.56.33.png Type: image/png Size: 32258 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.56.58.png Type: image/png Size: 96606 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-01-16 09.57.16.png Type: image/png Size: 40071 bytes Desc: not available URL: From chenbc20 at mails.tsinghua.edu.cn Thu Jan 26 01:28:20 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about analyzing directed network Message-ID: <2753b657.68fc7.185ed68198d.Coremail.chenbc20@mails.tsinghua.edu.cn> Dear developer: Thanks for your patience last time to answer my question since it was my first time using the mailing list. I have been reading papers about ERGM since and began to handle my own research in progress. However, when using the GOF command as well as other commands which includes "degree", it responds with error about it couldn't be used for directed network. However, I am dealing with a friendship cognition network while shouldn't be reciprocal, so I am wondering if there is a solution for further estimation of model fitness for directed network? Thanks a lot! Hope you have a great day! BC Chen 2023.1.26 -------------- next part -------------- An HTML attachment was scrubbed... URL: From rgilad at bgu.ac.il Thu Jan 26 03:15:00 2023 From: rgilad at bgu.ac.il (Gilad Ravid) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] network with edge attributes Message-ID: Hi, While trying to read (or create) a network with edge attributes, if there are more than 1000 edges, the attribute is not associated with the edges. for example the line x_1<-read.paj("1.net") works fine, and the edges have the attribute, while the line x_2<-read.paj("1_2.net") does not operate as expected. The only difference between the files is that the second one has one more edge (total of 1001 edges) Thanks, Gilad -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: 1_2.net Type: application/octet-stream Size: 15491 bytes Desc: 1_2.net URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: 1.net Type: application/octet-stream Size: 15483 bytes Desc: 1.net URL: From chenbc20 at mails.tsinghua.edu.cn Tue Mar 7 04:44:31 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about Statnet ergm model collapse Message-ID: <64fae2e8.45515.186bc1a155b.Coremail.chenbc20@mails.tsinghua.edu.cn> Statnet Developer: Good morning. Thank you so much for your patience to look into my question. I am encountering some questions beyond my capability so I have no choice but to turn to you for help. I was using ergm model, especially the dependent model, and my original code, which worked well, looked like "gw1_sup <- ergm(supnet ~ edges + nodefactor("edu") + nodefactor("income") + nodefactor("work") + nodefactor("religious") + nodefactor("gender") + nodecov("age") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = control.ergm(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), eval.loglik = T, verbose = T) " but when I turned the categorical variables "edu" and "income" into continuous variables and turn to "nodecov", like "gw1_sup <- ergm(supnet ~ edges + nodefactor("work") + nodefactor("religious") + nodefactor("gender") + nodecov("edu") + nodecov("income") + nodecov("age") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = control.ergm(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), eval.loglik = T, verbose = T) " the model at first run well, but at around 8/9 iteration, it collapsed and report: "Error in T2nullity && verbose : invalid 'x' type in 'x && y'". I browsed through the internet and some answers said it could be solved by deleting some of the variables, but those variables are important to my research, the only way I can think of is that I could go back to the old way by using nodefatcor for categorical variables. For a more clear code, you can see the question in github issue: https://github.com/statnet/ergm/issues/519 Thanks for your time and kindness!!! Best regards! -------------- next part -------------- An HTML attachment was scrubbed... URL: From fbr33 at cornell.edu Sat Mar 18 12:35:34 2023 From: fbr33 at cornell.edu (F Benjamin Rosche) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] How to replace network on the right-hand side in ergm.simulate? In-Reply-To: References: Message-ID: Dear statnet users, I have a question regarding the simulate.ergm() function. I fitted an ERGM that includes an edge covariate using ergm(g1 ~ edges + edgecov(g2, "X") + ...). To do a counterfactual analysis, I would like to replace g2 with another state of this network. While I understand that I can replace the network on the left-hand side in simulate.ergm() using simulate.ergm(ergm.fit, basis=g1*), I don't know how to replace the network on the right-hand side g2 with g2*. I have noticed that simulate.ergm() does not require g2 to be in the R environment - g2 must thus be stored in ergm.fit but the network stored in ergm.fit$network seems to be the network on the left-hand side. Thank you for any pointers that you can provide. Best wishes, Ben -- *Benjamin Rosche* Cornell University / Social Dynamics Lab / benrosche.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Sat Mar 18 20:18:41 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] How to replace network on the right-hand side in ergm.simulate? In-Reply-To: References: Message-ID: <6b8dc62e-8ced-9d97-7266-6439e44a0642@uci.edu> Hi, Ben - For edgecov, you may find it easier to simply pass a matrix as a predictor.? But more generally, the most flexible thing to do is always to simply call the simulate command with a formula implementing the model in lieu of the fitted ergm() object.? That way you can change whatever you like, on the fly, without having to worry about updating whatever was encoded in the ergm() object. So, e.g., instead of mymodel <- ergm(myobsnet ~ edges + edgecov("cardboard") + sunspots) sim <- simulate(mymodel) you'd have an expression like mymodel <- ergm(myobsnet ~ edges + edgecov("cardboard") + sunspots) sim <- simulate(mynewstartnet ~ edges + edgecov("newcardboard") + sunspots, coef=coef(mymodel)) where "newcardboard" is whatever you want your covariate to be (could be the old "cardboard" - reduce, reuse, recycle and all that), and "mynewstartnet" could be either the network you used to fit the graph or something else.? For instance, in settings where independence of draws is critical, you might initialize the simulation with e.g. a Bernoulli graph (produced by rgraph()) with the right target density, and take a single draw from each chain (being sure to set the burn-in to be long enough for good convergence).? And of course, it is trivial to use the above as a starting point for adding additional terms to a fitted model (e.g., for robustness tests). Another occasionally useful bit of Dark Arts is to coerce your ERGM formula to a string, manipulate the string, and then coerce the whole thing back to a formula (thence to use with simulate() or ergm()).? I've used that trick for many things, including to facilitate automated model comparison/selection.? R is very flexible about computing on the language, which one can use to one's advantage. Hope that helps, -Carter On 3/18/23 12:35 PM, F Benjamin Rosche wrote: > Dear statnet users, > > I have a question regarding the simulate.ergm() function. I fitted an > ERGM that includes an edge covariate using ergm(g1 ~ edges?+ > edgecov(g2, "X")?+ ...). > > To do a counterfactual analysis, I would like to replace g2 with > another state of this network. While I understand that I can replace > the network on the left-hand side in simulate.ergm() using > simulate.ergm(ergm.fit, basis=g1*), I don't know how to replace the > network on the right-hand side g2 with g2*. > > I have noticed that simulate.ergm() does not require g2 to be in the R > environment - g2 must thus be stored in ergm.fit but the network > stored in ergm.fit$network seems to be the network on the left-hand side. > > Thank you for any pointers that you can provide. > > Best wishes, > > Ben > -- > *Benjamin Rosche* > Cornell University / Social Dynamics Lab?/ benrosche.com > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!NVpfvHx6Ve7mI4BS6f0bokRWD37KA-KYGUMDezJQPYwj8vvcsabCBGEUb8kCuHIPuNZ8nM6mxrbtETI$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From fbr33 at cornell.edu Tue Mar 21 07:17:33 2023 From: fbr33 at cornell.edu (F Benjamin Rosche) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Degree constraint warning Message-ID: Dear statnet users, I have a question regarding degree constraints in an ergm. I am using constraints=~bd(minout=c(...)) to constrain the minimal outdegree of each node to a specific value. When I run the model, I get the following warning and it is not clear to me whether I can safely ignore this warning or whether I did not implement the constraint correctly: Warning: In is.na(a$minout) && is.na(a$minin): 'length(x) = 16 > 1' in coercion to 'logical(1)' Here is a reproducible example: library(statnet) data("florentine") ergm(flomarriage~edges, constraints=~bd(minout=c(0,0,0,0,2,1,3,0,0,0,0,0,0,0,0,0))) Thank you very much for any pointers you can provide! Ben -- Benjamin Rosche / Social Dynamics Lab / benrosche.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From chenbc20 at mails.tsinghua.edu.cn Sun Apr 9 06:30:58 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about Model Stuck in MCMLE Message-ID: <5b56edbe.af91f.1876636770d.Coremail.chenbc20@mails.tsinghua.edu.cn> Statnet Developer: Good morning. Thank you so much for your patience to look into my question. I am encountering some questions beyond my capability so I have no choice but to turn to you for help. I am encountering the problem of MCMLE stuck. My original code looked like gw1_help <- ergm(helpnet ~ edges + nodefactor("work") + nodefactor("income") + nodefactor("edu") + nodefactor("religious") + nodefactor("gender") + nodecov("age") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = control.ergm(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), eval.loglik = T, verbose = T) which worked qiuite well. However, once I add 'edgecov(sup_loca2_net, "distance") ' in, which looked like: gw1_help <- ergm(helpnet ~ edges + nodefactor("work") + nodefactor("income") + nodefactor("edu") + nodefactor("religious") + nodefactor("gender") + nodecov("age") + edgecov(sup_loca2_net, "distance") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = control.ergm(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), eval.loglik = T, verbose = T) the model stuck and said "MCMLE estimation stuck. There may be excessive correlation between model terms, suggesting a poor model for the observed data. If target.stats are specified, try increasing SAN parameter." I don't know what happened and I desperately want to know how to solve this problem, Thanks for your time and patience! Hope to hear from you soon! -------------- next part -------------- An HTML attachment was scrubbed... URL: From handcock at ucla.edu Sun Apr 9 08:25:25 2023 From: handcock at ucla.edu (Mark S. Handcock) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about Model Stuck in MCMLE In-Reply-To: <5b56edbe.af91f.1876636770d.Coremail.chenbc20@mails.tsinghua.edu.cn> References: <5b56edbe.af91f.1876636770d.Coremail.chenbc20@mails.tsinghua.edu.cn> Message-ID: <39ecf3be-49ba-8b3d-a0ac-39ebd64305db@stat.ucla.edu> An HTML attachment was scrubbed... URL: From buttsc at uci.edu Sun Apr 9 12:38:34 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about Model Stuck in MCMLE In-Reply-To: <5b56edbe.af91f.1876636770d.Coremail.chenbc20@mails.tsinghua.edu.cn> References: <5b56edbe.af91f.1876636770d.Coremail.chenbc20@mails.tsinghua.edu.cn> Message-ID: <4f515507-270e-125f-d1f0-4bd8675b31c7@uci.edu> If you are using distance as a predictor, you are positing that tie probability falls exponentially (approximately) in distance; this is probably too sharp an SIF. Try using the log distance instead. -Carter On 4/9/23 6:30 AM, ??? wrote: > Statnet Developer: > Good morning.?Thank you so much for your patience to look into my > question. I am encountering some questions beyond my capability so I > have no choice but to turn to you for help. > > I am encountering the problem of MCMLE stuck. My original code looked like > > gw1_help <- ergm(helpnet ~ edges + nodefactor("work") + > nodefactor("income") + nodefactor("edu") + nodefactor("religious") + > nodefactor("gender") + nodecov("age") + edgecov(kin_sup,"kinweight") + > gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = > control.ergm(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval > = 1000, seed = 567), eval.loglik = T, verbose = T) > > which worked qiuite well. However, once I add 'edgecov(sup_loca2_net, > "distance")?' in, which looked like: > > gw1_help <- ergm(helpnet ~ edges + nodefactor("work") + > nodefactor("income") + nodefactor("edu") + nodefactor("religious") + > nodefactor("gender") + nodecov("age") + edgecov(sup_loca2_net, > "distance") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + > gwesp(.1, T) + gwdsp(.1, T), control = control.ergm(MCMC.samplesize = > 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), > eval.loglik = T, verbose = T) > > the model stuck and said "*_MCMLE estimation stuck. There may be > excessive correlation between model terms, suggesting a poor model for > the observed data. If target.stats are specified, try increasing SAN > _**param*eter." > > I don't know what happened and I desperately want to know how to solve > this problem, Thanks for your time and patience! Hope to hear from you > soon! > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!MXKxthAW4qsCYVUIoNakj72_N0d2pl2p1VQqNWKwlCEVOKwOY_y_R8rqy2SuSzxRcojQkfOTzeaCjimCM1NbTpS-EGs$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From chenbc20 at mails.tsinghua.edu.cn Mon Apr 10 09:52:45 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Model Stuck Without Reason Message-ID: <7deb9b9b.38ea1.1876c1592a2.Coremail.chenbc20@mails.tsinghua.edu.cn> Dear Statnet developer: Thanks for your time! I am encountering a problem of my model, which yesterday Prof. Handcock told me to replace ergm with ergm.tepered: "gw1_help <- ergm.tapered(supnet ~ edges + nodefactor("work")+ nodefactor("income")+ nodefactor("edu")+ nodefactor("religious")+ nodefactor("gender")+ nodecov("age")+ edgecov(sup_loca2_net,"distance")+ edgecov(kin_sup,"kinweight")+ gwidegree(.1,T)+ gwesp(.1,T)+ gwdsp(.1,T), control = control.ergm.tapered(MCMC.samplesize =1e+5, MCMC.burnin =1e+6, MCMC.interval =1000, seed =567), eval.loglik =T, verbose =T)" I couldn't run this model on my macbookpro ( intel 4-core 16ram), so I put this code on an online platform, which is Chinese version of Colab. I used 8-core 32G CPU to run my code, and it stuck at 48th iteration. I put the plot of the code and the plot of CPU and RAM in the appendix, and I don't know what I should do to make it work. My network isn't very large, which nodes are around 300 and edges are less than 1000. Thanks!!! -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-04-11 00.52.05.png Type: image/png Size: 245890 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-04-11 00.52.23.png Type: image/png Size: 274565 bytes Desc: not available URL: From chenbc20 at mails.tsinghua.edu.cn Mon Apr 10 10:18:54 2023 From: chenbc20 at mails.tsinghua.edu.cn (=?UTF-8?B?6ZmI5Y2a5bed?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] MCMC sampling did not mix at all Message-ID: <73e0940f.38eb8.1876c2d80e5.Coremail.chenbc20@mails.tsinghua.edu.cn> Dear Statnet developer: Thanks for your time! I am encountering a problem of my model, which yesterday Prof. Handcock told me to replace ergm with ergm.tepered: "gw1_help <- ergm.tapered(supnet ~ edges + nodefactor("work") + nodefactor("income") + nodefactor("edu") + nodefactor("religious") + nodefactor("gender") + nodecov("age") + edgecov(sup_loca2_net,"distance") + edgecov(kin_sup,"kinweight") + gwidegree(.1, T) + gwesp(.1, T) + gwdsp(.1, T), control = control.ergm.tapered(MCMC.samplesize = 1e+5, MCMC.burnin = 1e+6, MCMC.interval = 1000, seed = 567), eval.loglik = T, verbose = T)" I couldn't run this model on my macbookpro ( intel 4-core 16ram), so I put this code on an online platform, which is Chinese version of Colab. I used 8-core 32G CPU to run my code, and it stuck at 48th iteration. I don't know what I should do to make it work. My network isn't very large, which nodes are around 300 and edges are less than 1000. After I wrote the first email, the model reported the error says: "Unconstrained MCMC sampling did not mix at all. Optimization cannot continue" and I also put the screenshot in the appendix. Please help me out with this problem! I have been suffered soooo much with this model issue. Thank you so much! -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: ??2023-04-11 01.17.03.png Type: image/png Size: 224132 bytes Desc: not available URL: From fbr33 at cornell.edu Mon Apr 10 14:30:01 2023 From: fbr33 at cornell.edu (F Benjamin Rosche) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Constraints: blockdiag() and odegrees Message-ID: Dear statnet users, I have a question regarding constraints in an ergm. To estimate one model across a few networks (indexed by nid), I am using: ergm(...., constraints=blockdiag("nid")) to reflect that nodes in different networks cannot form edges between each other. For various reasons, I have to add another constraint: *+odegrees, *which is currently not possible in combination with blockdiag(). I tried estimating separate models by network instead of using blockdiag() so that I can add odegrees. However, because I have relatively many parameters for relatively few observations per network, estimating just one model across networks leads to much more stable estimates. Is there any other strategy that I could take to estimate just one model but enforce that nodes in different networks cannot form edges between each other and have the same outdegree as the observed network? I am grateful for any pointers you can provide. Thank you, Ben -- *Benjamin Rosche * Doctoral student / Social Dynamics Lab / benrosche.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From linzhu1 at stanford.edu Fri Apr 14 00:13:32 2023 From: linzhu1 at stanford.edu (Lin Zhu) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] GWESP drops overtime using STERGM fitted with edges dissolution approximation method Message-ID: Dear Statnet group, Hope you are all doing well. As part of our modeling work, we are struggling to fit a STERGM to cross-sectional + duration data that includes a GWESP target. We originally used the edges dissolution approximation method, and see GWESP drop precipitously in simulated networks over time (we are using monthly time steps). We have not been able to get the model to finish running using the full STERGM (without the approximation). Increasing simulation time horizon by using a smaller timestep (eg., daily) results in an improved fit using the approximation method, but this approach would present computational challenges if we needed to run the simulation at daily timesteps. Do you have suggestions on how we could address this issue (maybe changes to settings that would help the full STERGM to finish, or workarounds for the approximation)? Thank you so much. Best, Lin Lin Zhu | Research Engineer Department of Health Policy | School of Medicine Center for Health Policy | Freeman Spogli Institute for International Studies Stanford University 615 Crothers Way Encina Commons, Office 113 Stanford, CA 94305-6006 650.736.8139 | linzhu1@stanford.edu [signature_670862627] -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: image001.png Type: image/png Size: 6796 bytes Desc: image001.png URL: From terin.mayer at gmail.com Wed Apr 19 08:03:19 2023 From: terin.mayer at gmail.com (Terin Mayer) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Specifying TERGM constraint on bipartite network In-Reply-To: References: Message-ID: <58B3F77F-93F8-4295-A5E5-4C61819AB23E@gmail.com> Hello Statnet Community? Greetings, and with much appreciation for all your work on this incredible collection of open-source resources! I am writing to request information and advice for estimating a TERGM on a bipartite network with a dyad-independent constraint on the sample space. I am attempting to fit a TERGM to a bipartite network of local units of government participating in watershed management organizations. I observe this network over a several decade time-span. During that range, some watershed management organizations (mode 2 nodes) are created and some disband. There are geographic constraints on the possible links, which I have represented as an edge-list and passed to the "fixallbut()? constraint operator. I am encountering some difficulties in getting the model to run. In certain circumstances initial MPLE estimation doesn?t complete (?Error in set.constr.type(lprec, rep(?>=?, NROW(X.bar))): ?types? and ?constraints? are not the same length.) In other attempts, the MPLE estimation appears to get stuck on ?evaluating the predictor and response matrix". In all cases, I get the warning ?Active vertex set varies from time point to time point. Estimation may not work.? Being new at this, I?m having some difficulty discerning when I should keep debugging vs. when I need to reconceptualize my data to fit existing package limitations. Specifically, my questions are: - Should I anticipate that fixing the vertex set will improve my chances of successful modeling? (i.e. not representing creation/destruction of the second mode in the bipartite scheme and thus not having changes in the active vertex set) - What is the correct way to specify the ?fixallbut(free.dyads)? constraint with a bipartite network? Is there a way of using permanent ids to ensure that the edgelist (which I build using node names) correctly picks out the vertexes in each network panel? I have refrained from providing a reproducible example, but would readily do so if that would help clarify my query. Thanks kindly for any help you can provide! Terin Terin Mayer PhD Candidate in Public Affairs, University of Minnesota UC Berkeley MPP ?18 terin.mayer@gmail.com terin@umn.edu | 612-812-0710 terinvmayer.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From m.h.a.p.janssen at tilburguniversity.edu Mon May 8 02:27:19 2023 From: m.h.a.p.janssen at tilburguniversity.edu (Mathieu Janssen) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm.ego with ego-ego ties Message-ID: Dear statnetters, My name is Mathieu, and I am a data science master's student at Tilburg University. For a research project I'm doing, I have constructed a network with Spotify data that focuses on artists' collaborations. I have two harvests of data. 1) specific artists and their collaborators, 2) info on all the artists and their collaborators. The data structure is ego nets. I have a few questions about the *ergm.ego *package. Thanks in advance to anyone that will spend a few minutes to help me out. My edge list looks like that: edgelist <- data.frame(c("ego1", "ego2", "ego3","ego4","ego5","ego5"), c("alter1", "alter2", "alter3", "alter2", "ego2", "alter1")) In plain English, I do not have alter-alter ties, but I do have ego-ego ties. Question 1: is ergm.ego the right package for that? If not, can you recommend a different one that would allow me to run ERGMs on this data structure? Or would you please suggest how to constrain a regular ERGM? If ergm.ego is the right one, is this the right way to import the data? library(ergm.ego) temp_n <- network::as.network.data.frame(edgelist, directed = FALSE) temp_n <- network::set.vertex.attribute(temp_n, "is_emerging", value = c(TRUE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE, FALSE)) temp_n.ego <- as.egor(temp_n, ego_name = c("is_emerging")) Using the as.egor() function, the result is not what I expect. Exactly those nodes with is_emerging == True should be egos, the rest alters. When the as.egor() function is called, some of the alters are included as egos, which is not desired. Therefore, I adjust the underlying Tibbles myself. old_egos <- as.data.frame(temp_n.ego$ego) new_egos <- old_egos[old_egos$is_emerging == TRUE,] new_egos <- as_tibble(new_egos) old_alters <- as.data.frame(temp_n.ego$alter) new_alters <- old_alters[old_alters$.egoID %in% new_egos$.egoID,] new_alters <- as_tibble(new_alters) artists.ego <- egor(egos = new_egos, new_alters, ID.vars = list(ego = ".egoID", alter = ".altID")) Question 2: is this the correct way to import this data type? As you can see, we have no data on the Alter-Alter ties, so they are left out of the final function call. Many thanks, Mathieu Janssen Jheronimus Academy of Data Science / Tilburg University A: JADS, Sint Janssingel 92, 5211 DA 's-Hertogenbosch, The Netherlands E: Mathieu-janssen@live.nl / m.h.a.p.janssen@tilburguniversity.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Tue May 9 02:51:02 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm.ego with ego-ego ties In-Reply-To: References: Message-ID: Dear Mathieu, There are a couple of issues here: 1. Whether ergm.ego is an appropriate inference approach depends on how you selected the artists (the "egos") and what is your target of inference. In other words, how are you willing to define the "population network" in the framework of egocentric ERGMs 2. From a purely technical perspective, for ergm.ego you need to construct an egor object using the 'egor' package. There is more than one way of doing that, please have a look at the documentation. 3. The implementation of ergm.ego essentially ignores the fact if two ego-alter ties are in fact the same tie in the population (i.e the two ego-networks overlap). If it happens only rarely and can be ignored vis a vis the sample size and population size, you should be good to go. If, on the other hand, it is a feature of the way you selected the egos (for example by a procedure resembling a snowball sample) then ergm.ego will not be appropriate. hope this helps Michal On Mon, May 8, 2023 at 11:28?AM Mathieu Janssen wrote: > > Dear statnetters, > > > > My name is Mathieu, and I am a data science master's student at Tilburg University. For a research project I'm doing, I have constructed a network with Spotify data that focuses on artists' collaborations. > > I have two harvests of data. 1) specific artists and their collaborators, 2) info on all the artists and their collaborators. The data structure is ego nets. > > I have a few questions about the ergm.ego package. Thanks in advance to anyone that will spend a few minutes to help me out. > > > > My edge list looks like that: > > > > edgelist <- data.frame(c("ego1", "ego2", "ego3","ego4","ego5","ego5"), c("alter1", "alter2", "alter3", "alter2", "ego2", "alter1")) > > > > In plain English, I do not have alter-alter ties, but I do have ego-ego ties. > > > > Question 1: is ergm.ego the right package for that? If not, can you recommend a different one that would allow me to run ERGMs on this data structure? Or would you please suggest how to constrain a regular ERGM? > > > > If ergm.ego is the right one, is this the right way to import the data? > > > > library(ergm.ego) > > temp_n <- network::as.network.data.frame(edgelist, directed = FALSE) > > temp_n <- network::set.vertex.attribute(temp_n, "is_emerging", value = c(TRUE, TRUE, TRUE, TRUE, TRUE, FALSE, FALSE, FALSE)) > > temp_n.ego <- as.egor(temp_n, ego_name = c("is_emerging")) > > > > Using the as.egor() function, the result is not what I expect. Exactly those nodes with is_emerging == True should be egos, the rest alters. When the as.egor() function is called, some of the alters are included as egos, which is not desired. Therefore, I adjust the underlying Tibbles myself. > > > > old_egos <- as.data.frame(temp_n.ego$ego) > > new_egos <- old_egos[old_egos$is_emerging == TRUE,] > > new_egos <- as_tibble(new_egos) > > > > old_alters <- as.data.frame(temp_n.ego$alter) > > new_alters <- old_alters[old_alters$.egoID %in% new_egos$.egoID,] > > new_alters <- as_tibble(new_alters) > > > > artists.ego <- egor(egos = new_egos, new_alters, ID.vars = list(ego = ".egoID", alter = ".altID")) > > > > Question 2: is this the correct way to import this data type? As you can see, we have no data on the Alter-Alter ties, so they are left out of the final function call. > > > > Many thanks, > > > > Mathieu Janssen > > Jheronimus Academy of Data Science / Tilburg University > > A: JADS, Sint Janssingel 92, 5211 DA 's-Hertogenbosch, The Netherlands > > E: Mathieu-janssen@live.nl / m.h.a.p.janssen@tilburguniversity.edu > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help From Jacob.Young.1 at asu.edu Sat Jul 29 11:37:03 2023 From: Jacob.Young.1 at asu.edu (Jacob Young) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about using constraints in ergm Message-ID: Hi all, I am estimating a model to assess whether the effect of a variable differs across two networks. Specifically, whether a nodeifactor() term for a variable ?white? (a binary White/Non-white node covariate). Here is a simple model: ergm( mynet ~ edges + mutual + nodeifactor( "white" ) # here is the model over both networks + S( ~ edges + mutual + nodeifactor( "white" ) ~ ( block == 1 ), # here is the constraints over the first network (i.e. block == 1) constraints = ~ blocks( "block", levels2 = c( 2, 3 ) ) ) My question is this: when I estimate the subgraph portion with S(, do I need to include the edges and mutual terms if I am only interested in the nodeifactor( "white" ) term? Please let me know if this question makes sense. Thanks! Jacob Young, PhD Associate Professor | School of Criminology and Criminal Justice Director of Curriculum | Institute for Social Science Research Watts College of Public Service & Community Solutions Arizona State University 411 N. Central Ave., Ste. 600 Phoenix, AZ 85004 -------------- next part -------------- An HTML attachment was scrubbed... URL: From chenshuohong at umass.edu Fri Aug 4 07:30:00 2023 From: chenshuohong at umass.edu (Chen-Shuo Hong) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] weird number of edges when using simulate Message-ID: Dear statnet users, I have a question regarding the simulate function for tergm. I fitted a tergm using a longitudinal network (2 waves). The models used both Form() and Persist() with CMLE, and include exogenous (e.g., nodematch) and endogenous terms (e.g., transitvieties, mutual, dgwdsp). The diagnosis showed the model fits the data pretty well. To obtain simulated wave 2 networks, I am using: ans1tm1_sim_first <- simulate(..., nsim = 1, nw.start = "first", output="final") However, the resulting simulated wave 2 networks produced more isolates and far less edges than observed wave 2 networks. I'm not entirely sure if this is because there is an unexpected bug in this function or if because of any other issues. I also explicitly specified parameters in simulate() and closed one at a time to test which parameter causes this issue. It turned out that including dgwdsp seems to play a non-trivial role in reducing the number of edges. When dgwdsp was retained, the number of edges became larger than the observed and the simulated network was quite compact. I am grateful for any insights you can provide. Thank you, Chen-Shuo -- Chen-Shuo Hong PhD Candidate in Sociology University of Massachusetts - Amherst chenshuohong@umass.edu -------------- next part -------------- An HTML attachment was scrubbed... URL: From vassey at usc.edu Wed Aug 9 03:59:32 2023 From: vassey at usc.edu (Julia Vassey) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: Dear All, This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) Thank you, -- Julia Vassey Health Behavior Research Department of Population and Public Health Sciences Keck School of Medicine University of Southern California -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Fri Aug 11 05:52:50 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: Hi Julia, Can you send the GOF plots attached? In what way does the ESP not fit well? Perhaps it is a matter of changing the decay parameter? Michal On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: > > > Dear All, > > This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. > > I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. > > I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. > > model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) > > Thank you, > > > -- > Julia Vassey > Health Behavior Research > Department of Population and Public Health Sciences > Keck School of Medicine > University of Southern California > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help From vassey at usc.edu Fri Aug 11 07:50:59 2023 From: vassey at usc.edu (Julia Vassey) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: Thank you Michal! Good to hear from you. Please, see the plots, attached. I have tried changing the decay parameter multiple times, and the model fit gets better when increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), but beyond 0.3 the model starts having issues with convergence. p values for all esps are always very low. Thank you for helping! Julia On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: > > Hi Julia, > > Can you send the GOF plots attached? In what way does the ESP not fit > well? Perhaps it is a matter of changing the decay parameter? > > Michal > > On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: > > > > > > Dear All, > > > > This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. > > > > I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. > > > > I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. > > > > model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) > > > > Thank you, > > > > > > -- > > Julia Vassey > > Health Behavior Research > > Department of Population and Public Health Sciences > > Keck School of Medicine > > University of Southern California > > > > _______________________________________________ > > statnet_help mailing list > > statnet_help@u.washington.edu > > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ -------------- next part -------------- A non-text attachment was scrubbed... Name: ergmmodel-gofplots.zip Type: application/zip Size: 51583 bytes Desc: not available URL: From michal2992 at gmail.com Sun Aug 13 06:36:14 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: Hi Julia, Thanks. Some questions and hypotheses about model specification: 1. The network is directed. Are you sure you want nodefactor (nodes in some groups have more ties than nodes in other groups irrespectively of tie direction) rather than nodeofactor (same but for outgoing ties only) or nodeifactor (same but for incoming ties only)? 2. Your model does not have any nodematch terms on the nodal attributes. Is that intentional? 3. The ESP distribution is notably relatively flat. One mechanism that might be responsible for that is strong community structure or homophily on some observed attribute -- triangles get closed within groups only and the count of ESP depends on group size. Do you have several cohesive groups of varying size in the network? m. On Fri, Aug 11, 2023 at 4:51?PM Julia Vassey wrote: > > Thank you Michal! Good to hear from you. > > Please, see the plots, attached. I have tried changing the decay > parameter multiple times, and the model fit gets better when > increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), > but beyond 0.3 the model starts having issues with convergence. > p values for all esps are always very low. > > Thank you for helping! > > Julia > > > On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: > > > > Hi Julia, > > > > Can you send the GOF plots attached? In what way does the ESP not fit > > well? Perhaps it is a matter of changing the decay parameter? > > > > Michal > > > > On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: > > > > > > > > > Dear All, > > > > > > This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. > > > > > > I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. > > > > > > I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. > > > > > > model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) > > > > > > Thank you, > > > > > > > > > -- > > > Julia Vassey > > > Health Behavior Research > > > Department of Population and Public Health Sciences > > > Keck School of Medicine > > > University of Southern California > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ From buttsc at uci.edu Sun Aug 13 13:38:51 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: <379e9e8d-333b-1d7f-e5a1-3bd05b307730@uci.edu> Hi, Julia - To add to Michal's excellent observations: ? - You seem to have a lot of out-isolates and in-isolates that you are not catching.? (They may be isolates per se - both out and in - but we can't tell from these plots.)?? Not accounting for that can lead to edges being too uniformly dispersed through the network.? You may thus want to consider idegree(0), odegree(0), and/or isolates() terms, if there is not a covariate that accounts for the structure. ? - As Michal observes, you clearly have some very locally dense clusters, and it is unlikely that you will capture them using a homogeneous clustering term - while exceptions exist, these are usually associated with inhomogeneity, generally clustering around a specific attribute, affiliation, etc.? His suggestions for dealing with that are very apposite.? But an even more basic suggestion is to do more exploratory investigation of your network.? Have you tried plotting the sociogram, and coloring the nodes by their attributes?? If the graph is too dense to easily see what is going on, the following can help: ???? - Increase the translucency of your edges.? This One Weird Trick is very simple, but very useful - if the graph is dense, don't be afraid to make the edges so transparent that they are barely visible!? When combined with varying the shape/color of nodes by attribute, this can be very revealing of internal structure, and particularly of where your clustering is coming from.? The edge.col option in plot.network or sna's gplot function can be used for this quite easily. ???? - Take the graph apart, using degree kcores.? The kcores() command in sna will give you the core number for every vertex in the network, using your favorite centrality measure; the default is Freeman degree, which is probably a good starting point for your network.? Try sequentially plotting the induced subgraphs for each core (i.e., everything in the 1 core, the 2 core, the 3 core, etc.), again coloring the nodes by their attributes.? (Page 24, figure 4 of this paper has an example of this approach: https://sciendo.com/article/10.21307/joss-2019-027)? What this will do is strip away the less cohesive parts of the network, revealing highly cohesive subgraphs that may otherwise be obscured by the surrounding structure.? If you look at figure 8 in that same paper, you'll see an example where that is used to reveal the presence of several highly cohesive, homophilous clusters that are not evident if you look at the whole graph (which is both very large and adorned with pendant trees).? Knowing the composition of the "deep" portions of the network will often give you strong indications of what is shaping it, and of what terms you will need in your model. ???? - Plot distributions of degree and other structural indices by attribute.? Simply looking at these distributions can be revealing - in addition to in/outdegree heterogeneity, examining e.g. cycle or clique membership can be another way to detect heterogeneity in local structure, all of which can inform your choice of graph statistics.? (Beyond nodal covariate effects and terms like nodematch/nodemix, note that inhomogeneous versions of e.g. degree and gwdegree exist, and can be useful when you have groups that interact in very different ways.? Even localtriangle can sometimes be helpful, if you have very small groups whose patterns of cohesion cannot be captured with mixing terms.) Those are just a few examples, but the general suggestion is to do more exploratory analysis to try to understand the network and its structure (and build your model around that), rather than trying to rely entirely on gof plots to infer what is going on. The latter does sometimes work, especially if the network is very simple, but complex cases usually require a combination of substantive insight (i.e., thinking about what the network is, where it came from, how it was measured, etc., and about how those things would be expected to shape the observed structure) and exploratory analysis (to reveal major quirks and inhomogeneities, provide clues about good term candidates, etc.).? Another benefit of putting more time into these pre-modeling efforts is that they usually give you a much better sense of what is interesting/important about the structure, which can greatly improve your model assessment strategy (by giving you more refined targets) and help you use and interpret the resulting model more effectively.? I know that, in my own work, I've often (almost always?) discovered that the things that turned out to be interesting/important about a network were not the ones that I expected, and that building effective models has really depended on developing those insights.? Fortunately, we have a very rich array of tools and concepts from classical social network analysis that can help us here!? When one's modeling efforts are stuck, I find it can often be useful to go back to basics in order to determine what to try next. Hope that helps, -Carter On 8/11/23 7:50 AM, Julia Vassey wrote: > Thank you Michal! Good to hear from you. > > Please, see the plots, attached. I have tried changing the decay > parameter multiple times, and the model fit gets better when > increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), > but beyond 0.3 the model starts having issues with convergence. > p values for all esps are always very low. > > Thank you for helping! > > Julia > > > On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: >> Hi Julia, >> >> Can you send the GOF plots attached? In what way does the ESP not fit >> well? Perhaps it is a matter of changing the decay parameter? >> >> Michal >> >> On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: >>> >>> Dear All, >>> >>> This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. >>> >>> I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. >>> >>> I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. >>> >>> model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) >>> >>> Thank you, >>> >>> >>> -- >>> Julia Vassey >>> Health Behavior Research >>> Department of Population and Public Health Sciences >>> Keck School of Medicine >>> University of Southern California >>> >>> _______________________________________________ >>> statnet_help mailing list >>> statnet_help@u.washington.edu >>> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ >>> >>> _______________________________________________ >>> statnet_help mailing list >>> statnet_help@u.washington.edu >>> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KFP_Dnvj4xmkPFViPcgvdRkqP6Ybm_N4eB9hriyYxxQhOOOtvoWrmXZPnl_zcUn14YpyXvo3NgopUw$ From vassey at usc.edu Sun Aug 13 13:56:16 2023 From: vassey at usc.edu (Julia Vassey) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: Thank you Michal and Carter! Appreciate all the suggestions. I do have cohesive groups, mostly by geographic regions and they are represented by color in this attached plot. I have not yet tried nodeifactor or nodeofactor, but I will. On Sun, Aug 13, 2023 at 9:43?AM Micha? Bojanowski wrote: > > Hi Julia, > > Thanks. Some questions and hypotheses about model specification: > > 1. The network is directed. Are you sure you want nodefactor (nodes in > some groups have more ties than nodes in other groups irrespectively > of tie direction) rather than nodeofactor (same but for outgoing ties > only) or nodeifactor (same but for incoming ties only)? > > 2. Your model does not have any nodematch terms on the nodal > attributes. Is that intentional? > > 3. The ESP distribution is notably relatively flat. One mechanism that > might be responsible for that is strong community structure or > homophily on some observed attribute -- triangles get closed within > groups only and the count of ESP depends on group size. Do you have > several cohesive groups of varying size in the network? > > m. > > On Fri, Aug 11, 2023 at 4:51?PM Julia Vassey wrote: > > > > Thank you Michal! Good to hear from you. > > > > Please, see the plots, attached. I have tried changing the decay > > parameter multiple times, and the model fit gets better when > > increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), > > but beyond 0.3 the model starts having issues with convergence. > > p values for all esps are always very low. > > > > Thank you for helping! > > > > Julia > > > > > > On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: > > > > > > Hi Julia, > > > > > > Can you send the GOF plots attached? In what way does the ESP not fit > > > well? Perhaps it is a matter of changing the decay parameter? > > > > > > Michal > > > > > > On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: > > > > > > > > > > > > Dear All, > > > > > > > > This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. > > > > > > > > I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. > > > > > > > > I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. > > > > > > > > model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) > > > > > > > > Thank you, > > > > > > > > > > > > -- > > > > Julia Vassey > > > > Health Behavior Research > > > > Department of Population and Public Health Sciences > > > > Keck School of Medicine > > > > University of Southern California > > > > > > > > _______________________________________________ > > > > statnet_help mailing list > > > > statnet_help@u.washington.edu > > > > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ -------------- next part -------------- A non-text attachment was scrubbed... Name: Screenshot 2023-08-13 at 4.48.48 PM.png Type: image/png Size: 165993 bytes Desc: not available URL: From buttsc at uci.edu Sun Aug 13 15:20:33 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: Message-ID: <34501f68-7cd4-3de5-5e4d-221fc85aee4b@uci.edu> Hi, Julia - Looking at your figure, you'll definitely want mixing terms (nodematch, nodemix), because your clustering is very strongly attribute-based.? But I also see that you have a situation where not /all/ nodes are participating in those subgroup interactions, so one key to getting this model to work is likely to be capturing the combination of the homophily/differential mixing among the nodes that do interact, and heterogeneity in propensity to interact at all. nodecov/nodefactor and degree terms can be useful for the latter, and you may indeed need an isolates term.? In any event, the figure you sent already provides a lot of helpful clues for modeling, so keep up with the visualizations and you are likely to find your way! Hope that helps, -Carter On 8/13/23 1:56 PM, Julia Vassey wrote: > Thank you Michal and Carter! Appreciate all the suggestions. I do have > cohesive groups, mostly by geographic regions and they are represented > by color in this attached plot. I have not yet tried nodeifactor or > nodeofactor, but I will. > > On Sun, Aug 13, 2023 at 9:43?AM Micha? Bojanowski wrote: >> Hi Julia, >> >> Thanks. Some questions and hypotheses about model specification: >> >> 1. The network is directed. Are you sure you want nodefactor (nodes in >> some groups have more ties than nodes in other groups irrespectively >> of tie direction) rather than nodeofactor (same but for outgoing ties >> only) or nodeifactor (same but for incoming ties only)? >> >> 2. Your model does not have any nodematch terms on the nodal >> attributes. Is that intentional? >> >> 3. The ESP distribution is notably relatively flat. One mechanism that >> might be responsible for that is strong community structure or >> homophily on some observed attribute -- triangles get closed within >> groups only and the count of ESP depends on group size. Do you have >> several cohesive groups of varying size in the network? >> >> m. >> >> On Fri, Aug 11, 2023 at 4:51?PM Julia Vassey wrote: >>> Thank you Michal! Good to hear from you. >>> >>> Please, see the plots, attached. I have tried changing the decay >>> parameter multiple times, and the model fit gets better when >>> increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), >>> but beyond 0.3 the model starts having issues with convergence. >>> p values for all esps are always very low. >>> >>> Thank you for helping! >>> >>> Julia >>> >>> >>> On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: >>>> Hi Julia, >>>> >>>> Can you send the GOF plots attached? In what way does the ESP not fit >>>> well? Perhaps it is a matter of changing the decay parameter? >>>> >>>> Michal >>>> >>>> On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: >>>>> >>>>> Dear All, >>>>> >>>>> This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. >>>>> >>>>> I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. >>>>> >>>>> I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. >>>>> >>>>> model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) >>>>> >>>>> Thank you, >>>>> >>>>> >>>>> -- >>>>> Julia Vassey >>>>> Health Behavior Research >>>>> Department of Population and Public Health Sciences >>>>> Keck School of Medicine >>>>> University of Southern California >>>>> >>>>> _______________________________________________ >>>>> statnet_help mailing list >>>>> statnet_help@u.washington.edu >>>>> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ >>>>> >>>>> _______________________________________________ >>>>> statnet_help mailing list >>>>> statnet_help@u.washington.edu >>>>> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Ie0ff82n4kWRW0ahOT_x2o11HXJ_IOxUBaL3xWEkPBpEcCE2rhK6tYDMmOZy_W2ko3PJ6Lx_6fQd2g$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From vassey at usc.edu Sun Aug 13 15:26:37 2023 From: vassey at usc.edu (Julia Vassey) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: <34501f68-7cd4-3de5-5e4d-221fc85aee4b@uci.edu> References: <34501f68-7cd4-3de5-5e4d-221fc85aee4b@uci.edu> Message-ID: Thank you!! On Sun, Aug 13, 2023 at 6:21?PM Carter T. Butts wrote: > > Hi, Julia - > > Looking at your figure, you'll definitely want mixing terms (nodematch, nodemix), because your clustering is very strongly attribute-based. But I also see that you have a situation where not all nodes are participating in those subgroup interactions, so one key to getting this model to work is likely to be capturing the combination of the homophily/differential mixing among the nodes that do interact, and heterogeneity in propensity to interact at all. nodecov/nodefactor and degree terms can be useful for the latter, and you may indeed need an isolates term. In any event, the figure you sent already provides a lot of helpful clues for modeling, so keep up with the visualizations and you are likely to find your way! > > Hope that helps, > > -Carter > > On 8/13/23 1:56 PM, Julia Vassey wrote: > > Thank you Michal and Carter! Appreciate all the suggestions. I do have > cohesive groups, mostly by geographic regions and they are represented > by color in this attached plot. I have not yet tried nodeifactor or > nodeofactor, but I will. > > On Sun, Aug 13, 2023 at 9:43?AM Micha? Bojanowski wrote: > > Hi Julia, > > Thanks. Some questions and hypotheses about model specification: > > 1. The network is directed. Are you sure you want nodefactor (nodes in > some groups have more ties than nodes in other groups irrespectively > of tie direction) rather than nodeofactor (same but for outgoing ties > only) or nodeifactor (same but for incoming ties only)? > > 2. Your model does not have any nodematch terms on the nodal > attributes. Is that intentional? > > 3. The ESP distribution is notably relatively flat. One mechanism that > might be responsible for that is strong community structure or > homophily on some observed attribute -- triangles get closed within > groups only and the count of ESP depends on group size. Do you have > several cohesive groups of varying size in the network? > > m. > > On Fri, Aug 11, 2023 at 4:51?PM Julia Vassey wrote: > > Thank you Michal! Good to hear from you. > > Please, see the plots, attached. I have tried changing the decay > parameter multiple times, and the model fit gets better when > increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), > but beyond 0.3 the model starts having issues with convergence. > p values for all esps are always very low. > > Thank you for helping! > > Julia > > > On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: > > Hi Julia, > > Can you send the GOF plots attached? In what way does the ESP not fit > well? Perhaps it is a matter of changing the decay parameter? > > Michal > > On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: > > Dear All, > > This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. > > I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. > > I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. > > model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) > > Thank you, > > > -- > Julia Vassey > Health Behavior Research > Department of Population and Public Health Sciences > Keck School of Medicine > University of Southern California > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Ie0ff82n4kWRW0ahOT_x2o11HXJ_IOxUBaL3xWEkPBpEcCE2rhK6tYDMmOZy_W2ko3PJ6Lx_6fQd2g$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!ram8cU6EoMuOY9pCHEvx41wCN3xMvqWhriZ7QWrL65e7tbhCzV1HRlWqhXVeoGTY7-RUCas_2EpJ2w$ From jmoody77 at duke.edu Sun Aug 13 19:00:41 2023 From: jmoody77 at duke.edu (James Moody) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ergm inquiry In-Reply-To: References: <34501f68-7cd4-3de5-5e4d-221fc85aee4b@uci.edu> Message-ID: Never underestimate the power and usefulness of basic descrptives and visualizations. The insight gained by seeing the clustering and heterogeneity is key; then model. Most of the time, it will confirm what you learned ? but starting with modems sans deep investment in learning the case is a recipe for frustration. Sent from my iPad > On Aug 13, 2023, at 6:27 PM, Julia Vassey wrote: > > ?Thank you!! > >> On Sun, Aug 13, 2023 at 6:21?PM Carter T. Butts wrote: >> >> Hi, Julia - >> >> Looking at your figure, you'll definitely want mixing terms (nodematch, nodemix), because your clustering is very strongly attribute-based. But I also see that you have a situation where not all nodes are participating in those subgroup interactions, so one key to getting this model to work is likely to be capturing the combination of the homophily/differential mixing among the nodes that do interact, and heterogeneity in propensity to interact at all. nodecov/nodefactor and degree terms can be useful for the latter, and you may indeed need an isolates term. In any event, the figure you sent already provides a lot of helpful clues for modeling, so keep up with the visualizations and you are likely to find your way! >> >> Hope that helps, >> >> -Carter >> >> On 8/13/23 1:56 PM, Julia Vassey wrote: >> >> Thank you Michal and Carter! Appreciate all the suggestions. I do have >> cohesive groups, mostly by geographic regions and they are represented >> by color in this attached plot. I have not yet tried nodeifactor or >> nodeofactor, but I will. >> >> On Sun, Aug 13, 2023 at 9:43?AM Micha? Bojanowski wrote: >> >> Hi Julia, >> >> Thanks. Some questions and hypotheses about model specification: >> >> 1. The network is directed. Are you sure you want nodefactor (nodes in >> some groups have more ties than nodes in other groups irrespectively >> of tie direction) rather than nodeofactor (same but for outgoing ties >> only) or nodeifactor (same but for incoming ties only)? >> >> 2. Your model does not have any nodematch terms on the nodal >> attributes. Is that intentional? >> >> 3. The ESP distribution is notably relatively flat. One mechanism that >> might be responsible for that is strong community structure or >> homophily on some observed attribute -- triangles get closed within >> groups only and the count of ESP depends on group size. Do you have >> several cohesive groups of varying size in the network? >> >> m. >> >> On Fri, Aug 11, 2023 at 4:51?PM Julia Vassey wrote: >> >> Thank you Michal! Good to hear from you. >> >> Please, see the plots, attached. I have tried changing the decay >> parameter multiple times, and the model fit gets better when >> increasing the parameter: from 0 to 0.3 (gwesp(0.3, fixed = TRUE)), >> but beyond 0.3 the model starts having issues with convergence. >> p values for all esps are always very low. >> >> Thank you for helping! >> >> Julia >> >> >> On Fri, Aug 11, 2023 at 8:53?AM Micha? Bojanowski wrote: >> >> Hi Julia, >> >> Can you send the GOF plots attached? In what way does the ESP not fit >> well? Perhaps it is a matter of changing the decay parameter? >> >> Michal >> >> On Wed, Aug 9, 2023 at 1:00?PM Julia Vassey wrote: >> >> Dear All, >> >> This is my first time posting a question in statnet help. If the question needs to be posted/sent to a different email please let me know. >> >> I have a question related to ergm. I am running an ergm model on a directed unipartite network of ~100 nodes (certain social media users) and ~700 edges. The model includes geolocation attributes of the nodes (geographic regions) and themes the nodes post about on social media. The model also includes terms (mutual, gwesp) provided in the code below. >> >> I am struggling with achieving a decent goodness of fit for edgewise shared partners using gof function. i/o degree, geodesic distance and model statistics look much better. I tried different things to try to improve edgewise shared partners, but nothing seems to work. The configuration below provides the best fit, however, I want to keep trying to improve the fit for edgewise shared partners. I appreciate any thoughts and comments on how to achieve this. >> >> model = ergm::ergm(netg_infl ~ edges + nodefactor('region_code', levels = -4) + nodematch('region_code', diff = T, levels = -3) + nodefactor('topic_sum_binary', levels = -LARGEST) + nodefactor('marijuana_recode', levels = -c(1,3)) + nodefactor('nature_recode', levels = -c(1,3)) + nodefactor('health_life_recode', levels = -c(1,3)) + nodefactor('gaming_recode', levels = -c(1,3)) + nodefactor('food_recode', levels = -c(1,3)) + nodefactor('clothing_recode', levels = -c(1,3)) + mutual + gwesp(0.28, fixed = TRUE) + offset(isolates), offset.coef = -Inf, control=control.ergm(parallel=2, parallel.type="PSOCK")) >> >> Thank you, >> >> >> -- >> Julia Vassey >> Health Behavior Research >> Department of Population and Public Health Sciences >> Keck School of Medicine >> University of Southern California >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!tbnR-2a3qwh_Vmaoc9B-qqMKQEZSxi4ehcV1JysdIymvDlKhFAh3fTMCFaO9ViZjQWKS75IRghzrMXLsKQ$ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Ie0ff82n4kWRW0ahOT_x2o11HXJ_IOxUBaL3xWEkPBpEcCE2rhK6tYDMmOZy_W2ko3PJ6Lx_6fQd2g$ >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!LIr3w8kk_Xxm!ram8cU6EoMuOY9pCHEvx41wCN3xMvqWhriZ7QWrL65e7tbhCzV1HRlWqhXVeoGTY7-RUCas_2EpJ2w$ > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!OToaGQ!u2yYz9zkzl-o-6HhWw1jM-0QCe1wVO-U4PXO_TnqwludGKbkesNkVhhYrKCO4779BZXBAP4tte1awwIyJA$ From S.Chambers3 at westernsydney.edu.au Sun Aug 20 20:08:52 2023 From: S.Chambers3 at westernsydney.edu.au (Simon Chambers) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Residual effect of prior ties Message-ID: Hello, I was hoping to understand if statnet?s temporal network tools have any capacity to model a residual effect of prior tie formation beyond the immediately preceding network. i.e. if a dyad had an edge at t1, which was then not present at t2, the likelihood of a tie at t3 may be influenced by the presence of an edge at t1 ? some networks may exhibit an increased likelihood (intermittent associations not necessarily reoccurring at the intervals represented by the panel data) whereas others may be less likely (once a tie is dissolved, it is less likely to re-form). Thanks, Simon. -- Dr Simon Chambers | Postdoctoral Research Fellow The MARCS Institute Western Sydney University M: +61 477 333 818 westernsydney.edu.au With respect for Aboriginal cultural protocol and out of recognition that its campuses occupy their traditional lands, Western Sydney University acknowledges the Darug, Tharawal (also historically referred to as Dharawal), Gandangarra and Wiradjuri peoples and thanks them for their support of its work in their lands (Greater Western Sydney and beyond). -------------- next part -------------- An HTML attachment was scrubbed... URL: From philip.leifeld at essex.ac.uk Sun Aug 20 23:59:46 2023 From: philip.leifeld at essex.ac.uk (Leifeld, Philip) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Residual effect of prior ties In-Reply-To: References: Message-ID: Hi Simon, You can do this with a TERGM using the btergm package (which is not part of statnet but builds on it). See parameter K in Equation 2 and the "lag" arguments of the model terms in Table 2 in the JStatSoft paper. There are two caveats: There are not many temporal model terms in the package (e.g., there are delayed reciprocity and dyadic stability, but no triadic closure over three time periods), and the first K networks in the sequence are used for forming covariates and hence cannot be on the left-hand side of the equation (see Section 5.4, "Temporal dependencies and the treatment of initial networks" in the JStatSoft paper). Best regards, Philip -- Philip Leifeld Professor, Department of Government University of Essex https://www.philipleifeld.com On 21/08/2023 04:08, Simon Chambers wrote: CAUTION: This email was sent from outside the University of Essex. Please do not click any links or open any attachments unless you recognise and trust the sender. If you are unsure whether the content of the email is safe or have any other queries, please contact the IT Helpdesk. Hello, I was hoping to understand if statnet?s temporal network tools have any capacity to model a residual effect of prior tie formation beyond the immediately preceding network. i.e. if a dyad had an edge at t1, which was then not present at t2, the likelihood of a tie at t3 may be influenced by the presence of an edge at t1 ? some networks may exhibit an increased likelihood (intermittent associations not necessarily reoccurring at the intervals represented by the panel data) whereas others may be less likely (once a tie is dissolved, it is less likely to re-form). Thanks, Simon. -- Dr Simon Chambers | Postdoctoral Research Fellow The MARCS Institute Western Sydney University M: +61 477 333 818 westernsydney.edu.au With respect for Aboriginal cultural protocol and out of recognition that its campuses occupy their traditional lands, Western Sydney University acknowledges the Darug, Tharawal (also historically referred to as Dharawal), Gandangarra and Wiradjuri peoples and thanks them for their support of its work in their lands (Greater Western Sydney and beyond). _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Mon Aug 21 01:40:00 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Residual effect of prior ties In-Reply-To: References: Message-ID: <3842e449-c955-4df1-620b-533e80301928@uci.edu> Hi, Simon - It may be useful to remember that you can define any lag effect of any complexity, or with any mix of prior states, using one or more appropriately designed edgecov terms.? The past is always a covariate to the present, as thus are functions of the past history. Hope that helps, -Carter On 8/20/23 8:08 PM, Simon Chambers wrote: > > Hello, > > I was hoping to understand if statnet?s temporal network tools have > any capacity to model a residual effect of prior tie formation beyond > the immediately preceding network. > > i.e. if a dyad had an edge at t1, which was then not present at t2, > the likelihood of a tie at t3 may be influenced by the presence of an > edge at t1 ? some networks may exhibit an increased likelihood > (intermittent associations not necessarily reoccurring at the > intervals represented by the panel data) whereas others may be less > likely (once a tie is dissolved, it is less likely to re-form). > > Thanks, Simon. > > -- > > *Dr Simon Chambers*?| Postdoctoral Research Fellow > > The MARCS Institute > > Western Sydney University > > M: +61 477 333 818 > > westernsydney.edu.au > > > /With respect for Aboriginal cultural protocol and out of recognition > that its campuses occupy their traditional lands, Western Sydney > University acknowledges the Darug, Tharawal (also historically > referred to as Dharawal), Gandangarra?and Wiradjuri peoples and thanks > them for their support of its work in their lands (Greater Western > Sydney and beyond)./ > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!K49Lshq8OuB5Tr39yrHZgOFED48vpE2DQdR2xFqvTzRIwNFi1J_yGORW7jIrK98_ydwgVaBIuskHpzCeP1Num03tS9duaA8$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From S.Chambers3 at westernsydney.edu.au Mon Aug 21 18:43:33 2023 From: S.Chambers3 at westernsydney.edu.au (Simon Chambers) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] [External] Re: Residual effect of prior ties In-Reply-To: <3842e449-c955-4df1-620b-533e80301928@uci.edu> References: <3842e449-c955-4df1-620b-533e80301928@uci.edu> Message-ID: Hi Carter, Thanks for your suggestion. Am I right in my understanding that the input to edgecov (e.g. a matrix) is singular ? such that there is not the capacity to include this term in a tergm in a manner which updates over time, such that successive waves get a progressively updated version of the past? As such, to the extent that it is a mechanism to model the effects of prior states, it is more suited to doing so in standard ergms where multiple edgecovs can be designed to model particular aspects of past history? Regards, Simon. From: statnet_help on behalf of Carter T. Butts Date: Monday, 21 August 2023 at 6:41 pm To: statnet_help@u.washington.edu Subject: [External] Re: [statnet_help] Residual effect of prior ties Hi, Simon - It may be useful to remember that you can define any lag effect of any complexity, or with any mix of prior states, using one or more appropriately designed edgecov terms. The past is always a covariate to the present, as thus are functions of the past history. Hope that helps, -Carter On 8/20/23 8:08 PM, Simon Chambers wrote: Hello, I was hoping to understand if statnet?s temporal network tools have any capacity to model a residual effect of prior tie formation beyond the immediately preceding network. i.e. if a dyad had an edge at t1, which was then not present at t2, the likelihood of a tie at t3 may be influenced by the presence of an edge at t1 ? some networks may exhibit an increased likelihood (intermittent associations not necessarily reoccurring at the intervals represented by the panel data) whereas others may be less likely (once a tie is dissolved, it is less likely to re-form). Thanks, Simon. -- Dr Simon Chambers | Postdoctoral Research Fellow The MARCS Institute Western Sydney University M: +61 477 333 818 westernsydney.edu.au With respect for Aboriginal cultural protocol and out of recognition that its campuses occupy their traditional lands, Western Sydney University acknowledges the Darug, Tharawal (also historically referred to as Dharawal), Gandangarra and Wiradjuri peoples and thanks them for their support of its work in their lands (Greater Western Sydney and beyond). _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!K49Lshq8OuB5Tr39yrHZgOFED48vpE2DQdR2xFqvTzRIwNFi1J_yGORW7jIrK98_ydwgVaBIuskHpzCeP1Num03tS9duaA8$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From William.Leung2 at lshtm.ac.uk Thu Sep 7 18:58:20 2023 From: William.Leung2 at lshtm.ac.uk (William Leung) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] AIC and BIC from package:ergm.ego Message-ID: To the Statnet team, I have been fitting ERGMs with package:ergm.ego and would now like to do some formal, data driven model selection using AIC or BIC. I am aware that this is calculated within package:ergm, but is this the case for package:ergm.ego? If not, can this be simply calculated from the log likelihood (`model$loglikelihood`) as per convention? (e.g. AIC = ? 2 ? ln likelihood +2 k +1), or is this an oversimplification? Many thanks and best wishes, Will Sent from Mail for Windows -------------- next part -------------- An HTML attachment was scrubbed... URL: From p.krivitsky at unsw.edu.au Sun Sep 17 21:53:57 2023 From: p.krivitsky at unsw.edu.au (Pavel Krivitsky) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] AIC and BIC from package:ergm.ego In-Reply-To: References: Message-ID: Dear William, ergm.ego's inference is not, strictly speaking, likelihood-based (see Krivitsky and Morris, 2017), so it doesn't report AIC, BIC, or other likelihood-based quantities. I hope this helps, Pavel On Fri, 2023-09-08 at 01:58 +0000, William Leung wrote: To the Statnet team, I have been fitting ERGMs with package:ergm.ego and would now like to do some formal, data driven model selection using AIC or BIC. I am aware that this is calculated within package:ergm, but is this the case for package:ergm.ego? If not, can this be simply calculated from the log likelihood (`model$loglikelihood`) as per convention? (e.g.AIC = ? 2 ? ln likelihood +2 k +1), or is this an oversimplification? Many thanks and best wishes, Will Sent from Mail for Windows _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From dtrapido at uw.edu Mon Sep 25 18:29:03 2023 From: dtrapido at uw.edu (Denis Trapido) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Comparing homophily for select groups Message-ID: Hi- I have a network directed, non-valued (0/1) ties and a node attribute, ethnicity. I want to use ERGM to compare the propensities of two ethic groups, A and B, to send ties to ethnicity C. In other words, I want to know if members of group A are more likely than members of group B to send ties to C. How should I go about it? Best, Denis Denis Trapido Associate Professor Management and Organization School of Business University of Washington Bothell -------------- next part -------------- An HTML attachment was scrubbed... URL: From companalysis2012 at gmail.com Tue Oct 17 16:45:00 2023 From: companalysis2012 at gmail.com (SJ C) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about adding moderating effect in REM Message-ID: Hi all, I am using REM(relevant event model) for the first time in my research project, and also this is my first time sending questions to statnet_help. My question is about entering a moderating effect in REM. I consulted REM tutorials uploaded in Statnet website and other workshop materials, but was not able to find relevant information. For instance, how can we test whether the RRecSnd effect is moderated by a particular attribute (e.g., education level) of the original sender in REM? It would be greatly appreciated if I can have any responses. Thank you! Sincerely, Choi -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Tue Oct 17 19:06:01 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about adding moderating effect in REM In-Reply-To: References: Message-ID: <27ab70b1-8e99-b01e-723e-35e285358950@uci.edu> Hi, Choi - To be clear, I presume that by "moderation" you mean an interaction effect (i.e., a product term between e.g. the RRecSnd statistic and another statistic).? This can be done, but currently it is DIY: what you have to do is compute the statistics you want, and enter them as dynamic edge (aka event) covariates.? (It is a basic property of these models that any and all effects can be implemented as (dynamic) edge covariates, which is a handy thing to keep in mind if you want to implement your own statistics.) There is an internal function accum.rrl that may be handy for this purpose; it is part of the black magic of the package (and thus not very documented), but can be used.? If you call accum.rrl with an eventlist, it will return a list with two elements.? The first is a list, with one entry per event in the data, whose ith entry is an ordered list of the most recent senders for every node that has received an event (/going into/ the ith event).? The second is the corresponding list for most recent receivers for every node that has sent an event (again, the ith entry is the state headed into the ith event).? Nodes that have never sent/received by a given event do not have entries.? This function is used "backstage" to help compute the hazards for RRecSnd and RSndSnd, so can be helpful for you if you want to make your own term.? But of course, it is just a tool for tabulation, and you can write your own if you prefer. Hope that helps! -Carter On 10/17/23 4:45 PM, SJ C wrote: > Hi all, > > I am using REM(relevant event model) for the first time in my?research > project, and also this is my first time sending questions to statnet_help. > > My question is about entering a moderating effect in REM. > I consulted REM tutorials uploaded in Statnet website and other > workshop materials, but was not able to find relevant information. > > For instance, how can we test whether the RRecSnd?effect is moderated > by a particular attribute (e.g., education level) of the original > sender in REM? > > It would be greatly?appreciated if I can have any responses. > Thank you! > > Sincerely, > Choi > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!PshxIW-NTXvNytOo2UNgN4BoECliq11-0VSkV_z-sFJoai8EOjfULJwZkLbKfA2fj43MiNC_6GfJ1WxJ9mMkHCpc$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From aditya_khanna at brown.edu Wed Oct 25 13:06:41 2023 From: aditya_khanna at brown.edu (Khanna, Aditya) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ERGM error Message-ID: Hello Statnet Dev Team and All, I have some code for an ERGM that I developed in v 3.10.4 (and R 3.6.0). I didn't upgrade it to the latest ERGM because it had a custom term and used a very large underlying dataset. Until recently, the code and the environment both worked just fine, I periodically re-ran the code on new data as updates were needed. Today I tried to run the code again, but it, and simpler versions of the ERGM (i.e., a model as simple as just including the edges term), but all failed. I then tried running a standard tutorial example, and it produces the same error: > data(florentine) > summary(flomarriage ~ edges) edges 20 > ergm(formula = flomarriage ~ edges) Error in rep(rle(v), r, scale = "run") : invalid 'times' argument > I have been managing the project using renv, so I don't know what underlying thing might have changed.I would be grateful for any thoughts on how to debug this. The sessionInfo() is included below. Aditya > sessionInfo() R version 3.6.0 (2019-04-26) Platform: x86_64-pc-linux-gnu (64-bit) Running under: Red Hat Enterprise Linux Server 7.9 (Maipo) Matrix products: default BLAS/LAPACK: /oscar/rt/7.2/opt/intel/2017.0/compilers_and_libraries_2017.0.098/linux/mkl/lib/intel64_lin/libmkl_gf_lp64.so locale: [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8 [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8 [7] LC_PAPER=en_US.UTF-8 LC_NAME=C [9] LC_ADDRESS=C LC_TELEPHONE=C [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C attached base packages: [1] stats graphics grDevices datasets utils methods base other attached packages: [1] here_1.0.1 ergm.userterms_3.1.1 statnet.common_4.7.0 [4] dplyr_1.1.0 ergm_3.10.4 network_1.18.1 loaded via a namespace (and not attached): [1] magrittr_2.0.3 MASS_7.3-51.4 tidyselect_1.2.0 lattice_0.20-38 [5] R6_2.5.1 rlang_1.1.1 fansi_1.0.4 tools_3.6.0 [9] parallel_3.6.0 grid_3.6.0 lpSolve_5.6.18 utf8_1.2.3 [13] cli_3.6.0 coda_0.19-4 withr_2.5.0 rprojroot_2.0.3 [17] tibble_3.1.8 lifecycle_1.0.3 Matrix_1.2-17 purrr_1.0.1 [21] vctrs_0.5.2 trust_0.1-8 robustbase_0.95-0 glue_1.6.2 [25] DEoptimR_1.0-11 compiler_3.6.0 pillar_1.8.1 generics_0.1.3 [29] jsonlite_1.8.7 renv_1.0.0 pkgconfig_2.0.3 > -------------- next part -------------- An HTML attachment was scrubbed... URL: From aditya_khanna at brown.edu Thu Oct 26 08:41:53 2023 From: aditya_khanna at brown.edu (Aditya Khanna) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ERGM error In-Reply-To: References: Message-ID: Hi All, I created the environment again with a fresh install of the package versions I am using. The flomarriage example and my own code are both working now. Aditya > On Oct 25, 2023, at 4:06 PM, Khanna, Aditya wrote: > > Hello Statnet Dev Team and All, > > I have some code for an ERGM that I developed in v 3.10.4 (and R 3.6.0). I didn't upgrade it to the latest ERGM because it had a custom term and used a very large underlying dataset. Until recently, the code and the environment both worked just fine, I periodically re-ran the code on new data as updates were needed. > > Today I tried to run the code again, but it, and simpler versions of the ERGM (i.e., a model as simple as just including the edges term), but all failed. I then tried running a standard tutorial example, and it produces the same error: > > > data(florentine) > > summary(flomarriage ~ edges) > edges > 20 > > ergm(formula = flomarriage ~ edges) > Error in rep(rle(v), r, scale = "run") : invalid 'times' argument > > > > I have been managing the project using renv, so I don't know what underlying thing might have changed.I would be grateful for any thoughts on how to debug this. The sessionInfo() is included below. > Aditya > > > sessionInfo() > R version 3.6.0 (2019-04-26) > Platform: x86_64-pc-linux-gnu (64-bit) > Running under: Red Hat Enterprise Linux Server 7.9 (Maipo) > > Matrix products: default > BLAS/LAPACK: /oscar/rt/7.2/opt/intel/2017.0/compilers_and_libraries_2017.0.098/linux/mkl/lib/intel64_lin/libmkl_gf_lp64.so > > locale: > [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C > [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8 > [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8 > [7] LC_PAPER=en_US.UTF-8 LC_NAME=C > [9] LC_ADDRESS=C LC_TELEPHONE=C > [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C > > attached base packages: > [1] stats graphics grDevices datasets utils methods base > > other attached packages: > [1] here_1.0.1 ergm.userterms_3.1.1 statnet.common_4.7.0 > [4] dplyr_1.1.0 ergm_3.10.4 network_1.18.1 > > loaded via a namespace (and not attached): > [1] magrittr_2.0.3 MASS_7.3-51.4 tidyselect_1.2.0 lattice_0.20-38 > [5] R6_2.5.1 rlang_1.1.1 fansi_1.0.4 tools_3.6.0 > [9] parallel_3.6.0 grid_3.6.0 lpSolve_5.6.18 utf8_1.2.3 > [13] cli_3.6.0 coda_0.19-4 withr_2.5.0 rprojroot_2.0.3 > [17] tibble_3.1.8 lifecycle_1.0.3 Matrix_1.2-17 purrr_1.0.1 > [21] vctrs_0.5.2 trust_0.1-8 robustbase_0.95-0 glue_1.6.2 > [25] DEoptimR_1.0-11 compiler_3.6.0 pillar_1.8.1 generics_0.1.3 > [29] jsonlite_1.8.7 renv_1.0.0 pkgconfig_2.0.3 > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Fri Oct 27 04:12:56 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ERGM error In-Reply-To: References: Message-ID: Dear Aditya, Yeah, running old versions of the Statnet in an even older version of R might be tricky... It's strange that suddenly stopped working though, given that you managed the environment with 'renv'. You must have updated something, right? It was all already some time but I believe the issues might be related to the rle functionality which for the moment was part of statnet.common (or ergm) and was later externalized to a separate package 'rle'. I'm glad that all worked-out in the end. Best, Michal On Thu, Oct 26, 2023 at 5:43?PM Aditya Khanna wrote: > > Hi All, > > I created the environment again with a fresh install of the package versions I am using. The flomarriage example and my own code are both working now. > Aditya > > On Oct 25, 2023, at 4:06 PM, Khanna, Aditya wrote: > > Hello Statnet Dev Team and All, > > I have some code for an ERGM that I developed in v 3.10.4 (and R 3.6.0). I didn't upgrade it to the latest ERGM because it had a custom term and used a very large underlying dataset. Until recently, the code and the environment both worked just fine, I periodically re-ran the code on new data as updates were needed. > > Today I tried to run the code again, but it, and simpler versions of the ERGM (i.e., a model as simple as just including the edges term), but all failed. I then tried running a standard tutorial example, and it produces the same error: > > > data(florentine) > > summary(flomarriage ~ edges) > edges > 20 > > ergm(formula = flomarriage ~ edges) > Error in rep(rle(v), r, scale = "run") : invalid 'times' argument > > > > I have been managing the project using renv, so I don't know what underlying thing might have changed.I would be grateful for any thoughts on how to debug this. The sessionInfo() is included below. > Aditya > > > sessionInfo() > R version 3.6.0 (2019-04-26) > Platform: x86_64-pc-linux-gnu (64-bit) > Running under: Red Hat Enterprise Linux Server 7.9 (Maipo) > > Matrix products: default > BLAS/LAPACK: /oscar/rt/7.2/opt/intel/2017.0/compilers_and_libraries_2017.0.098/linux/mkl/lib/intel64_lin/libmkl_gf_lp64.so > > locale: > [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C > [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8 > [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8 > [7] LC_PAPER=en_US.UTF-8 LC_NAME=C > [9] LC_ADDRESS=C LC_TELEPHONE=C > [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C > > attached base packages: > [1] stats graphics grDevices datasets utils methods base > > other attached packages: > [1] here_1.0.1 ergm.userterms_3.1.1 statnet.common_4.7.0 > [4] dplyr_1.1.0 ergm_3.10.4 network_1.18.1 > > loaded via a namespace (and not attached): > [1] magrittr_2.0.3 MASS_7.3-51.4 tidyselect_1.2.0 lattice_0.20-38 > [5] R6_2.5.1 rlang_1.1.1 fansi_1.0.4 tools_3.6.0 > [9] parallel_3.6.0 grid_3.6.0 lpSolve_5.6.18 utf8_1.2.3 > [13] cli_3.6.0 coda_0.19-4 withr_2.5.0 rprojroot_2.0.3 > [17] tibble_3.1.8 lifecycle_1.0.3 Matrix_1.2-17 purrr_1.0.1 > [21] vctrs_0.5.2 trust_0.1-8 robustbase_0.95-0 glue_1.6.2 > [25] DEoptimR_1.0-11 compiler_3.6.0 pillar_1.8.1 generics_0.1.3 > [29] jsonlite_1.8.7 renv_1.0.0 pkgconfig_2.0.3 > > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help From aditya_khanna at brown.edu Fri Oct 27 05:44:43 2023 From: aditya_khanna at brown.edu (Khanna, Aditya) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] ERGM error In-Reply-To: References: Message-ID: Thanks Michal! I had installed the languageserver and httpgd packages to use the vscode extension for R. I haven?t install them in my fresh environment that is working. Aditya On Fri, Oct 27, 2023 at 7:13 AM Micha? Bojanowski wrote: > Dear Aditya, > > Yeah, running old versions of the Statnet in an even older version of > R might be tricky... It's strange that suddenly stopped working > though, given that you managed the environment with 'renv'. You must > have updated something, right? > > It was all already some time but I believe the issues might be related > to the rle functionality which for the moment was part of > statnet.common (or ergm) and was later externalized to a separate > package 'rle'. > > I'm glad that all worked-out in the end. > > Best, > Michal > > On Thu, Oct 26, 2023 at 5:43?PM Aditya Khanna > wrote: > > > > Hi All, > > > > I created the environment again with a fresh install of the package > versions I am using. The flomarriage example and my own code are both > working now. > > Aditya > > > > On Oct 25, 2023, at 4:06 PM, Khanna, Aditya > wrote: > > > > Hello Statnet Dev Team and All, > > > > I have some code for an ERGM that I developed in v 3.10.4 (and R 3.6.0). > I didn't upgrade it to the latest ERGM because it had a custom term and > used a very large underlying dataset. Until recently, the code and the > environment both worked just fine, I periodically re-ran the code on new > data as updates were needed. > > > > Today I tried to run the code again, but it, and simpler versions of the > ERGM (i.e., a model as simple as just including the edges term), but all > failed. I then tried running a standard tutorial example, and it produces > the same error: > > > > > data(florentine) > > > summary(flomarriage ~ edges) > > edges > > 20 > > > ergm(formula = flomarriage ~ edges) > > Error in rep(rle(v), r, scale = "run") : invalid 'times' argument > > > > > > > I have been managing the project using renv, so I don't know what > underlying thing might have changed.I would be grateful for any thoughts on > how to debug this. The sessionInfo() is included below. > > Aditya > > > > > sessionInfo() > > R version 3.6.0 (2019-04-26) > > Platform: x86_64-pc-linux-gnu (64-bit) > > Running under: Red Hat Enterprise Linux Server 7.9 (Maipo) > > > > Matrix products: default > > BLAS/LAPACK: > /oscar/rt/7.2/opt/intel/2017.0/compilers_and_libraries_2017.0.098/linux/mkl/lib/intel64_lin/libmkl_gf_lp64.so > > > > locale: > > [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C > > [3] LC_TIME=en_US.UTF-8 LC_COLLATE=en_US.UTF-8 > > [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8 > > [7] LC_PAPER=en_US.UTF-8 LC_NAME=C > > [9] LC_ADDRESS=C LC_TELEPHONE=C > > [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C > > > > attached base packages: > > [1] stats graphics grDevices datasets utils methods base > > > > other attached packages: > > [1] here_1.0.1 ergm.userterms_3.1.1 statnet.common_4.7.0 > > [4] dplyr_1.1.0 ergm_3.10.4 network_1.18.1 > > > > loaded via a namespace (and not attached): > > [1] magrittr_2.0.3 MASS_7.3-51.4 tidyselect_1.2.0 > lattice_0.20-38 > > [5] R6_2.5.1 rlang_1.1.1 fansi_1.0.4 tools_3.6.0 > > [9] parallel_3.6.0 grid_3.6.0 lpSolve_5.6.18 utf8_1.2.3 > > [13] cli_3.6.0 coda_0.19-4 withr_2.5.0 > rprojroot_2.0.3 > > [17] tibble_3.1.8 lifecycle_1.0.3 Matrix_1.2-17 purrr_1.0.1 > > [21] vctrs_0.5.2 trust_0.1-8 robustbase_0.95-0 glue_1.6.2 > > [25] DEoptimR_1.0-11 compiler_3.6.0 pillar_1.8.1 generics_0.1.3 > > [29] jsonlite_1.8.7 renv_1.0.0 pkgconfig_2.0.3 > > > > > > > > > _______________________________________________ > > statnet_help mailing list > > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From companalysis2012 at gmail.com Mon Oct 30 02:31:30 2023 From: companalysis2012 at gmail.com (SJ C) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about adding moderating effect in REM In-Reply-To: <27ab70b1-8e99-b01e-723e-35e285358950@uci.edu> References: <27ab70b1-8e99-b01e-723e-35e285358950@uci.edu> Message-ID: Dear Carter, I appreciate your detailed response! Based on my understanding of your explanation, I can create an interaction term by multiplying two matrices (e.g., one matrix of RRecSnd retrieved from accum.rrl and the other matrix which is transformed from node attributes) and enter the interaction term as a CovEvent in the rem.dyad. Am I correct? Assuming that I am correct, I entered the interaction term and the attribute-transformed matrix as if they are the covariates of CovEvent in rem.dyad as follows: covar = list(CovEvent = cbind(attribute-transformed matrix, interaction matrix)) However, this code did not work. If I separate this code into two as follows, the rem.dyad gives me a result of only one variable. covar = list(CovEvent = attribute-transformed matrix, CovEvent = interaction matrix) Since we have to enter lower-order terms of the interaction term, I need to enter these two terms as CovEvent. Can you please let me know how to handle this? Thank you for your time paying attention to this! Sincerely, Choi 2023? 10? 18? (?) ?? 11:06, Carter T. Butts ?? ??: > Hi, Choi - > > To be clear, I presume that by "moderation" you mean an interaction effect > (i.e., a product term between e.g. the RRecSnd statistic and another > statistic). This can be done, but currently it is DIY: what you have to do > is compute the statistics you want, and enter them as dynamic edge (aka > event) covariates. (It is a basic property of these models that any and > all effects can be implemented as (dynamic) edge covariates, which is a > handy thing to keep in mind if you want to implement your own statistics.) > > There is an internal function accum.rrl that may be handy for this > purpose; it is part of the black magic of the package (and thus not very > documented), but can be used. If you call accum.rrl with an eventlist, it > will return a list with two elements. The first is a list, with one entry > per event in the data, whose ith entry is an ordered list of the most > recent senders for every node that has received an event (*going into* > the ith event). The second is the corresponding list for most recent > receivers for every node that has sent an event (again, the ith entry is > the state headed into the ith event). Nodes that have never sent/received > by a given event do not have entries. This function is used "backstage" to > help compute the hazards for RRecSnd and RSndSnd, so can be helpful for you > if you want to make your own term. But of course, it is just a tool for > tabulation, and you can write your own if you prefer. > > Hope that helps! > > -Carter > On 10/17/23 4:45 PM, SJ C wrote: > > Hi all, > > I am using REM(relevant event model) for the first time in my research > project, and also this is my first time sending questions to statnet_help. > > My question is about entering a moderating effect in REM. > I consulted REM tutorials uploaded in Statnet website and other workshop > materials, but was not able to find relevant information. > > For instance, how can we test whether the RRecSnd effect is moderated by a > particular attribute (e.g., education level) of the original sender in REM? > > It would be greatly appreciated if I can have any responses. > Thank you! > > Sincerely, > Choi > > > _______________________________________________ > statnet_help mailing liststatnet_help@u.washington.eduhttps://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!PshxIW-NTXvNytOo2UNgN4BoECliq11-0VSkV_z-sFJoai8EOjfULJwZkLbKfA2fj43MiNC_6GfJ1WxJ9mMkHCpc$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Mon Oct 30 21:00:36 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:51 2024 Subject: [statnet_help] Question about adding moderating effect in REM In-Reply-To: References: <27ab70b1-8e99-b01e-723e-35e285358950@uci.edu> Message-ID: <7f071c2d-388b-fc6b-4bfd-d5a7e169f02e@uci.edu> Hi, Choi - Your covariate is time varying, so will need to be specified in an array of dimension m x p x n x n, where n is the number of nodes, p is the number of covariates (here 1, if you are only using the one effect), and m is the number of events in the data set.? The value X[i,j,k,l] in this array is then the value of the jth covariate for the k->l interaction leading into the ith event.? (See ?rem.dyad for documentation on format.)? You can construct that array using the information in accum.rrl, or by other means (but either way, some processing is involved).? Once you have the covariate array, you indeed pass it to the covar argument by a CovEvent entry.? In your case, it looks like your data was not formatted correctly for this purpose; in that case, it will not function correctly. Hope that helps! -Carter On 10/30/23 2:31 AM, SJ C wrote: > Dear Carter, > > I appreciate your detailed response! > > Based on my understanding of your explanation, I can create an > interaction term by multiplying two matrices (e.g., one matrix of > RRecSnd retrieved from accum.rrl and the other matrix which is > transformed?from node attributes) and enter the interaction term as a > CovEvent in the rem.dyad. Am I correct? > > Assuming that I am correct, I entered the interaction term and the > attribute-transformed matrix as if they are the covariates of CovEvent > in rem.dyad as follows: > covar = list(CovEvent = cbind(attribute-transformed matrix, > interaction matrix)) > > However, this code did not work. > If I separate this code into two as follows, the rem.dyad gives me a > result of only one variable. > covar = list(CovEvent = attribute-transformed matrix, CovEvent = > interaction matrix) > > Since we have to enter lower-order terms?of the interaction term, I > need to enter these two terms as CovEvent. > Can you please let me know how to handle this? > > Thank you for your time paying attention to this! > > Sincerely, > Choi > > 2023? 10? 18? (?) ?? 11:06, Carter T. Butts ?? > ??: > > Hi, Choi - > > To be clear, I presume that by "moderation" you mean an > interaction effect (i.e., a product term between e.g. the RRecSnd > statistic and another statistic).? This can be done, but currently > it is DIY: what you have to do is compute the statistics you want, > and enter them as dynamic edge (aka event) covariates.? (It is a > basic property of these models that any and all effects can be > implemented as (dynamic) edge covariates, which is a handy thing > to keep in mind if you want to implement your own statistics.) > > There is an internal function accum.rrl that may be handy for this > purpose; it is part of the black magic of the package (and thus > not very documented), but can be used. If you call accum.rrl with > an eventlist, it will return a list with two elements.? The first > is a list, with one entry per event in the data, whose ith entry > is an ordered list of the most recent senders for every node that > has received an event (/going into/ the ith event).? The second is > the corresponding list for most recent receivers for every node > that has sent an event (again, the ith entry is the state headed > into the ith event).? Nodes that have never sent/received by a > given event do not have entries.? This function is used > "backstage" to help compute the hazards for RRecSnd and RSndSnd, > so can be helpful for you if you want to make your own term.? But > of course, it is just a tool for tabulation, and you can write > your own if you prefer. > > Hope that helps! > > -Carter > > On 10/17/23 4:45 PM, SJ C wrote: >> Hi all, >> >> I am using REM(relevant event model) for the first time in >> my?research project, and also this is my first time sending >> questions to statnet_help. >> >> My question is about entering a moderating effect in REM. >> I consulted REM tutorials uploaded in Statnet website and other >> workshop materials, but was not able to find relevant information. >> >> For instance, how can we test whether the RRecSnd?effect is >> moderated by a particular attribute (e.g., education level) of >> the original sender in REM? >> >> It would be greatly?appreciated if I can have any responses. >> Thank you! >> >> Sincerely, >> Choi >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!PshxIW-NTXvNytOo2UNgN4BoECliq11-0VSkV_z-sFJoai8EOjfULJwZkLbKfA2fj43MiNC_6GfJ1WxJ9mMkHCpc$ > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From marcel.dick at smail.uni-koeln.de Wed Nov 15 05:40:37 2023 From: marcel.dick at smail.uni-koeln.de (Marcel Dick) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] statnet package: 'stergm'-code, but R calculates a TERGM Message-ID: <20231115144037.Horde.HdkFy-1ATUISep3T1-176Jk@webmail.uni-koeln.de> Dear statnet team, dear colleagues, My name is Marcel Dick and I am doing an external doctorate in social sciences at the University of Cologne. Among other things, I am working with R and your statnet package. I would like to investigate networks in school classes at two different points in time t1 and t2 using Separable Temporal Exponential Random Graphs (STERGMs). I used example 3 from the paper by Mr. Krivitksy and Mr. Goodreau of the statnet development team "STERGM - Separable Temporal ERGMs for modeling discrete relational dynamics with statnet" from 2016 (pp. 19-23). I have taken and adapted the code for the STERGM as shown in the example. The STERGM runs through, but R calculates a TERGM and also gives me such an output, so that I do not get separate information for a formation model and a dissolution model under 'summary()'. I have attached a corresponding screenshot. I have already talked to my supervisors but they are just as perplexed as I am as to why R calculates a TERGM although the command contains a STERGM and this code runs through without an error message. Is this perhaps a bug or is the problem already known? If so, how can I solve the problem so that R really calculates a STERGM? In the TERGM workshop on the statnet package I can only find a short appendix on STERGMS. Many thanks for your help! Best regards Marcel Dick ________________________________________ Marcel Dick, M.Ed., M.A. ext. Promovend Universit?t zu K?ln Humanwissenschaftliche Fakult?t Department Erziehungs- und Sozialwissenschaften Lehrbereich Sozialwissenschaften mit den Schwerpunkten Politikwissenschaft, Soziologie und ?konomische Bildung Lehrstuhl f?r Erziehungs- und Kultursoziologie / Soziologie E-Mail: marcel.dick@smail.uni-koeln.de -------------- next part -------------- A non-text attachment was scrubbed... Name: Krivitsky+Goodreau_2016_STERGM.pdf Type: application/pdf Size: 637449 bytes Desc: not available URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: R-output_summary('object')_object='STERGM_pos'_2023-11-15.jpg Type: image/jpeg Size: 442565 bytes Desc: not available URL: From morrism at uw.edu Wed Nov 15 12:30:43 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] statnet package: 'stergm'-code, but R calculates a TERGM In-Reply-To: <20231115144037.Horde.HdkFy-1ATUISep3T1-176Jk@webmail.uni-koeln.de> References: <20231115144037.Horde.HdkFy-1ATUISep3T1-176Jk@webmail.uni-koeln.de> Message-ID: Hi Marcel -- Thanks for reaching out! The tergm package had a major update a couple of years ago, so some things have changed. Note that STERGMs are a subset of TERGMs. So your model is a TERGM, just a specific type. At the time that paper was written in 2016 the tergm package had more limited functionality. As more functionality was added, the syntax had to change. I believe the current package still allows the old tergm::stergm() syntax for backwards compatibility for old scripts (@pavel can you confirm this)? But we don't recommend using it; use the updated syntax for any new work. The biggest difference you?ll notice is that this typically involves more than one formula, and the formulas are specified using the new temporal ?operators? available since version 4.0 of the tergm package. If your model is separable, the tergm package will recognize that and fit accordingly. We have a tutorial with examples of separable models on the Statnet website: https://statnet.org/workshop-tergm/ Please make sure you have updated both R and all of your statnet packages to the most recent versions. Then see if the tutorial gives you what you need to make progress. Feel free to reach out again if you have questions! best, Martina On Wed, Nov 15, 2023 at 5:41?AM Marcel Dick wrote: > Dear statnet team, dear colleagues, > > > My name is Marcel Dick and I am doing an external doctorate in social > sciences at the University of Cologne. Among other things, I am > working with R and your statnet package. > > I would like to investigate networks in school classes at two > different points in time t1 and t2 using Separable Temporal > Exponential Random Graphs (STERGMs). I used example 3 from the paper > by Mr. Krivitksy and Mr. Goodreau of the statnet development team > "STERGM - Separable Temporal ERGMs for modeling discrete relational > dynamics with statnet" from 2016 (pp. 19-23). I have taken and adapted > the code for the STERGM as shown in the example. The STERGM runs > through, but R calculates a TERGM and also gives me such an output, so > that I do not get separate information for a formation model and a > dissolution model under 'summary()'. I have attached a corresponding > screenshot. > > I have already talked to my supervisors but they are just as perplexed > as I am as to why R calculates a TERGM although the command contains a > STERGM and this code runs through without an error message. Is this > perhaps a bug or is the problem already known? If so, how can I solve > the problem so that R really calculates a STERGM? > > In the TERGM workshop on the statnet package I can only find a short > appendix on STERGMS. > > Many thanks for your help! > > > Best regards > Marcel Dick > > ________________________________________ > > Marcel Dick, M.Ed., M.A. > ext. Promovend > > Universit?t zu K?ln > Humanwissenschaftliche Fakult?t > Department Erziehungs- und Sozialwissenschaften > Lehrbereich Sozialwissenschaften mit den Schwerpunkten > Politikwissenschaft, Soziologie und ?konomische Bildung > Lehrstuhl f?r Erziehungs- und Kultursoziologie / Soziologie > > E-Mail: marcel.dick@smail.uni-koeln.de > -------------- next part -------------- An HTML attachment was scrubbed... URL: From carine.pachoud at hotmail.fr Thu Nov 23 05:50:55 2023 From: carine.pachoud at hotmail.fr (Carine Pachoud) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: , , Message-ID: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud [https://s-install.avcdn.net/ipm/preview/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif] Sans virus.www.avast.com -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Thu Nov 23 06:13:05 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud wrote: > Dear ERGM users, > > I previously worked with ERGM on complete networks. Today I am dealing > with an egocentric network and the package ego.ergm. I have 55 nodes et 274 > edges. I do not know the size of the population. > > It's certainly a very simple adjustment to make but I'm stuck on the > population size. When I want to run my model, I get the following error > message: > > Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : > missing value where TRUE/FALSE needed > > I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. > > Could you help me ? > > thanks a lot! > > best regards, > Carine Pachoud > > > > > > Sans virus.www.avast.com > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Thu Nov 23 06:55:12 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] Fwd: ego.ergm popsize In-Reply-To: References: Message-ID: Forwarding the reply back to the mailing list. ---------- Forwarded message --------- From: Carine Pachoud Date: Thu, Nov 23, 2023 at 3:43?PM Subject: RE: [statnet_help] ego.ergm popsize To: Micha? Bojanowski Hello Michal, thank you for your quick answer ! Here the call : m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) summary(m17) I tried again adding the actual ppopsize : 990 and I get a new message error : Constructing pseudopopulation network. Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > summary(m17) Error: object 'm17' not found This message appears for each of the terms. The study focuses on a territorial agrifood system, including a diversity of actors (farmers, elected officials, civil society, advisors...) who have different visions of agriculture. In order to understand sustainable agrifood transformations, I try to see if the different groups holding different visions interact and how is structured the network (ERGM). I used a snowball sampling starting from 15 actors, central and in each typology. best regards, Carine Pachoud ________________________________ De : Micha? Bojanowski Envoy? : jeudi 23 novembre 2023 11:13 ? : Carine Pachoud Cc : statnet_help@u.washington.edu Objet : Re: [statnet_help] ego.ergm popsize Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud wrote: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud Sans virus.www.avast.com _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help From carine.pachoud at hotmail.fr Thu Nov 23 07:06:49 2023 From: carine.pachoud at hotmail.fr (Carine Pachoud) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Dear Micha?, here the summary for f.ego: > summary(f.ego) 55 Egos/ Ego Networks 430 Alters Min. Netsize 1 Average Netsize 7.81818181818182 Max. Netsize 31 Average Density 0.674972800136199 Alter survey design: Maximum nominations: Inf I tried on undirected network : m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov ("LienE")+ nodefactor ("View")+ nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) and now I get the following message: Constructing pseudopopulation network. Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). Run `rlang::last_trace()` to see where the error occurred.> rlang::last_trace() Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). I do not understand why the attribute is invalid (I tried with the package ERGM and it worked). thank you for the reference, I will read it! best regards, Carine ________________________________ De : Micha? Bojanowski Envoy? : jeudi 23 novembre 2023 11:54 ? : Carine Pachoud Objet : Re: [statnet_help] ego.ergm popsize ... and: 4. The fact that you used snowball sampling to collect data makes inference with ergm.ego questionable because it assumes that the data is a probabilistic sample of egos. You may need a different model such as https://doi.org/10.1016/j.socnet.2015.11.003. On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: Thanks Carine. Couple of issues/questions based on what you sent: 1. Can you please send the output of summary(f.ego) rather than m17? You use f.ego when fitting the model. 2. The egocentric ERGM implemented in ergm.ego assumes the network is undirected. Thus, you'd need to use the undirected variants of the terms such as nodefactor only (not nodeifactor and nodeofactor separately). 3. Based on what you send for (1) I should be able to tell what's the problem with ppopsize and popsize. Michal On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: Hello Michal, thank you for your quick answer ! Here the call : m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) summary(m17) I tried again adding the actual ppopsize : 990 and I get a new message error : Constructing pseudopopulation network. Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > summary(m17) Error: object 'm17' not found This message appears for each of the terms. The study focuses on a territorial agrifood system, including a diversity of actors (farmers, elected officials, civil society, advisors...) who have different visions of agriculture. In order to understand sustainable agrifood transformations, I try to see if the different groups holding different visions interact and how is structured the network (ERGM). I used a snowball sampling starting from 15 actors, central and in each typology. best regards, Carine Pachoud ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:13 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud [https://s-install.avcdn.net/ipm/preview/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif] Sans virus.www.avast.com _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Thu Nov 23 07:12:30 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Thanks. The output of summary() suggests that you don't have any nodal attribute variables in the `f.ego` object. No `Type` variable in particular, so ergm.ego is not finding them when fitting the model. Can you send the output of `str(f.ego)`? Are you sure you've produced that object correctly? On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud wrote: > Dear Micha?, > > here the summary for f.ego: > > > summary(f.ego)55 Egos/ Ego Networks > 430 Alters > Min. Netsize 1 > Average Netsize 7.81818181818182 > Max. Netsize 31 > Average Density 0.674972800136199 > Alter survey design: > Maximum nominations: Inf > > > I tried on undirected network : > m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ > nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ > nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov > ("LienE")+ nodefactor ("View")+ nodematch("View") + > gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = > control.ergm.ego(ppopsize=990)) > > and now I get the following message: > > Constructing pseudopopulation network.*Error** in `ergm_Init_abort()`:*! In unknown function: ?Type? is/are not valid nodal attribute(s).Run `*rlang::last_trace()*` to see where the error occurred.> rlang::last_trace()***Error** in **`ergm_Init_abort()`**:*! In unknown function: ?Type? is/are not valid nodal attribute(s). > > > I do not understand why the attribute is invalid (I tried with the package > ERGM and it worked). > > thank you for the reference, I will read it! > > best regards, > Carine > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:54 > *? :* Carine Pachoud > *Objet :* Re: [statnet_help] ego.ergm popsize > > ... and: > > 4. The fact that you used snowball sampling to collect data makes > inference with ergm.ego questionable because it assumes that the data is a > probabilistic sample of egos. You may need a different model such as > https://doi.org/10.1016/j.socnet.2015.11.003. > > On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: > > Thanks Carine. Couple of issues/questions based on what you sent: > > 1. Can you please send the output of summary(f.ego) rather than m17? You > use f.ego when fitting the model. > 2. The egocentric ERGM implemented in ergm.ego assumes the network is > undirected. Thus, you'd need to use the undirected variants of the terms > such as nodefactor only (not nodeifactor and nodeofactor separately). > 3. Based on what you send for (1) I should be able to tell what's the > problem with ppopsize and popsize. > > Michal > > On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: > > Hello Michal, > > thank you for your quick answer ! > > Here the call : > m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor > ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ > nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ > nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ > nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor > ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp > (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) > summary(m17) > > I tried again adding the actual ppopsize : 990 and I get a new message > error : > > Constructing pseudopopulation network. > Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > > > summary(m17)Error: object 'm17' not found > > > This message appears for each of the terms. > > The study focuses on a territorial agrifood system, including a diversity > of actors (farmers, elected officials, civil society, advisors...) who have > different visions of agriculture. In order to understand sustainable > agrifood transformations, I try to see if the different groups holding > different visions interact and how is structured the network (ERGM). I used > a snowball sampling starting from 15 actors, central and in each typology. > > best regards, > Carine Pachoud > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:13 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Hello Carine, > > Can you please send the exact offending call that is triggering the error > together with the output of calling summary() on your data object? > > In general, population size (as opposed to the pseudo-population size) > should correspond to the size of the population network (which is not > observed in its entirety) you want to make inferences about and from which > you sampled the egos. This inference is the main goal of the model > implemented in ergm.ego. If such a population network does not exist or is > hard to conceive as a theoretical device it might be difficult to interpret > (some of) the results. What's the purpose of your analysis? > > Best, > Micha? > > > On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: > > Dear ERGM users, > > I previously worked with ERGM on complete networks. Today I am dealing > with an egocentric network and the package ego.ergm. I have 55 nodes et 274 > edges. I do not know the size of the population. > > It's certainly a very simple adjustment to make but I'm stuck on the > population size. When I want to run my model, I get the following error > message: > > Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : > missing value where TRUE/FALSE needed > > I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. > > Could you help me ? > > thanks a lot! > > best regards, > Carine Pachoud > > > > > > Sans virus.www.avast.com > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From carine.pachoud at hotmail.fr Thu Nov 23 07:22:37 2023 From: carine.pachoud at hotmail.fr (Carine Pachoud) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: here str(f.ego): List of 3 $ ego : tibble [55 ? 2] (S3: tbl_df/tbl/data.frame) ..$ .egoID : int [1:55] 1 2 3 4 5 6 7 8 9 10 ... ..$ vertex.names: chr [1:55] "1" "10" "11" "12" ... $ alter: tibble [430 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .altID : int [1:430] 30 25 18 10 36 17 16 3 51 32 ... ..$ .egoID : int [1:430] 1 1 1 1 1 1 1 1 2 2 ... ..$ vertex.names: chr [1:430] "36" "31" "25" "18" ... $ aatie: tibble [1,140 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .egoID: int [1:1140] 1 1 1 1 1 1 1 1 1 1 ... ..$ .srcID: int [1:1140] 30 25 25 25 25 18 18 18 10 10 ... ..$ .tgtID: int [1:1140] 25 18 10 16 30 17 10 25 3 25 ... - attr(*, "class")= chr [1:2] "egor" "list" - attr(*, "alter_design")=List of 1 ..$ max: num Inf - attr(*, "active")= chr "ego" Ok it is surely different from ERGM. Do you know the commands to add node attributes with ego.ergm (numeric and character) ? thanks! Carine ________________________________ De : Micha? Bojanowski Envoy? : jeudi 23 novembre 2023 12:12 ? : Carine Pachoud Cc : statnet_help@u.washington.edu Objet : Re: [statnet_help] ego.ergm popsize Thanks. The output of summary() suggests that you don't have any nodal attribute variables in the `f.ego` object. No `Type` variable in particular, so ergm.ego is not finding them when fitting the model. Can you send the output of `str(f.ego)`? Are you sure you've produced that object correctly? On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud > wrote: Dear Micha?, here the summary for f.ego: > summary(f.ego) 55 Egos/ Ego Networks 430 Alters Min. Netsize 1 Average Netsize 7.81818181818182 Max. Netsize 31 Average Density 0.674972800136199 Alter survey design: Maximum nominations: Inf I tried on undirected network : m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov ("LienE")+ nodefactor ("View")+ nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) and now I get the following message: Constructing pseudopopulation network. Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). Run `rlang::last_trace()` to see where the error occurred.> rlang::last_trace() Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). I do not understand why the attribute is invalid (I tried with the package ERGM and it worked). thank you for the reference, I will read it! best regards, Carine ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:54 ? : Carine Pachoud > Objet : Re: [statnet_help] ego.ergm popsize ... and: 4. The fact that you used snowball sampling to collect data makes inference with ergm.ego questionable because it assumes that the data is a probabilistic sample of egos. You may need a different model such as https://doi.org/10.1016/j.socnet.2015.11.003. On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: Thanks Carine. Couple of issues/questions based on what you sent: 1. Can you please send the output of summary(f.ego) rather than m17? You use f.ego when fitting the model. 2. The egocentric ERGM implemented in ergm.ego assumes the network is undirected. Thus, you'd need to use the undirected variants of the terms such as nodefactor only (not nodeifactor and nodeofactor separately). 3. Based on what you send for (1) I should be able to tell what's the problem with ppopsize and popsize. Michal On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: Hello Michal, thank you for your quick answer ! Here the call : m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) summary(m17) I tried again adding the actual ppopsize : 990 and I get a new message error : Constructing pseudopopulation network. Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > summary(m17) Error: object 'm17' not found This message appears for each of the terms. The study focuses on a territorial agrifood system, including a diversity of actors (farmers, elected officials, civil society, advisors...) who have different visions of agriculture. In order to understand sustainable agrifood transformations, I try to see if the different groups holding different visions interact and how is structured the network (ERGM). I used a snowball sampling starting from 15 actors, central and in each typology. best regards, Carine Pachoud ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:13 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud [https://s-install.avcdn.net/ipm/preview/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif] Sans virus.www.avast.com _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Thu Nov 23 07:27:49 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Yeah, the attribute variables are not in there. How did you create that object? Is your data originally as a `network` object? It will be easier to recreate it properly from the source rather than re-add what's missing. On Thu, Nov 23, 2023 at 4:22?PM Carine Pachoud wrote: > here str(f.ego): > > List of 3 $ ego : tibble [55 ? 2] (S3: tbl_df/tbl/data.frame) ..$ .egoID > : int [1:55] 1 2 3 4 5 6 7 8 9 10 ... ..$ vertex.names: chr [1:55] "1" "10" > "11" "12" ... $ alter: tibble [430 ? 3] (S3: tbl_df/tbl/data.frame) ..$ > .altID : int [1:430] 30 25 18 10 36 17 16 3 51 32 ... ..$ .egoID : int > [1:430] 1 1 1 1 1 1 1 1 2 2 ... ..$ vertex.names: chr [1:430] "36" "31" > "25" "18" ... $ aatie: tibble [1,140 ? 3] (S3: tbl_df/tbl/data.frame) ..$ > .egoID: int [1:1140] 1 1 1 1 1 1 1 1 1 1 ... ..$ .srcID: int [1:1140] 30 25 > 25 25 25 18 18 18 10 10 ... ..$ .tgtID: int [1:1140] 25 18 10 16 30 17 10 > 25 3 25 ... - attr(*, "class")= chr [1:2] "egor" "list" - attr(*, > "alter_design")=List of 1 ..$ max: num Inf - attr(*, "active")= chr "ego" > > Ok it is surely different from ERGM. Do you know the commands to add node > attributes with ego.ergm (numeric and character) ? > > thanks! > Carine > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 12:12 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Thanks. The output of summary() suggests that you don't have any nodal > attribute variables in the `f.ego` object. No `Type` variable in > particular, so ergm.ego is not finding them when fitting the model. Can you > send the output of `str(f.ego)`? Are you sure you've produced that object > correctly? > > On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud > wrote: > > Dear Micha?, > > here the summary for f.ego: > > > summary(f.ego)55 Egos/ Ego Networks > 430 Alters > Min. Netsize 1 > Average Netsize 7.81818181818182 > Max. Netsize 31 > Average Density 0.674972800136199 > Alter survey design: > Maximum nominations: Inf > > > I tried on undirected network : > m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ > nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ > nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov > ("LienE")+ nodefactor ("View")+ nodematch("View") + > gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = > control.ergm.ego(ppopsize=990)) > > and now I get the following message: > > Constructing pseudopopulation network.*Error** in `ergm_Init_abort()`:*! In unknown function: ?Type? is/are not valid nodal attribute(s).Run `*rlang::last_trace()*` to see where the error occurred.> rlang::last_trace()***Error** in **`ergm_Init_abort()`**:*! In unknown function: ?Type? is/are not valid nodal attribute(s). > > > I do not understand why the attribute is invalid (I tried with the package > ERGM and it worked). > > thank you for the reference, I will read it! > > best regards, > Carine > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:54 > *? :* Carine Pachoud > *Objet :* Re: [statnet_help] ego.ergm popsize > > ... and: > > 4. The fact that you used snowball sampling to collect data makes > inference with ergm.ego questionable because it assumes that the data is a > probabilistic sample of egos. You may need a different model such as > https://doi.org/10.1016/j.socnet.2015.11.003. > > On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: > > Thanks Carine. Couple of issues/questions based on what you sent: > > 1. Can you please send the output of summary(f.ego) rather than m17? You > use f.ego when fitting the model. > 2. The egocentric ERGM implemented in ergm.ego assumes the network is > undirected. Thus, you'd need to use the undirected variants of the terms > such as nodefactor only (not nodeifactor and nodeofactor separately). > 3. Based on what you send for (1) I should be able to tell what's the > problem with ppopsize and popsize. > > Michal > > On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: > > Hello Michal, > > thank you for your quick answer ! > > Here the call : > m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor > ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ > nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ > nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ > nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor > ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp > (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) > summary(m17) > > I tried again adding the actual ppopsize : 990 and I get a new message > error : > > Constructing pseudopopulation network. > Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > > > summary(m17)Error: object 'm17' not found > > > This message appears for each of the terms. > > The study focuses on a territorial agrifood system, including a diversity > of actors (farmers, elected officials, civil society, advisors...) who have > different visions of agriculture. In order to understand sustainable > agrifood transformations, I try to see if the different groups holding > different visions interact and how is structured the network (ERGM). I used > a snowball sampling starting from 15 actors, central and in each typology. > > best regards, > Carine Pachoud > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:13 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Hello Carine, > > Can you please send the exact offending call that is triggering the error > together with the output of calling summary() on your data object? > > In general, population size (as opposed to the pseudo-population size) > should correspond to the size of the population network (which is not > observed in its entirety) you want to make inferences about and from which > you sampled the egos. This inference is the main goal of the model > implemented in ergm.ego. If such a population network does not exist or is > hard to conceive as a theoretical device it might be difficult to interpret > (some of) the results. What's the purpose of your analysis? > > Best, > Micha? > > > On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: > > Dear ERGM users, > > I previously worked with ERGM on complete networks. Today I am dealing > with an egocentric network and the package ego.ergm. I have 55 nodes et 274 > edges. I do not know the size of the population. > > It's certainly a very simple adjustment to make but I'm stuck on the > population size. When I want to run my model, I get the following error > message: > > Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : > missing value where TRUE/FALSE needed > > I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. > > Could you help me ? > > thanks a lot! > > best regards, > Carine Pachoud > > > > > > Sans virus.www.avast.com > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From carine.pachoud at hotmail.fr Thu Nov 23 07:36:09 2023 From: carine.pachoud at hotmail.fr (Carine Pachoud) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Ok. The nodes and the attributes are in a csv file and the edge list in another csv file. As I'm more familiar with complete networks and the ERGM package, I created the object in the same way and tried to integrate the attributes on the ERGM package model # Load links lien<-read.table("C:/Users/admin/Dropbox/Transformont/network/interne-discussion.csv", sep=";", header=TRUE) lien[,1]<-as.character(lien[,1]) lien[,2]<-as.character(lien[,2]) # Load nodes (actors) attributes noeud<-read.table("C:/Users/admin/Dropbox/Transformont/network/node-interviewed.csv", sep=";", header=TRUE, stringsAsFactors=FALSE) row.names(noeud)<-as.character(noeud$node) nrow(noeud) # create the network object f <- as.network(as.matrix(lien[,c(1:2)]), directed=F, loops = F, matrix.type="edgelist") f plot(f) f.ego<-as.egor(f) f.ego f.ego$ego f.ego$alter # Attach nodes attributes f.ego %v% "Nom" <- as.character(noeud[,"nom"]) ... ________________________________ De : Micha? Bojanowski Envoy? : jeudi 23 novembre 2023 12:27 ? : Carine Pachoud Cc : statnet_help@u.washington.edu Objet : Re: [statnet_help] ego.ergm popsize Yeah, the attribute variables are not in there. How did you create that object? Is your data originally as a `network` object? It will be easier to recreate it properly from the source rather than re-add what's missing. On Thu, Nov 23, 2023 at 4:22?PM Carine Pachoud > wrote: here str(f.ego): List of 3 $ ego : tibble [55 ? 2] (S3: tbl_df/tbl/data.frame) ..$ .egoID : int [1:55] 1 2 3 4 5 6 7 8 9 10 ... ..$ vertex.names: chr [1:55] "1" "10" "11" "12" ... $ alter: tibble [430 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .altID : int [1:430] 30 25 18 10 36 17 16 3 51 32 ... ..$ .egoID : int [1:430] 1 1 1 1 1 1 1 1 2 2 ... ..$ vertex.names: chr [1:430] "36" "31" "25" "18" ... $ aatie: tibble [1,140 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .egoID: int [1:1140] 1 1 1 1 1 1 1 1 1 1 ... ..$ .srcID: int [1:1140] 30 25 25 25 25 18 18 18 10 10 ... ..$ .tgtID: int [1:1140] 25 18 10 16 30 17 10 25 3 25 ... - attr(*, "class")= chr [1:2] "egor" "list" - attr(*, "alter_design")=List of 1 ..$ max: num Inf - attr(*, "active")= chr "ego" Ok it is surely different from ERGM. Do you know the commands to add node attributes with ego.ergm (numeric and character) ? thanks! Carine ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 12:12 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Thanks. The output of summary() suggests that you don't have any nodal attribute variables in the `f.ego` object. No `Type` variable in particular, so ergm.ego is not finding them when fitting the model. Can you send the output of `str(f.ego)`? Are you sure you've produced that object correctly? On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud > wrote: Dear Micha?, here the summary for f.ego: > summary(f.ego) 55 Egos/ Ego Networks 430 Alters Min. Netsize 1 Average Netsize 7.81818181818182 Max. Netsize 31 Average Density 0.674972800136199 Alter survey design: Maximum nominations: Inf I tried on undirected network : m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov ("LienE")+ nodefactor ("View")+ nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) and now I get the following message: Constructing pseudopopulation network. Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). Run `rlang::last_trace()` to see where the error occurred.> rlang::last_trace() Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). I do not understand why the attribute is invalid (I tried with the package ERGM and it worked). thank you for the reference, I will read it! best regards, Carine ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:54 ? : Carine Pachoud > Objet : Re: [statnet_help] ego.ergm popsize ... and: 4. The fact that you used snowball sampling to collect data makes inference with ergm.ego questionable because it assumes that the data is a probabilistic sample of egos. You may need a different model such as https://doi.org/10.1016/j.socnet.2015.11.003. On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: Thanks Carine. Couple of issues/questions based on what you sent: 1. Can you please send the output of summary(f.ego) rather than m17? You use f.ego when fitting the model. 2. The egocentric ERGM implemented in ergm.ego assumes the network is undirected. Thus, you'd need to use the undirected variants of the terms such as nodefactor only (not nodeifactor and nodeofactor separately). 3. Based on what you send for (1) I should be able to tell what's the problem with ppopsize and popsize. Michal On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: Hello Michal, thank you for your quick answer ! Here the call : m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) summary(m17) I tried again adding the actual ppopsize : 990 and I get a new message error : Constructing pseudopopulation network. Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > summary(m17) Error: object 'm17' not found This message appears for each of the terms. The study focuses on a territorial agrifood system, including a diversity of actors (farmers, elected officials, civil society, advisors...) who have different visions of agriculture. In order to understand sustainable agrifood transformations, I try to see if the different groups holding different visions interact and how is structured the network (ERGM). I used a snowball sampling starting from 15 actors, central and in each typology. best regards, Carine Pachoud ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:13 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud [https://s-install.avcdn.net/ipm/preview/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif] Sans virus.www.avast.com _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From michal2992 at gmail.com Thu Nov 23 07:49:25 2023 From: michal2992 at gmail.com (=?UTF-8?Q?Micha=C5=82_Bojanowski?=) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: Nope, `%v%` will not work that way. Technically it might be easiest to construct the 'egor' object directly from the edge list and attribute CSV like in the example at https://egor.tillt.net/articles/using_egor.html Still, a more serious issue is that effectively every person in your data becomes an ego, while in your snowball sample, I presume, this was not the case. This leads me even more in the direction that the egocentric ERGM does not seem appropriate for your data. How many "waves" did the snowball have? On Thu, Nov 23, 2023 at 4:36?PM Carine Pachoud wrote: > Ok. The nodes and the attributes are in a csv file and the edge list in > another csv file. As I'm more familiar with complete networks and the ERGM > package, I created the object in the same way and tried to integrate the > attributes on the ERGM package model > > > # Load links > lien<-read.table("C:/Users/admin/Dropbox/Transformont/network/interne-discussion.csv", > sep=";", header=TRUE) > lien[,1]<-as.character(lien[,1]) > lien[,2]<-as.character(lien[,2]) > > # Load nodes (actors) attributes > noeud<-read.table("C:/Users/admin/Dropbox/Transformont/network/node-interviewed.csv", > sep=";", header=TRUE, stringsAsFactors=FALSE) > row.names(noeud)<-as.character(noeud$node) > nrow(noeud) > > # create the network object > f <- as.network(as.matrix(lien[,c(1:2)]), directed=F, loops = F, > matrix.type="edgelist") > f > plot(f) > > f.ego<-as.egor(f) > f.ego > f.ego$ego > f.ego$alter > > # Attach nodes attributes > f.ego %v% "Nom" <- as.character(noeud[,"nom"]) > ... > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 12:27 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Yeah, the attribute variables are not in there. How did you create that > object? Is your data originally as a `network` object? It will be easier to > recreate it properly from the source rather than re-add what's missing. > > On Thu, Nov 23, 2023 at 4:22?PM Carine Pachoud > wrote: > > here str(f.ego): > > List of 3 $ ego : tibble [55 ? 2] (S3: tbl_df/tbl/data.frame) ..$ .egoID > : int [1:55] 1 2 3 4 5 6 7 8 9 10 ... ..$ vertex.names: chr [1:55] "1" "10" > "11" "12" ... $ alter: tibble [430 ? 3] (S3: tbl_df/tbl/data.frame) ..$ > .altID : int [1:430] 30 25 18 10 36 17 16 3 51 32 ... ..$ .egoID : int > [1:430] 1 1 1 1 1 1 1 1 2 2 ... ..$ vertex.names: chr [1:430] "36" "31" > "25" "18" ... $ aatie: tibble [1,140 ? 3] (S3: tbl_df/tbl/data.frame) ..$ > .egoID: int [1:1140] 1 1 1 1 1 1 1 1 1 1 ... ..$ .srcID: int [1:1140] 30 25 > 25 25 25 18 18 18 10 10 ... ..$ .tgtID: int [1:1140] 25 18 10 16 30 17 10 > 25 3 25 ... - attr(*, "class")= chr [1:2] "egor" "list" - attr(*, > "alter_design")=List of 1 ..$ max: num Inf - attr(*, "active")= chr "ego" > > Ok it is surely different from ERGM. Do you know the commands to add node > attributes with ego.ergm (numeric and character) ? > > thanks! > Carine > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 12:12 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Thanks. The output of summary() suggests that you don't have any nodal > attribute variables in the `f.ego` object. No `Type` variable in > particular, so ergm.ego is not finding them when fitting the model. Can you > send the output of `str(f.ego)`? Are you sure you've produced that object > correctly? > > On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud > wrote: > > Dear Micha?, > > here the summary for f.ego: > > > summary(f.ego)55 Egos/ Ego Networks > 430 Alters > Min. Netsize 1 > Average Netsize 7.81818181818182 > Max. Netsize 31 > Average Density 0.674972800136199 > Alter survey design: > Maximum nominations: Inf > > > I tried on undirected network : > m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ > nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ > nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov > ("LienE")+ nodefactor ("View")+ nodematch("View") + > gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = > control.ergm.ego(ppopsize=990)) > > and now I get the following message: > > Constructing pseudopopulation network.*Error** in `ergm_Init_abort()`:*! In unknown function: ?Type? is/are not valid nodal attribute(s).Run `*rlang::last_trace()*` to see where the error occurred.> rlang::last_trace()***Error** in **`ergm_Init_abort()`**:*! In unknown function: ?Type? is/are not valid nodal attribute(s). > > > I do not understand why the attribute is invalid (I tried with the package > ERGM and it worked). > > thank you for the reference, I will read it! > > best regards, > Carine > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:54 > *? :* Carine Pachoud > *Objet :* Re: [statnet_help] ego.ergm popsize > > ... and: > > 4. The fact that you used snowball sampling to collect data makes > inference with ergm.ego questionable because it assumes that the data is a > probabilistic sample of egos. You may need a different model such as > https://doi.org/10.1016/j.socnet.2015.11.003. > > On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: > > Thanks Carine. Couple of issues/questions based on what you sent: > > 1. Can you please send the output of summary(f.ego) rather than m17? You > use f.ego when fitting the model. > 2. The egocentric ERGM implemented in ergm.ego assumes the network is > undirected. Thus, you'd need to use the undirected variants of the terms > such as nodefactor only (not nodeifactor and nodeofactor separately). > 3. Based on what you send for (1) I should be able to tell what's the > problem with ppopsize and popsize. > > Michal > > On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: > > Hello Michal, > > thank you for your quick answer ! > > Here the call : > m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor > ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ > nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ > nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ > nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor > ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp > (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) > summary(m17) > > I tried again adding the actual ppopsize : 990 and I get a new message > error : > > Constructing pseudopopulation network. > Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > > > summary(m17)Error: object 'm17' not found > > > This message appears for each of the terms. > > The study focuses on a territorial agrifood system, including a diversity > of actors (farmers, elected officials, civil society, advisors...) who have > different visions of agriculture. In order to understand sustainable > agrifood transformations, I try to see if the different groups holding > different visions interact and how is structured the network (ERGM). I used > a snowball sampling starting from 15 actors, central and in each typology. > > best regards, > Carine Pachoud > > > ------------------------------ > *De :* Micha? Bojanowski > *Envoy? :* jeudi 23 novembre 2023 11:13 > *? :* Carine Pachoud > *Cc :* statnet_help@u.washington.edu > *Objet :* Re: [statnet_help] ego.ergm popsize > > Hello Carine, > > Can you please send the exact offending call that is triggering the error > together with the output of calling summary() on your data object? > > In general, population size (as opposed to the pseudo-population size) > should correspond to the size of the population network (which is not > observed in its entirety) you want to make inferences about and from which > you sampled the egos. This inference is the main goal of the model > implemented in ergm.ego. If such a population network does not exist or is > hard to conceive as a theoretical device it might be difficult to interpret > (some of) the results. What's the purpose of your analysis? > > Best, > Micha? > > > On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: > > Dear ERGM users, > > I previously worked with ERGM on complete networks. Today I am dealing > with an egocentric network and the package ego.ergm. I have 55 nodes et 274 > edges. I do not know the size of the population. > > It's certainly a very simple adjustment to make but I'm stuck on the > population size. When I want to run my model, I get the following error > message: > > Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : > missing value where TRUE/FALSE needed > > I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. > > Could you help me ? > > thanks a lot! > > best regards, > Carine Pachoud > > > > > > Sans virus.www.avast.com > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From carine.pachoud at hotmail.fr Thu Nov 23 07:57:52 2023 From: carine.pachoud at hotmail.fr (Carine Pachoud) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] ego.ergm popsize In-Reply-To: References: Message-ID: ok thank you. Sorry this tool is new for me. I have 4 waves. I have two tables of nodes, one with all the nodes (egos and alters) and one with the egos only (the 55 that I have interviewed). My question is how can I use ego.ergm if I do not have information on the alters (their attributes) ? regards, Carine ________________________________ De : Micha? Bojanowski Envoy? : jeudi 23 novembre 2023 12:49 ? : Carine Pachoud Cc : statnet_help@u.washington.edu Objet : Re: [statnet_help] ego.ergm popsize Nope, `%v%` will not work that way. Technically it might be easiest to construct the 'egor' object directly from the edge list and attribute CSV like in the example at https://egor.tillt.net/articles/using_egor.html Still, a more serious issue is that effectively every person in your data becomes an ego, while in your snowball sample, I presume, this was not the case. This leads me even more in the direction that the egocentric ERGM does not seem appropriate for your data. How many "waves" did the snowball have? On Thu, Nov 23, 2023 at 4:36?PM Carine Pachoud > wrote: Ok. The nodes and the attributes are in a csv file and the edge list in another csv file. As I'm more familiar with complete networks and the ERGM package, I created the object in the same way and tried to integrate the attributes on the ERGM package model # Load links lien<-read.table("C:/Users/admin/Dropbox/Transformont/network/interne-discussion.csv", sep=";", header=TRUE) lien[,1]<-as.character(lien[,1]) lien[,2]<-as.character(lien[,2]) # Load nodes (actors) attributes noeud<-read.table("C:/Users/admin/Dropbox/Transformont/network/node-interviewed.csv", sep=";", header=TRUE, stringsAsFactors=FALSE) row.names(noeud)<-as.character(noeud$node) nrow(noeud) # create the network object f <- as.network(as.matrix(lien[,c(1:2)]), directed=F, loops = F, matrix.type="edgelist") f plot(f) f.ego<-as.egor(f) f.ego f.ego$ego f.ego$alter # Attach nodes attributes f.ego %v% "Nom" <- as.character(noeud[,"nom"]) ... ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 12:27 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Yeah, the attribute variables are not in there. How did you create that object? Is your data originally as a `network` object? It will be easier to recreate it properly from the source rather than re-add what's missing. On Thu, Nov 23, 2023 at 4:22?PM Carine Pachoud > wrote: here str(f.ego): List of 3 $ ego : tibble [55 ? 2] (S3: tbl_df/tbl/data.frame) ..$ .egoID : int [1:55] 1 2 3 4 5 6 7 8 9 10 ... ..$ vertex.names: chr [1:55] "1" "10" "11" "12" ... $ alter: tibble [430 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .altID : int [1:430] 30 25 18 10 36 17 16 3 51 32 ... ..$ .egoID : int [1:430] 1 1 1 1 1 1 1 1 2 2 ... ..$ vertex.names: chr [1:430] "36" "31" "25" "18" ... $ aatie: tibble [1,140 ? 3] (S3: tbl_df/tbl/data.frame) ..$ .egoID: int [1:1140] 1 1 1 1 1 1 1 1 1 1 ... ..$ .srcID: int [1:1140] 30 25 25 25 25 18 18 18 10 10 ... ..$ .tgtID: int [1:1140] 25 18 10 16 30 17 10 25 3 25 ... - attr(*, "class")= chr [1:2] "egor" "list" - attr(*, "alter_design")=List of 1 ..$ max: num Inf - attr(*, "active")= chr "ego" Ok it is surely different from ERGM. Do you know the commands to add node attributes with ego.ergm (numeric and character) ? thanks! Carine ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 12:12 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Thanks. The output of summary() suggests that you don't have any nodal attribute variables in the `f.ego` object. No `Type` variable in particular, so ergm.ego is not finding them when fitting the model. Can you send the output of `str(f.ego)`? Are you sure you've produced that object correctly? On Thu, Nov 23, 2023 at 4:06?PM Carine Pachoud > wrote: Dear Micha?, here the summary for f.ego: > summary(f.ego) 55 Egos/ Ego Networks 430 Alters Min. Netsize 1 Average Netsize 7.81818181818182 Max. Netsize 31 Average Density 0.674972800136199 Alter survey design: Maximum nominations: Inf I tried on undirected network : m17 <- ergm.ego(f.ego ~ edges+ nodefactor ("Type")+ nodematch("Type")+ nodefactor ("Date")+ nodematch ("Date")+ nodefactor("Commune")+ nodematch("Commune")+ nodefactor ("Mairie")+ nodematch("Mairie")+ nodecov ("LienE")+ nodefactor ("View")+ nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) and now I get the following message: Constructing pseudopopulation network. Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). Run `rlang::last_trace()` to see where the error occurred.> rlang::last_trace() Error in `ergm_Init_abort()`:! In unknown function: ?Type? is/are not valid nodal attribute(s). I do not understand why the attribute is invalid (I tried with the package ERGM and it worked). thank you for the reference, I will read it! best regards, Carine ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:54 ? : Carine Pachoud > Objet : Re: [statnet_help] ego.ergm popsize ... and: 4. The fact that you used snowball sampling to collect data makes inference with ergm.ego questionable because it assumes that the data is a probabilistic sample of egos. You may need a different model such as https://doi.org/10.1016/j.socnet.2015.11.003. On Thu, Nov 23, 2023 at 3:47?PM Micha? Bojanowski > wrote: Thanks Carine. Couple of issues/questions based on what you sent: 1. Can you please send the output of summary(f.ego) rather than m17? You use f.ego when fitting the model. 2. The egocentric ERGM implemented in ergm.ego assumes the network is undirected. Thus, you'd need to use the undirected variants of the terms such as nodefactor only (not nodeifactor and nodeofactor separately). 3. Based on what you send for (1) I should be able to tell what's the problem with ppopsize and popsize. Michal On Thu, Nov 23, 2023 at 3:43?PM Carine Pachoud > wrote: Hello Michal, thank you for your quick answer ! Here the call : m17 <- ergm.ego(f.ego ~ edges+ nodeifactor ("Type")+ nodeofactor ("Type")+nodematch("Type")+ nodeifactor ("Date")+ nodeofactor ("Date")+ nodematch ("Date")+ nodeifactor("Commune")+nodeofactor("Commune")+ nodematch("Commune")+ nodeifactor ("Mairie")+ nodeofactor("Mairie")+ nodematch("Mairie")+ nodeicov ("LienE")+ nodeocov ("LienE")+nodeifactor ("View")+ nodeofactor ("View")+nodematch("View") + gwesp(0.1,fixed=T)+gwdsp (0.1,fixed=T),control = control.ergm.ego(ppopsize=990)) summary(m17) I tried again adding the actual ppopsize : 990 and I get a new message error : Constructing pseudopopulation network. Error: Egocentric statistic ?nodeifactor? function ?EgoStat.nodeifactor? not found. > summary(m17) Error: object 'm17' not found This message appears for each of the terms. The study focuses on a territorial agrifood system, including a diversity of actors (farmers, elected officials, civil society, advisors...) who have different visions of agriculture. In order to understand sustainable agrifood transformations, I try to see if the different groups holding different visions interact and how is structured the network (ERGM). I used a snowball sampling starting from 15 actors, central and in each typology. best regards, Carine Pachoud ________________________________ De : Micha? Bojanowski > Envoy? : jeudi 23 novembre 2023 11:13 ? : Carine Pachoud > Cc : statnet_help@u.washington.edu > Objet : Re: [statnet_help] ego.ergm popsize Hello Carine, Can you please send the exact offending call that is triggering the error together with the output of calling summary() on your data object? In general, population size (as opposed to the pseudo-population size) should correspond to the size of the population network (which is not observed in its entirety) you want to make inferences about and from which you sampled the egos. This inference is the main goal of the model implemented in ergm.ego. If such a population network does not exist or is hard to conceive as a theoretical device it might be difficult to interpret (some of) the results. What's the purpose of your analysis? Best, Micha? On Thu, Nov 23, 2023 at 2:51?PM Carine Pachoud > wrote: Dear ERGM users, I previously worked with ERGM on complete networks. Today I am dealing with an egocentric network and the package ego.ergm. I have 55 nodes et 274 edges. I do not know the size of the population. It's certainly a very simple adjustment to make but I'm stuck on the population size. When I want to run my model, I get the following error message: Error in if (ppopsize < sampsize && !is.data.frame(control$ppopsize)) warning("Using a smaller pseudopopulation size than sample size usually does not make sense.") else if (ppopsize == : missing value where TRUE/FALSE needed I tried to fix the ppopsize with control = control.ergm.ego(ppopsize=1000) and also to fix the popsize (popsize=N, control=snctrl(ppopsize=N)), based on your tutorials, but I still have this error message. Could you help me ? thanks a lot! best regards, Carine Pachoud [https://s-install.avcdn.net/ipm/preview/icons/icon-envelope-tick-round-orange-animated-no-repeat-v1.gif] Sans virus.www.avast.com _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From waxenecker at fss.muni.cz Thu Nov 30 09:58:31 2023 From: waxenecker at fss.muni.cz (Harald Waxenecker) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... Message-ID: Dear ?statnet community?, Our research focuses on tie formation and elite cohesion, specifically examining interlocking directorates and kinship relations. The dependent bipartite business network comprises 6,902 individuals and 5,178 companies, exhibiting sparsity (density = 0.00012) and fragmentation with 4,455 components, including 3,850 isolates in the first mode (persons). The attached documents contain descriptives and the component size distribution from the observed network. The fragmented structure is important, as other network layers, like kinship relations, are expected to contribute to the cohesion of this business network. We apply ERGM to model these processes, but we struggle to capture the fragmented structure of the observed network. The component size distribution of the simulated network differs significantly. In addition, the goodness-of-fit (GOF) for k-stars (in both modes) and geodesic distances (Inf) shows significant results. All these results are also attached. We've explored various options, including constraints, MCMC propositions, and simulated annealing, but haven't achieved success. Please, we would like to ask for your help to improve our model. Thank you! Kind regards, Harald --- Harald Waxenecker Masaryk University | Faculty of social studies Department of Environment Studies A: Jostova 10 | 602 00 Brno | Czech Republic E: waxenecker@fss.muni.cz -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: component.size.dist.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: model0.gof.pdf Type: application/pdf Size: 55632 bytes Desc: model0.gof.pdf URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: model0.gof.print.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: model0.mcmc.pdf Type: application/pdf Size: 2042132 bytes Desc: model0.mcmc.pdf URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: model0.summary.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: observed_network.png Type: image/png Size: 59419 bytes Desc: observed_network.png URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: observed.network.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: simulated_network.png Type: image/png Size: 36506 bytes Desc: simulated_network.png URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: simulated.network.txt URL: From kraft.tom at gmail.com Thu Nov 30 11:10:04 2023 From: kraft.tom at gmail.com (Tom Kraft) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: Hi Harald, I do not fully understand what makes this a bipartite network based on the description you have given, but given your attached graphs of the observed and simulated networks it seems likely that you are failing to capture important homophily processes. One example would be homophily within companies (or some other trait), which seems like an obvious candidate that could lead to the distinct clustering you see in your observed network. Similarly there may be other important individual attributes beyond the structural properties you have included in your model. Forgive me if I am misunderstanding your question, Tom On Thu, Nov 30, 2023 at 10:59?AM Harald Waxenecker wrote: > Dear ?statnet community?, > > > > Our research focuses on tie formation and elite cohesion, specifically > examining interlocking directorates and kinship relations. The dependent > bipartite business network comprises 6,902 individuals and 5,178 companies, > exhibiting sparsity (density = 0.00012) and fragmentation with 4,455 > components, including 3,850 isolates in the first mode (persons). The > attached documents contain descriptives and the component size distribution > from the observed network. > > > > The fragmented structure is important, as other network layers, like > kinship relations, are expected to contribute to the cohesion of this > business network. We apply ERGM to model these processes, but we struggle > to capture the fragmented structure of the observed network. The component > size distribution of the simulated network differs significantly. In > addition, the goodness-of-fit (GOF) for k-stars (in both modes) and > geodesic distances (Inf) shows significant results. All these results are > also attached. > > > > We've explored various options, including constraints, MCMC propositions, > and simulated annealing, but haven't achieved success. Please, we would > like to ask for your help to improve our model. Thank you! > > > > Kind regards, > > Harald > > > > > > > > --- > > > > *Harald Waxenecker * > > *Masaryk University | Faculty of social studies* > Department of Environment Studies > A: Jostova 10 | 602 00 Brno | Czech Republic > E: waxenecker@fss.muni.cz > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From morrism at uw.edu Thu Nov 30 11:26:36 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: Hi Harald, Thanks for reaching out, and for sending all the info. That's really helpful for us in answering your questions. I haven't used the bipartite modeling options for a while, as we have transitioned to using a different approach in our own research. But based on the simulated network from the model, I would guess that the bipartite constraint is not operating properly -- and that's why you're getting a very dense giant component. Have you verified that there are no cross-group ties in the simulated data? I also think your model is probably not well-specified, in the sense that it relies exclusively on dyad-dependent terms. Models like this are known to produce "near degenerate" networks -- that is, networks that are almost complete, or almost empty, or artificially regular. These kinds of models would not produce the observed network that you have. As Tom Kraft points out, a better specification would probably include some of the dyad-independent terms like "nodematch" to control homophily. But I suspect you might actually get better results by taking the approach we've been taking -- which uses offsets to control bipartite behavior rather than bipartite network definitions. I'll have more to say on this -- but am consulting with some of my colleagues first :) best, Martina agree with Tom Kraft's point On Thu, Nov 30, 2023 at 10:00?AM Harald Waxenecker wrote: > Dear ?statnet community?, > > > > Our research focuses on tie formation and elite cohesion, specifically > examining interlocking directorates and kinship relations. The dependent > bipartite business network comprises 6,902 individuals and 5,178 companies, > exhibiting sparsity (density = 0.00012) and fragmentation with 4,455 > components, including 3,850 isolates in the first mode (persons). The > attached documents contain descriptives and the component size distribution > from the observed network. > > > > The fragmented structure is important, as other network layers, like > kinship relations, are expected to contribute to the cohesion of this > business network. We apply ERGM to model these processes, but we struggle > to capture the fragmented structure of the observed network. The component > size distribution of the simulated network differs significantly. In > addition, the goodness-of-fit (GOF) for k-stars (in both modes) and > geodesic distances (Inf) shows significant results. All these results are > also attached. > > > > We've explored various options, including constraints, MCMC propositions, > and simulated annealing, but haven't achieved success. Please, we would > like to ask for your help to improve our model. Thank you! > > > > Kind regards, > > Harald > > > > > > > > --- > > > > *Harald Waxenecker * > > *Masaryk University | Faculty of social studies* > Department of Environment Studies > A: Jostova 10 | 602 00 Brno | Czech Republic > E: waxenecker@fss.muni.cz > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Thu Nov 30 16:32:47 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: Hi, Harald - Coexistence of large complex components does not generally occur unless something drives the fragmentation, and this is what your models are telling you: the terms you are currently using do not include the forces that are sufficient to reproduce your component size distribution.? That means that you need to think about why your network is split into fragments, and include terms that capture the relevant social forces.? Thinking about likely mechanisms is step zero, so do that before anything else!? Guided by your substantive knowledge of what is likely going on, you will next (as others have said) want to look at covariate effects relating to differential mixing, since those are your most obvious and most important sources of heterogeneity.? If you find that there is still more fragmentation that can be explained by other means, you may need to consider model terms relating directly to component count or size.? These are still somewhat experimental, and are currently sequestered in an add-on package called ergm.components (https://github.com/statnet/ergm.components). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded.? Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. Hope that helps, -Carter On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > Dear ?statnet community?, > > Our research focuses on tie formation and elite cohesion, specifically > examining interlocking directorates and kinship relations. The > dependent bipartite business network comprises 6,902 individuals and > 5,178 companies, exhibiting sparsity (density = 0.00012) and > fragmentation with 4,455 components, including 3,850 isolates in the > first mode (persons). The attached documents contain descriptives and > the component size distribution from the observed network. > > The fragmented structure is important, as other network layers, like > kinship relations, are expected to contribute to the cohesion of this > business network. We apply ERGM to model these processes, but we > struggle to capture the fragmented structure of the observed network. > The component size distribution of the?simulated network?differs > significantly. In addition, the goodness-of-fit (GOF) for k-stars (in > both modes) and geodesic distances (Inf) shows significant results. > All these results are also attached. > > We've explored various options, including constraints, MCMC > propositions, and simulated annealing, but haven't achieved success. > Please, we would like to ask for your help to improve our model. Thank > you! > > Kind regards, > > Harald > > --- > > *Harald Waxenecker > > * > > *Masaryk University | Faculty of social studies* > Department of Environment Studies > A:?Jostova 10 | 602 00 Brno | Czech Republic > E: waxenecker@fss.muni.cz > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From morrism at uw.edu Thu Nov 30 17:13:36 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: Hi Harald, I'm looking for some clarification here, which I think Tom Kraft might also have wondered about. You say: Our research focuses on tie formation and elite cohesion, specifically examining interlocking directorates and kinship relations. The dependent bipartite business network comprises 6,902 individuals and 5,178 companies, exhibiting sparsity (density = 0.00012) and fragmentation with 4,455 components, including 3,850 isolates in the first mode (persons) For a bipartite network ties are allowed only between modes (persons, companies), not within. It's clear how interlocking directorates would meet that criteria. But kinship relations would be among persons, so within-mode, not between, and this would not be a bipartite network. Is the model you've sent us for the interlocking directorships only? And by isolates in the person mode, do you mean persons who are not affiliated with any of the companies? If so, then it's a bit odd to include them in the bipartite network. I'm wondering if this problem is better posed as a multilevel network (not my area of expertise). thanks, Martina On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts wrote: > Hi, Harald - > > Coexistence of large complex components does not generally occur unless > something drives the fragmentation, and this is what your models are > telling you: the terms you are currently using do not include the forces > that are sufficient to reproduce your component size distribution. That > means that you need to think about why your network is split into > fragments, and include terms that capture the relevant social forces. > Thinking about likely mechanisms is step zero, so do that before anything > else! Guided by your substantive knowledge of what is likely going on, you > will next (as others have said) want to look at covariate effects relating > to differential mixing, since those are your most obvious and most > important sources of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. These > are still somewhat experimental, and are currently sequestered in an add-on > package called ergm.components (https://github.com/statnet/ergm.components > ). > However, this package can be installed from github (see the github page), > and the terms will work automagically with ergm() and friends once the > package is loaded. Depending on your situation, you may need or want to > examine the components() or compsizesum() terms, both of which are > documented within the package. > > Hope that helps, > > -Carter > On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > Dear ?statnet community?, > > > > Our research focuses on tie formation and elite cohesion, specifically > examining interlocking directorates and kinship relations. The dependent > bipartite business network comprises 6,902 individuals and 5,178 companies, > exhibiting sparsity (density = 0.00012) and fragmentation with 4,455 > components, including 3,850 isolates in the first mode (persons). The > attached documents contain descriptives and the component size distribution > from the observed network. > > > > The fragmented structure is important, as other network layers, like > kinship relations, are expected to contribute to the cohesion of this > business network. We apply ERGM to model these processes, but we struggle > to capture the fragmented structure of the observed network. The component > size distribution of the simulated network differs significantly. In > addition, the goodness-of-fit (GOF) for k-stars (in both modes) and > geodesic distances (Inf) shows significant results. All these results are > also attached. > > > > We've explored various options, including constraints, MCMC propositions, > and simulated annealing, but haven't achieved success. Please, we would > like to ask for your help to improve our model. Thank you! > > > > Kind regards, > > Harald > > > > > > > > --- > > > > *Harald Waxenecker * > > *Masaryk University | Faculty of social studies* > Department of Environment Studies > A: Jostova 10 | 602 00 Brno | Czech Republic > E: waxenecker@fss.muni.cz > > > > _______________________________________________ > statnet_help mailing liststatnet_help@u.washington.eduhttps://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From daniel.gotthardt at uni-hamburg.de Fri Dec 1 12:53:38 2023 From: daniel.gotthardt at uni-hamburg.de (Daniel Gotthardt) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: <9e721580-41e6-47c4-ab69-6ba4ec38b4e3@uni-hamburg.de> Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within.? It's clear how interlocking directorates would > meet that criteria.? But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies?? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution.? That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces.? Thinking about likely mechanisms is step > zero, so do that before anything else!? Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity.? If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > ). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded.? Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the?simulated >> network?differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A:?Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz ____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de From waxenecker at fss.muni.cz Mon Dec 4 00:28:12 2023 From: waxenecker at fss.muni.cz (Harald Waxenecker) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... Message-ID: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de -------------- next part -------------- An HTML attachment was scrubbed... URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: component.size.dist.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: model0.gof.pdf Type: application/pdf Size: 55226 bytes Desc: model0.gof.pdf URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: model0.gof.print.txt URL: -------------- next part -------------- A non-text attachment was scrubbed... Name: model0.mcmc.pdf Type: application/pdf Size: 3516323 bytes Desc: model0.mcmc.pdf URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: model0.summary.txt URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: observed_network.summary.txt URL: -------------- next part -------------- An embedded and charset-unspecified text was scrubbed... Name: simulated_network.summary.txt URL: From morrism at uw.edu Thu Dec 7 14:45:59 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker wrote: > Dear Tom, Martina, Carter and Daniel > > Thank you for your supportive answers. > > > > First, I will try to address some of your questions. The dependent network > is a bipartite business network (6902 persons x 5178 companies), based > exclusively on interlocking directorates. This dependent bipartite network > represents the business ties of elite members in their home country. We > include two covariates for the first node set (persons): *traditional > surname* and *gender*. Isolates in this network represent elite members > without any business ties. We belief that isolated nodes are meaningful in > this network; e.g., women are often constrained to ?reproduction? rather > than participating in ?production? (businesses). However, in different > network layers they contribute to elite cohesion. > > > > Regarding these different layers: we have six more networks. The first is > a one-mode kinship network (6902x6902), and the others are bipartite > networks (based on interlocks), where persons form the first node set and > entities the second. Hence, all matrices share a consistent number of rows > (n = 6902), while the number of columns varies according to the number of > entities in each network layer: offshore companies in Panama (n = 1537), > business associations (n = 128), non-profit organizations (n = 236), > political parties (n = 55), and public entities (n = 431). > > > > We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) > https://doi.org/10.1017/S0143814X18000181 > , > to test whether a common property (?homophily?) of the nodes in the first > node set, such as a shared attribute (gender, traditional surname), a > direct tie (kinship relation), or a mutual membership in other bipartite > layers (offshore companies, business associations, etc.) contribute to the > probability of two individuals forming ties with the same company in the > dependent network. > > > > Regarding the modeling process, it?s true that the model we shared relies > only on dyad-dependent terms. We always ?come back? to this model > specification because all our attempts, which certainly were also based > primarily on dyad-dependent terms, did not produce better results. We > explored various options, including nodematch to control for component > membership to split the network into smaller fragments. Then we > incorporated component membership of the nodes as constraint to induce > network fragmentation. While this partially improved network fragmentation, > problems with goodness-of-fit persisted. Additionally, we encountered some > computational limitations while running these options. > > > > Now, we have incorporated several of your recommendations, introducing > dyad-independent terms and utilizing components() from the ergm.components > package. Please find the new outcomes (model 0) attached. We've also > attached summary files and component distribution for a comparative > analysis between the observed network and the simulated network. > > > > We also tried to include the terms compsizesum() and dimers() into the > model; however, we observe degeneracy issues. In addition, we still could > not get results with bridges(), because it seems to be very time consuming > and/or needs much computational capacity. > > > > I think this bridges-term relates somehow to your question @Martina about > cross-group ties in the simulated data. Or maybe I am wrong. Please, could > you explain that in more detail? Thanks. > > > > Thank you again for your support. Looking very forward to read your > thoughts and advice. > > > > Kind regards, > > Harald > > > > > > > > > > > > > > > > > > El 1/12/23, 21:53, "[NOMBRE]" escribi?: > > Hello Harald, > > > > if I understand you correctly you have a within-mode network as well as > > a bipartite network. James Hollway et al. (2017) has described an > > approach to handle these kinds of combined networks as multilevel social > > spaces with stochastic actor-oriented models: > > > https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 > > > - There are also some tricks to transform these types of networks into > > an extended multimodal network matrix, exemplified e.g. in Knoke et al. > > (2021): > > > https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 > > > > > I personally don't know of any ergm model that can handle this kind of > > co-evolution of one-mode and two-mode networks but some kind of > > multilevel ergms (see Wang et al. (2013) > > https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051 > ) > > > might be the way to go: - I'm sure others here know more about the > > capabilities of ergm.multi though. > > > > If these kinship structures explain the fragmentation of the bipartite > > network, you might need to include them either directly with the > > approaches above or construct some corresponding dyadic or monadic > > covariates to represent the kinship structure in your single level network. > > > > Best Regards, > > > > Daniel > > > > Am 01.12.2023 um 02:13 schrieb Martina Morris: > > > > > > Hi Harald, > > > > > > I'm looking for some clarification here, which I think Tom Kraft might > > > also have wondered about. > > > > > > You say: > > >> > > >> Our research focuses on tie formation and elite cohesion, specifically > > >> examining interlocking directorates and kinship relations. The > > >> dependent bipartite business network comprises 6,902 individuals and > > >> 5,178 companies, exhibiting sparsity (density = 0.00012) and > > >> fragmentation with 4,455 components, including 3,850 isolates in the > > >> first mode (persons) > > >> > > > For a bipartite network ties are allowed only between modes (persons, > > > companies), not within. It's clear how interlocking directorates would > > > meet that criteria. But kinship relations would be among persons, so > > > within-mode, not between, and this would not be a bipartite network. > > > > > > Is the model you've sent us for the interlocking directorships only? > > > And by isolates in the person mode, do you mean persons who are not > > > affiliated with any of the companies? If so, then it's a bit odd to > > > include them in the bipartite network. > > > > > > I'm wondering if this problem is better posed as a multilevel network > > > (not my area of expertise). > > > > > > thanks, > > > Martina > > > > > > > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > > > wrote: > > > > > > __ > > > > > > Hi, Harald - > > > > > > Coexistence of large complex components does not generally occur > > > unless something drives the fragmentation, and this is what your > > > models are telling you: the terms you are currently using do not > > > include the forces that are sufficient to reproduce your component > > > size distribution. That means that you need to think about why your > > > network is split into fragments, and include terms that capture the > > > relevant social forces. Thinking about likely mechanisms is step > > > zero, so do that before anything else! Guided by your substantive > > > knowledge of what is likely going on, you will next (as others have > > > said) want to look at covariate effects relating to differential > > > mixing, since those are your most obvious and most important sources > > > of heterogeneity. If you find that there is still more > > > fragmentation that can be explained by other means, you may need to > > > consider model terms relating directly to component count or size. > > > These are still somewhat experimental, and are currently sequestered > > > in an add-on package called ergm.components > > > (https://github.com/statnet/ergm.components > > > > < > https://urldefense.com/v3/__https://github.com/statnet/ergm.components__;!!K-Hz7m0Vt54!iKts-XLv39sY0gvmpW6MWLIxNMCNKjKQKOhJszIbp3PIy_J5mdLCs0MytfHsBu-cjnQjk997tCRX0MMs6LDW$ > >). > However, this package can be installed from github (see the github page), > and the terms will work automagically with ergm() and friends once the > package is loaded. Depending on your situation, you may need or want to > examine the components() or compsizesum() terms, both of which are > documented within the package. > > > > > > Hope that helps, > > > > > > -Carter > > > > > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > >> > > >> Dear ?statnet community?,____ > > >> > > >> __ __ > > >> > > >> Our research focuses on tie formation and elite cohesion, > > >> specifically examining interlocking directorates and kinship > > >> relations. The dependent bipartite business network comprises > > >> 6,902 individuals and 5,178 companies, exhibiting sparsity > > >> (density = 0.00012) and fragmentation with 4,455 components, > > >> including 3,850 isolates in the first mode (persons). The attached > > >> documents contain descriptives and the component size distribution > > >> from the observed network.____ > > >> > > >> ____ > > >> > > >> The fragmented structure is important, as other network layers, > > >> like kinship relations, are expected to contribute to the cohesion > > >> of this business network. We apply ERGM to model these processes, > > >> but we struggle to capture the fragmented structure of the > > >> observed network. The component size distribution of the simulated > > >> network differs significantly. In addition, the goodness-of-fit > > >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) > > >> shows significant results. All these results are also attached.____ > > >> > > >> ____ > > >> > > >> We've explored various options, including constraints, MCMC > > >> propositions, and simulated annealing, but haven't achieved > > >> success. Please, we would like to ask for your help to improve our > > >> model. Thank you!____ > > >> > > >> __ __ > > >> > > >> Kind regards,____ > > >> > > >> Harald____ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> --- ____ > > >> > > >> *Harald Waxenecker > > >> > > >> *____ > > >> > > >> *Masaryk University | Faculty of social studies* > > >> Department of Environment Studies > > >> A: Jostova 10 | 602 00 Brno | Czech Republic > > >> E: waxenecker@fss.muni.cz ____ > > >> > > >> __ __ > > >> > > >> > > >> _______________________________________________ > > >> statnet_help mailing list > > >> statnet_help@u.washington.edu statnet_help@u.washington.edu> > > >> > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > < > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > > > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > -- > > > > Daniel Gotthardt, M.A. > > > > Wissenschaftlicher Mitarbeiter / Research Associate > > > > Universit?t Hamburg > > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > > Business, Economics and Social Sciences > > Fachbereich Sozialwissenschaften / Department of Social Sciences > > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > > Social Science > > > > Max-Brauer-Allee 60 > > 22765 Hamburg > > www.uni-hamburg.de > > > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From daniel.gotthardt at uni-hamburg.de Thu Dec 7 20:52:37 2023 From: daniel.gotthardt at uni-hamburg.de (Gotthardt, Daniel) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: , Message-ID: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Thu Dec 7 22:06:46 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: Message-ID: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets.? Nontrivial /global/ automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now.? (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994).? Beautiful idea with some lovely technical results, but one with few if any real-world success stories.? Sometimes, things just don't work out.)? I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford.? H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated).? H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence.? These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: > Dear Harald, > > after Martinas very insightful message and considering that you have > kinship and business ties but not so many node covariates, I am > wondering if you need or should think of structural equivalance as a > driving factor. With White and others there is a strong tradition of > focussing on this for kinship networks and DiMaggio and Burt have > studied the importance oft business roles and structural position. In > your case that probably means non-local forms of equivalence > (automorphic, role, etc) that might matter directly in the network > behavior or could represent unmeasured node attributes. Feature and > embedding based measures are more scalable and now allow to measure > those concepts better in larger networks. > > To the best of my knowledge this is not considered offen in generative > network models and i don't think that we can include those > less-localized mechanisms directly (yet). Plesae let me know if this > is a direction that makes sense for you from a theoretical point of > view and also something that could be identified in your data. I am > currently working on this in the context oft actor-oriented models but > am interested in the potential of ergms in this regard as well.? At > least as exogenous covariates this might be possible but otherwise we > might violate conditional independence (Hammersley-Clifford theorem). > I am curious to hear about the thoughts of experienced ergm modelers > on this, though. > > Best Regards, > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > Business, Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. > Digital Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > > ------------------------------------------------------------------------ > *Von:* Martina Morris > *Gesendet:* Donnerstag, 7. Dezember 2023 23:45:59 > *An:* Harald Waxenecker > *Cc:* Gotthardt, Daniel; statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > Hi Harald, > > You do have a complicated analysis here, and I'm a bit under-equipped > to help you Dx what is going on, as I don't have much experience with > either bipartite or multi-level nets (let alone both together!). > > What I can say, though, is that factor and covariate effects on the > nodes are, in the non-multilevel context, one of the most important > brakes on the feedback effects caused by dyad-dependent terms,?making > them more well-behaved and more likely to produce the kinds of > networks we actually observe (caveat: sometimes those dependent > effects are needed, see Carter's work on amyloid fibrils). > > In this case, it seems like you don't have many attributes to work > with -- indeed, only on one of the modes.? For gender, I would fit as > a factor btw, not a quantitative covariate, tho if there are only 2 > levels this will not have much impact.? But when I think about the > goals of board composition in non-profits (the closest I get to your > world), it's clear that gender is not the only attribute?that > influences board member invitations -- and I would expect the same > would be true here.? You might try adding? family name as a > bxnodefactor (will pick up both family size and family activity level > differentials), or sociality for either (or both) modes (to condition > on the degree of each node).? Your additional terms can then be > interpreted as effects operating beyond these differences in degree.? > Degree distributions definitely influence component size > distributions, up to a point, so if your model is not getting these > right, you can start there. > > Thinking about the orgs, it seems there must be org attributes that > influence the size and composition of the board.? Org size, sector, > geographic location, age, specialization, etc. -- I can imagine all of > these would influence board memberships.? Properties these nodes show > in the other nets you have might be able to be represented on the > cheap here as nodal attributes in this network. If these effects are > at work -- and if you're not including them in the model, it is a form > of mis-specification that compromises all of the other model estimates. > > Then there's homophily, which works differently in bip nets -- for > one, it's a dyad-dependent term.? But it's also more complicated to > think about.? Perhaps families might choose to specialize in an org > sector, or maybe the opposite, they aim to integrate across sectors.? > Orgs might want diversity (on some measure) for members, which would > show up as anti-homophily in bip two-paths.? Again though, this would > require more measured attributes for both orgs and persons. > > Adding model terms like components is different.? In my modeling > world, we want our (parsimonious) models to represent the mechanistic > effects that may actually generate the ties in the network.? For us, > component size distributions are an *output* of a network formation > process, not the generating mechanism (people aren't creating ties > with the explicit intent of structuring the network component size > distributions, with one key exception, and that we do model).? We > instead use the component size distribution as a goodness-of-fit > indicator -- to test whether the mechanistic terms we included in our > model?reproduce these higher order excluded network stats. > > But your context may be different.? When an org board is formed, if > there is an explicit strategy to create specific component structures > in the overall network then those intentions should be included as > model terms.? I can imagine that bridging structural holes might be > one of those strategies.? But again, not my area of expertise. > > I'm not sure how much any of this helps your specific issues.? But > when models don't fit the data properly, it's worth thinking about > specification from first principles. So I hope this helps. > > best, > Martina > > On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: > > Dear Tom, Martina, Carter and Daniel > > Thank you for your supportive answers. __ > > First, I will try to address some of your questions. The dependent > network is a bipartite business network (6902 persons x 5178 > companies), based exclusively on interlocking directorates. This > dependent bipartite network represents the business ties of elite > members in their home country. We include two covariates for the > first node set (persons): /traditional surname/ and /gender/. > Isolates in this network represent elite members without any > business ties. We belief that isolated nodes are meaningful in > this network; e.g., women are often constrained to ?reproduction? > rather than participating in ?production? (businesses). However, > in different network layers they contribute to elite cohesion. > > Regarding these different layers: we have six more networks. The > first is a one-mode kinship network (6902x6902), and the others > are bipartite networks (based on interlocks), where persons form > the first node set and entities the second. Hence, all matrices > share a consistent number of rows (n = 6902), while the number of > columns varies according to the number of entities in each network > layer: offshore companies in Panama (n = 1537), business > associations (n = 128), non-profit organizations (n = 236), > political parties (n = 55), and public entities (n = 431). > > We employ ?bipartite homophily terms?, as proposed by Metz et al. > (2018) https://doi.org/10.1017/S0143814X18000181 > , > to test whether a common property (?homophily?) of the nodes in > the first node set, such as a shared attribute (gender, > traditional surname), a direct tie (kinship relation), or a mutual > membership in other bipartite layers (offshore companies, business > associations, etc.) contribute to the probability of two > individuals forming ties with the same company in the dependent > network. > > Regarding the modeling process, it?s true that the model we shared > relies only on dyad-dependent terms.We always ?come back? to this > model specification because all our attempts, which certainly were > also based primarily on dyad-dependent terms, did not produce > better results. We explored various options, including nodematch > to control for component membership to split the network into > smaller fragments. Then we incorporated component membership of > the nodes as constraint to induce network fragmentation. While > this partially improved network fragmentation, problems with > goodness-of-fit persisted. Additionally, we encountered some > computational limitations while running these options. > > Now, we have incorporated several of your recommendations, > introducing dyad-independent terms and utilizing components() from > the ergm.components package. Please find the new outcomes (model > 0) attached. We've also attached summary files and component > distribution for a comparative analysis between the observed > network and the simulated network. > > We also tried to include the terms compsizesum() and dimers() into > the model; however, we observe degeneracy issues. In addition, we > still could not get results with bridges(), because it seems to be > very time consuming and/or needs much computational capacity. > > I think this bridges-term relates somehow to your question > @Martina about cross-group ties in the simulated data. Or maybe I > am wrong. Please, could you explain that in more detail? Thanks. > > Thank you again for your support. Looking very forward to read > your thoughts and advice. > > Kind regards, > > Harald > > El 1/12/23, 21:53, "[NOMBRE]" > escribi?: > > Hello Harald, > > if I understand you correctly you have a within-mode network as > well as > > a bipartite network. James Hollway et al. (2017) has described an > > approach to handle these kinds of combined networks as multilevel > social > > spaces with stochastic actor-oriented models: > > https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 > > > > - There are also some tricks to transform these types of networks > into > > an extended multimodal network matrix, exemplified e.g. in Knoke > et al. > > (2021): > > https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 > > > I personally don't know of any ergm model that can handle this > kind of > > co-evolution of one-mode and two-mode networks but some kind of > > multilevel ergms (see Wang et al. (2013) > > https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051 > ) > > > might be the way to go: - I'm sure others here know more about the > > capabilities of ergm.multi though. > > If these kinship structures explain the fragmentation of the > bipartite > > network, you might need to include them either directly with the > > approaches above or construct some corresponding dyadic or monadic > > covariates to represent the kinship structure in your single level > network. > > Best Regards, > > Daniel > > Am 01.12.2023 um 02:13 schrieb Martina Morris: > > > > > > Hi Harald, > > > > > > I'm looking for some clarification here, which I think Tom Kraft > might > > > also have wondered about. > > > > > > You say: > > >> > > >> Our research focuses on tie formation and elite cohesion, > specifically > > >> examining interlocking directorates and kinship relations. The > > >> dependent bipartite business network comprises 6,902 > individuals and > > >> 5,178 companies, exhibiting sparsity (density = 0.00012) and > > >> fragmentation with 4,455 components, including 3,850 isolates > in the > > >> first mode (persons) > > >> > > > For a bipartite network ties are allowed only between modes > (persons, > > > companies), not within. It's clear how interlocking directorates > would > > > meet that criteria.? But kinship relations would be among > persons, so > > > within-mode, not between, and this would not be a bipartite network. > > > > > > Is the model you've sent us for the interlocking directorships > only? > > > And by isolates in the person mode, do you mean persons who are not > > > affiliated with any of the companies?? If so, then it's a bit > odd to > > > include them in the bipartite network. > > > > > > I'm wondering if this problem is better posed as a multilevel > network > > > (not my area of expertise). > > > > > > thanks, > > > Martina > > > > > > > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > > > wrote: > > > > > >???? __ > > > > > >???? Hi, Harald - > > > > > >???? Coexistence of large complex components does not generally occur > > >???? unless something drives the fragmentation, and this is what your > > >???? models are telling you: the terms you are currently using do not > > >???? include the forces that are sufficient to reproduce your > component > > >???? size distribution. That means that you need to think about > why your > > >???? network is split into fragments, and include terms that > capture the > > >???? relevant social forces.? Thinking about likely mechanisms is > step > > >???? zero, so do that before anything else!? Guided by your > substantive > > >???? knowledge of what is likely going on, you will next (as > others have > > >???? said) want to look at covariate effects relating to differential > > >???? mixing, since those are your most obvious and most important > sources > > >???? of heterogeneity.? If you find that there is still more > > >???? fragmentation that can be explained by other means, you may > need to > > >???? consider model terms relating directly to component count or > size. > > >???? These are still somewhat experimental, and are currently > sequestered > > >???? in an add-on package called ergm.components > > >???? (https://github.com/statnet/ergm.components > > > >???? > >). > However, this package can be installed from github (see the github > page), and the terms will work automagically with ergm() and > friends once the package is loaded.? Depending on your situation, > you may need or want to examine the components() or compsizesum() > terms, both of which are documented within the package. > > > > > >???? Hope that helps, > > > > > >???? -Carter > > > > > >???? On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > >> > > >>???? Dear ?statnet community?,____ > > >> > > >>???? __ __ > > >> > > >>???? Our research focuses on tie formation and elite cohesion, > > >>???? specifically examining interlocking directorates and kinship > > >>???? relations. The dependent bipartite business network comprises > > >>???? 6,902 individuals and 5,178 companies, exhibiting sparsity > > >>???? (density = 0.00012) and fragmentation with 4,455 components, > > >>???? including 3,850 isolates in the first mode (persons). The > attached > > >>???? documents contain descriptives and the component size > distribution > > >>???? from the observed network.____ > > >> > > >>???? ____ > > >> > > >>???? The fragmented structure is important, as other network layers, > > >>???? like kinship relations, are expected to contribute to the > cohesion > > >>???? of this business network. We apply ERGM to model these > processes, > > >>???? but we struggle to capture the fragmented structure of the > > >>???? observed network. The component size distribution of > the?simulated > > >>???? network?differs significantly. In addition, the goodness-of-fit > > >>???? (GOF) for k-stars (in both modes) and geodesic distances (Inf) > > >>???? shows significant results. All these results are also > attached.____ > > >> > > >>???? ____ > > >> > > >>???? We've explored various options, including constraints, MCMC > > >>???? propositions, and simulated annealing, but haven't achieved > > >>???? success. Please, we would like to ask for your help to > improve our > > >>???? model. Thank you!____ > > >> > > >>???? __ __ > > >> > > >>???? Kind regards,____ > > >> > > >>???? Harald____ > > >> > > >>???? __ __ > > >> > > >>???? __ __ > > >> > > >>???? __ __ > > >> > > >>???? --- ____ > > >> > > >>???? *Harald Waxenecker > > >> > > >>???? *____ > > >> > > >>???? *Masaryk University | Faculty of social studies* > > >>???? Department of Environment Studies > > >>???? A:?Jostova 10 | 602 00 Brno | Czech Republic > > >>???? E: waxenecker@fss.muni.cz ____ > > >> > > >>???? __ __ > > >> > > >> > > >> _______________________________________________ > > >>???? statnet_help mailing list > > >> > statnet_help@u.washington.edu?? > > >> > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > ?? > > > > > _______________________________________________ > > >???? statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > >???? > > > > > > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > -- > > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > > Business, Economics and Social Sciences > > Fachbereich Sozialwissenschaften / Department of Social Sciences > > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. > Digital > > Social Science > > Max-Brauer-Allee 60 > > 22765 Hamburg > > www.uni-hamburg.de > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From daniel.gotthardt at uni-hamburg.de Thu Dec 7 23:02:04 2023 From: daniel.gotthardt at uni-hamburg.de (Gotthardt, Daniel) Date: Mon Mar 25 10:47:52 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> References: , <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> Message-ID: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From buttsc at uci.edu Fri Dec 8 01:53:06 2023 From: buttsc at uci.edu (Carter T. Butts) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> Message-ID: <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks.?? I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like.? Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases).? My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants.?? At any rate, models for differential mixing with discrete group structure are well-trod.? As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term).? The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern.? Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments.? (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize.? But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions."? It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so).? So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad.? If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals.? Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one.? (It will, in particular, have a form of Markov graph dependence.)? Nothing says that your model has to belong to /any/ particular dependence class - unless you want to impose such a condition.? Of course, if you /do /want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph).? In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong.? Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result.?? Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out.? (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.)?? For instance, k-cycles can be long-range, for large k.? The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints).? All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.).? If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family.? But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: > Hello Carter, > > i agree that stricter types oft equivalence are very rare and I would > personally also look at either generalized blockmodeling or actually > just measures of structural or positional similarity - but indeed not > only local ones (which are already included in ergm of course). I did > mention them here because most results of the relevance of more global > equivalence structures I know have been found in especially kinship > research and organisational science (Krackhardt & Porter 1986 and e.g. > in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). > There has also been some recent research in foreign trade and > political conflicts that indicate that block structures might matter > (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am > curious though which tools you are thinking about for implementing > aspects oft generalized block structures? > > Regarding hammersley-clifford I mostly wanted to be careful here, but > I did think that H-C and extensions like social circuit dependency > (which allows partial depensence) did matter to ensure some > (conditional) independence assumption with a few parameters (one for > each clique of the dependence graph) in ergms (see e.g. Koskinen & > Daraganova 2012 and Block er al. 2019). I thought dependencies (far) > beyond the local neigborhood might violate these properties. This is > probably beyond Harald's concerns but I would be happy if you could > indicate any literature to alleviate my misunderstanding. > > Best Regards > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > Business, Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. > Digital Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > > ------------------------------------------------------------------------ > *Von:* statnet_help > im Auftrag von Carter T. Butts > *Gesendet:* Freitag, 8. Dezember 2023 07:06:46 > *An:* statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > > Local automorphism orbits and their associations with covariates can > be modeled using graphlet statistics; see e.g. ergm.graphlets.? > Nontrivial /global/ automorphisms are extremely rare in typical social > networks, so such terms would be unlikely to be useful - what one > might call the "strong algebraic paradigm" of network analysis (the > idea that we could explain most social network structure in terms of > small numbers of roles, as defined through algebraic equivalences) was > a very compelling idea that didn't really work out, and I don't think > many folks are pushing in that direction right now.? (See also > compositional factorization, as famously illustrated by the semigroup > on the cover of Wasserman and Faust (1994). Beautiful idea with some > lovely technical results, but one with few if any real-world success > stories.? Sometimes, things just don't work out.)? I think there could > be some potential uses for terms for adherence to (confirmatory) > generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj > tradition), though some of this can already be emulated using existing > tools; there has also been a relative dearth of empirical cases in > which complex block types have been shown to be important for > capturing network structure.? If such cases were to become more often > encountered, this would naturally motivate more work to model them. > > With respect to your second comment, I am not sure what you mean by > "violating" Hammersley-Clifford.? H-C provides one way of establishing > an equivalence between sets of network statistics and associated > dependence conditions; Pip Pattison, Gary Robbins, and others have > obtained various refinements to the original result (allowing for more > subtle conditions to be treated).? H-C and friends simply say > (effectively) that certain classes of statistics implement certain > kinds of dependence.? These are important results for constructing and > interpreting statistics, but they are not rules that can be violated. > > Hope that clarifies things, > > -Carter > > On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: >> Dear Harald, >> >> after Martinas very insightful message and considering that you have >> kinship and business ties but not so many node covariates, I am >> wondering if you need or should think of structural equivalance as a >> driving factor. With White and others there is a strong tradition of >> focussing on this for kinship networks and DiMaggio and Burt have >> studied the importance oft business roles and structural position. In >> your case that probably means non-local forms of equivalence >> (automorphic, role, etc) that might matter directly in the network >> behavior or could represent unmeasured node attributes. Feature and >> embedding based measures are more scalable and now allow to measure >> those concepts better in larger networks. >> >> To the best of my knowledge this is not considered offen in >> generative network models and i don't think that we can include those >> less-localized mechanisms directly (yet). Plesae let me know if this >> is a direction that makes sense for you from a theoretical point of >> view and also something that could be identified in your data. I am >> currently working on this in the context oft actor-oriented models >> but am interested in the potential of ergms in this regard as well.? >> At least as exogenous covariates this might be possible but otherwise >> we might violate conditional independence (Hammersley-Clifford >> theorem). I am curious to hear about the thoughts of experienced ergm >> modelers on this, though. >> >> Best Regards, >> Daniel >> >> -- >> Daniel Gotthardt, M.A. >> >> Wissenschaftlicher Mitarbeiter / Research Associate >> >> Universit?t Hamburg >> Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of >> Business, Economics and Social Sciences >> Fachbereich Sozialwissenschaften / Department of Social Sciences >> Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. >> Digital Social Science >> >> Max-Brauer-Allee 60 >> 22765 Hamburg >> www.uni-hamburg.de >> >> >> ------------------------------------------------------------------------ >> *Von:* Martina Morris >> *Gesendet:* Donnerstag, 7. Dezember 2023 23:45:59 >> *An:* Harald Waxenecker >> *Cc:* Gotthardt, Daniel; statnet_help@u.washington.edu >> *Betreff:* Re: [statnet_help] fragmented bipartite network... >> Hi Harald, >> >> You do have a complicated analysis here, and I'm a bit under-equipped >> to help you Dx what is going on, as I don't have much experience with >> either bipartite or multi-level nets (let alone both together!). >> >> What I can say, though, is that factor and covariate effects on the >> nodes are, in the non-multilevel context, one of the most important >> brakes on the feedback effects caused by dyad-dependent terms,?making >> them more well-behaved and more likely to produce the kinds of >> networks we actually observe (caveat: sometimes those dependent >> effects are needed, see Carter's work on amyloid fibrils). >> >> In this case, it seems like you don't have many attributes to work >> with -- indeed, only on one of the modes.? For gender, I would fit as >> a factor btw, not a quantitative covariate, tho if there are only 2 >> levels this will not have much impact.? But when I think about the >> goals of board composition in non-profits (the closest I get to your >> world), it's clear that gender is not the only attribute?that >> influences board member invitations -- and I would expect the same >> would be true here.? You might try adding? family name as a >> bxnodefactor (will pick up both family size and family activity level >> differentials), or sociality for either (or both) modes (to condition >> on the degree of each node).? Your additional terms can then be >> interpreted as effects operating beyond these differences in degree. >> Degree distributions definitely influence component size >> distributions, up to a point, so if your model is not getting these >> right, you can start there. >> >> Thinking about the orgs, it seems there must be org attributes that >> influence the size and composition of the board.? Org size, sector, >> geographic location, age, specialization, etc. -- I can imagine all >> of these would influence board memberships.? Properties these nodes >> show in the other nets you have might be able to be represented on >> the cheap here as nodal attributes in this network. If these effects >> are at work -- and if you're not including them in the model, it is a >> form of mis-specification that compromises all of the other model >> estimates. >> >> Then there's homophily, which works differently in bip nets -- for >> one, it's a dyad-dependent term.? But it's also more complicated to >> think about.? Perhaps families might choose to specialize in an org >> sector, or maybe the opposite, they aim to integrate across sectors.? >> Orgs might want diversity (on some measure) for members, which would >> show up as anti-homophily in bip two-paths.? Again though, this would >> require more measured attributes for both orgs and persons. >> >> Adding model terms like components is different.? In my modeling >> world, we want our (parsimonious) models to represent the mechanistic >> effects that may actually generate the ties in the network.? For us, >> component size distributions are an *output* of a network formation >> process, not the generating mechanism (people aren't creating ties >> with the explicit intent of structuring the network component size >> distributions, with one key exception, and that we do model).? We >> instead use the component size distribution as a goodness-of-fit >> indicator -- to test whether the mechanistic terms we included in our >> model?reproduce these higher order excluded network stats. >> >> But your context may be different.? When an org board is formed, if >> there is an explicit strategy to create specific component structures >> in the overall network then those intentions should be included as >> model terms.? I can imagine that bridging structural holes might be >> one of those strategies.? But again, not my area of expertise. >> >> I'm not sure how much any of this helps your specific issues.? But >> when models don't fit the data properly, it's worth thinking about >> specification from first principles.? So I hope this helps. >> >> best, >> Martina >> >> On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker >> wrote: >> >> Dear Tom, Martina, Carter and Daniel >> >> Thank you for your supportive answers. __ >> >> First, I will try to address some of your questions. The >> dependent network is a bipartite business network (6902 persons x >> 5178 companies), based exclusively on interlocking directorates. >> This dependent bipartite network represents the business ties of >> elite members in their home country. We include two covariates >> for the first node set (persons): /traditional surname/ and >> /gender/. Isolates in this network represent elite members >> without any business ties. We belief that isolated nodes are >> meaningful in this network; e.g., women are often constrained to >> ?reproduction? rather than participating in ?production? >> (businesses). However, in different network layers they >> contribute to elite cohesion. >> >> Regarding these different layers: we have six more networks. The >> first is a one-mode kinship network (6902x6902), and the others >> are bipartite networks (based on interlocks), where persons form >> the first node set and entities the second. Hence, all matrices >> share a consistent number of rows (n = 6902), while the number of >> columns varies according to the number of entities in each >> network layer: offshore companies in Panama (n = 1537), business >> associations (n = 128), non-profit organizations (n = 236), >> political parties (n = 55), and public entities (n = 431). >> >> We employ ?bipartite homophily terms?, as proposed by Metz et al. >> (2018) https://doi.org/10.1017/S0143814X18000181 >> , >> to test whether a common property (?homophily?) of the nodes in >> the first node set, such as a shared attribute (gender, >> traditional surname), a direct tie (kinship relation), or a >> mutual membership in other bipartite layers (offshore companies, >> business associations, etc.) contribute to the probability of two >> individuals forming ties with the same company in the dependent >> network. >> >> Regarding the modeling process, it?s true that the model we >> shared relies only on dyad-dependent terms.We always ?come back? >> to this model specification because all our attempts, which >> certainly were also based primarily on dyad-dependent terms, did >> not produce better results. We explored various options, >> including nodematch to control for component membership to split >> the network into smaller fragments. Then we incorporated >> component membership of the nodes as constraint to induce network >> fragmentation. While this partially improved network >> fragmentation, problems with goodness-of-fit persisted. >> Additionally, we encountered some computational limitations while >> running these options. >> >> Now, we have incorporated several of your recommendations, >> introducing dyad-independent terms and utilizing components() >> from the ergm.components package. Please find the new outcomes >> (model 0) attached. We've also attached summary files and >> component distribution for a comparative analysis between the >> observed network and the simulated network. >> >> We also tried to include the terms compsizesum() and dimers() >> into the model; however, we observe degeneracy issues. In >> addition, we still could not get results with bridges(), because >> it seems to be very time consuming and/or needs much >> computational capacity. >> >> I think this bridges-term relates somehow to your question >> @Martina about cross-group ties in the simulated data. Or maybe I >> am wrong. Please, could you explain that in more detail? Thanks. >> >> Thank you again for your support. Looking very forward to read >> your thoughts and advice. >> >> Kind regards, >> >> Harald >> >> El 1/12/23, 21:53, "[NOMBRE]" >> escribi?: >> >> Hello Harald, >> >> if I understand you correctly you have a within-mode network as >> well as >> >> a bipartite network. James Hollway et al. (2017) has described an >> >> approach to handle these kinds of combined networks as multilevel >> social >> >> spaces with stochastic actor-oriented models: >> >> https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 >> >> >> >> - There are also some tricks to transform these types of networks >> into >> >> an extended multimodal network matrix, exemplified e.g. in Knoke >> et al. >> >> (2021): >> >> https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 >> >> >> I personally don't know of any ergm model that can handle this >> kind of >> >> co-evolution of one-mode and two-mode networks but some kind of >> >> multilevel ergms (see Wang et al. (2013) >> >> https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051 >> ) >> >> >> might be the way to go: - I'm sure others here know more about the >> >> capabilities of ergm.multi though. >> >> If these kinship structures explain the fragmentation of the >> bipartite >> >> network, you might need to include them either directly with the >> >> approaches above or construct some corresponding dyadic or monadic >> >> covariates to represent the kinship structure in your single >> level network. >> >> Best Regards, >> >> Daniel >> >> Am 01.12.2023 um 02:13 schrieb Martina Morris: >> >> > >> >> > Hi Harald, >> >> > >> >> > I'm looking for some clarification here, which I think Tom >> Kraft might >> >> > also have wondered about. >> >> > >> >> > You say: >> >> >> >> >> >> Our research focuses on tie formation and elite cohesion, >> specifically >> >> >> examining interlocking directorates and kinship relations. The >> >> >> dependent bipartite business network comprises 6,902 >> individuals and >> >> >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> >> >> fragmentation with 4,455 components, including 3,850 isolates >> in the >> >> >> first mode (persons) >> >> >> >> >> > For a bipartite network ties are allowed only between modes >> (persons, >> >> > companies), not within.? It's clear how interlocking >> directorates would >> >> > meet that criteria. But kinship relations would be among >> persons, so >> >> > within-mode, not between, and this would not be a bipartite >> network. >> >> > >> >> > Is the model you've sent us for the interlocking directorships >> only? >> >> > And by isolates in the person mode, do you mean persons who are >> not >> >> > affiliated with any of the companies?? If so, then it's a bit >> odd to >> >> > include them in the bipartite network. >> >> > >> >> > I'm wondering if this problem is better posed as a multilevel >> network >> >> > (not my area of expertise). >> >> > >> >> > thanks, >> >> > Martina >> >> > >> >> > >> >> > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> > > wrote: >> >> > >> >> >???? __ >> >> > >> >> >???? Hi, Harald - >> >> > >> >> >???? Coexistence of large complex components does not generally >> occur >> >> >???? unless something drives the fragmentation, and this is what >> your >> >> >???? models are telling you: the terms you are currently using >> do not >> >> >???? include the forces that are sufficient to reproduce your >> component >> >> >???? size distribution.? That means that you need to think about >> why your >> >> >???? network is split into fragments, and include terms that >> capture the >> >> >???? relevant social forces.? Thinking about likely mechanisms >> is step >> >> >???? zero, so do that before anything else!? Guided by your >> substantive >> >> >???? knowledge of what is likely going on, you will next (as >> others have >> >> >???? said) want to look at covariate effects relating to >> differential >> >> >???? mixing, since those are your most obvious and most >> important sources >> >> >???? of heterogeneity. If you find that there is still more >> >> >???? fragmentation that can be explained by other means, you may >> need to >> >> >???? consider model terms relating directly to component count >> or size. >> >> >???? These are still somewhat experimental, and are currently >> sequestered >> >> >???? in an add-on package called ergm.components >> >> >???? (https://github.com/statnet/ergm.components >> >> >> >???? >> > >). >> However, this package can be installed from github (see the >> github page), and the terms will work automagically with ergm() >> and friends once the package is loaded.? Depending on your >> situation, you may need or want to examine the components() or >> compsizesum() terms, both of which are documented within the package. >> >> > >> >> >???? Hope that helps, >> >> > >> >> >???? -Carter >> >> > >> >> >???? On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> >> >> >> >>???? Dear ?statnet community?,____ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >>???? Our research focuses on tie formation and elite cohesion, >> >> >>???? specifically examining interlocking directorates and kinship >> >> >>???? relations. The dependent bipartite business network comprises >> >> >>???? 6,902 individuals and 5,178 companies, exhibiting sparsity >> >> >>???? (density = 0.00012) and fragmentation with 4,455 components, >> >> >>???? including 3,850 isolates in the first mode (persons). The >> attached >> >> >>???? documents contain descriptives and the component size >> distribution >> >> >>???? from the observed network.____ >> >> >> >> >> >>???? ____ >> >> >> >> >> >>???? The fragmented structure is important, as other network >> layers, >> >> >>???? like kinship relations, are expected to contribute to the >> cohesion >> >> >>???? of this business network. We apply ERGM to model these >> processes, >> >> >>???? but we struggle to capture the fragmented structure of the >> >> >>???? observed network. The component size distribution of >> the?simulated >> >> >> network?differs significantly. In addition, the goodness-of-fit >> >> >>???? (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> >> >>???? shows significant results. All these results are also >> attached.____ >> >> >> >> >> >>???? ____ >> >> >> >> >> >>???? We've explored various options, including constraints, MCMC >> >> >>???? propositions, and simulated annealing, but haven't achieved >> >> >>???? success. Please, we would like to ask for your help to >> improve our >> >> >>???? model. Thank you!____ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >>???? Kind regards,____ >> >> >> >> >> >>???? Harald____ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >>???? --- ____ >> >> >> >> >> >>???? *Harald Waxenecker >> >> >> >> >> >>???? *____ >> >> >> >> >> >>???? *Masaryk University | Faculty of social studies* >> >> >>???? Department of Environment Studies >> >> >>???? A:?Jostova 10 | 602 00 Brno | Czech Republic >> >> >>???? E: waxenecker@fss.muni.cz ____ >> >> >> >> >> >>???? __ __ >> >> >> >> >> >> >> >> >> _______________________________________________ >> >> >>???? statnet_help mailing list >> >> >> >> statnet_help@u.washington.edu?? >> >> >> >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ >> ??> > >> >> >> > _______________________________________________ >> >> >???? statnet_help mailing list >> >> > statnet_help@u.washington.edu >> >> >> > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help >> >> >> >???? >> >> >> >> > >> >> > >> >> > _______________________________________________ >> >> > statnet_help mailing list >> >> > statnet_help@u.washington.edu >> >> > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help >> >> >> -- >> >> Daniel Gotthardt, M.A. >> >> Wissenschaftlicher Mitarbeiter / Research Associate >> >> Universit?t Hamburg >> >> Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of >> >> Business, Economics and Social Sciences >> >> Fachbereich Sozialwissenschaften / Department of Social Sciences >> >> Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. >> Digital >> >> Social Science >> >> Max-Brauer-Allee 60 >> >> 22765 Hamburg >> >> www.uni-hamburg.de >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> http://mailman13.u.washington.edu/mailman/listinfo/statnet_help >> >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From jmoody77 at duke.edu Fri Dec 8 06:28:07 2023 From: jmoody77 at duke.edu (James Moody) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Message-ID: Fun discussion, thanks for sharing, always learn something in these sorts of posts. As to this this application per se; a couple of pragmatic (i.e. may not be elegant!) ideas: - theory should be able to inform some unlikely mixing that one could specify using a mixingmatrix term or two, no? So family, private/public, industry, etc. - For many business group applications, the actual family name is embedded in many of the subsidiaries (Tata group, tata inc, tata industries, etc.) so a name-similarity score could help (if you have nodenames) - The interlock limit will be size of the boards. While its possible to change the size of each board in a company, its not trivial, and I think you can justifiably take that as exogenous in the time-frame you have. I?m betting most of your small components are single family companies without external board memberships. Those create small stars in the bipartiate network (cliques in the projection). So that would imply: a) a hard-constraint on target degree. You could just fix that as a constraint. Again, not elegant (Carter?s cutting at joints and all), but likely true. b) a size mixing logic. Family-only/small-board cliques are isolated, leaving big-with-big, so there?s effectively a two-mode degree assortativity here. If you can?t induce this by an attribute (family name/ownership), then use assortativity on degree. - Cheating a little, but you could make component membership at attribute and hard-code mixing within/between. That means you can?t model what drives membership in the largest components vs. the small fractions, but, again, this is such a weird case (from a graph expectation sense), as anything that had even a little random noise in it would link across those small components, so the restriction here is almost certainly a legal/possibility restriction that should be treated as exogenous. - that?s, of course, just the crudest version of Daniel?s idea ? find a structural pattern that implies high/low probability of mixing across modes and hard-code it. I.e. do some old-fashioned inductive modeling of your network before the ERGM to generate classes of cases based on your best effort to induce the (to you) invisible restrictions patterning the ties, then add those back into the model as appropriate node/edge attributes. PTs Jim From: statnet_help On Behalf Of Carter T. Butts Sent: Friday, December 8, 2023 4:53 AM To: statnet_help@u.washington.edu Subject: Re: [statnet_help] fragmented bipartite network... Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks. I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like. Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases). My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants. At any rate, models for differential mixing with discrete group structure are well-trod. As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term). The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern. Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments. (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize. But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions." It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so). So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad. If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals. Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one. (It will, in particular, have a form of Markov graph dependence.) Nothing says that your model has to belong to any particular dependence class - unless you want to impose such a condition. Of course, if you do want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph). In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong. Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result. Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out. (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.) For instance, k-cycles can be long-range, for large k. The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints). All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.). If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family. But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ -------------- next part -------------- An HTML attachment was scrubbed... URL: From morrism at uw.edu Fri Dec 8 12:45:54 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Message-ID: This is a great conversation; many thanks to the contributors. As I read through the proposed stats, though, I keep stumbling on the bipartite bit: how would some of these translate into bip net terms? I appreciate Jim's effort to bring this back to practical advice. So, some really basic thoughts here. There are two general types of blocks: those based on exogenous attributes, and those based on endogenous processes. I think the reason we're circling around the idea of blocks is that these depictions tend to capture the clustering observed in real world networks, and that blocking can help explain why dyad-dependent effects operate locally, rather than globally across a network. The exogenous type of block is captured by nodemix and nodematch type terms in ergm (which have a number of different specifications). In the bip net context these terms become more complicated as they no longer represent the crosstabulation of pairwise nodal attributes, but instead a crosstab of the terminal node attributes of a 2-mode triad. What's interesting about the bip net version of these terms is that this 2-path configuration is also a building block of equivalence. More on this below. The endogenous type of block is captured as latent block structures in hergms (for the ergm framework, other frameworks are out there). HERGMs are an interesting approach to identifying observed or latent neighborhoods of dependence (https://www.jstatsoft.org/article/view/v085i01), but I don't know if the package (or the models) can handle bipartite nets. I've added Michael Schweinberger to this email in case he would like to comment. Back to the exogenous blocking then. Family name could be a powerful blocking effect (e.g. Jim's example of Tata), showing up in this bip net as org board memberships shared by people with the same family name. Ignoring the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner (NSP) statistics. If two people shared all of their org memberships, they are structurally equivalent (whether they share an exogenous attribute or not) -- and more generally, the more NSPs, the higher the equivalence. And if the nodal name attribute is not driving these 2 paths, these high value NSPs are indicators of latent structure. The 2-paths can also be used to examine the org equivalence pattern in the same way. And my intuition would be that, conditioned on density, NSP distributions with higher means or longer tails would lead to fragmentation in the network. So, that makes me think perhaps the place to start is with EDA -- look at the NSP distributions, for both persons and orgs. Compare these to the expected distributions under a simple null random graph. If the distributions differ significantly, then start to look for exogenous effects that help to explain the deviation from the null (using the bip homophily terms with some more attributes on the nodes of both modes). And look into whether endogenously defined blocks (a la HERGM) can be used for bip nets. For me, the ideal would be to identify the latent blocks, and then explain almost all of that blocking in terms of exogenous/observed attributes. The blocks capture the structure. The explicit exogenous effects "explain" it. best, mm On Fri, Dec 8, 2023 at 6:28?AM James Moody wrote: > Fun discussion, thanks for sharing, always learn something in these sorts > of posts. > > > > As to this this application per se; a couple of pragmatic (i.e. may not be > elegant!) ideas: > > > > - theory should be able to inform some unlikely mixing that one could > specify using a mixingmatrix term or two, no? So family, private/public, > industry, etc. > > - For many business group applications, the actual family name is > embedded in many of the subsidiaries (Tata group, tata inc, tata > industries, etc.) so a name-similarity score could help (if you have > nodenames) > > - The interlock limit will be size of the boards. While its possible to > change the size of each board in a company, its not trivial, and I think > you can justifiably take that as exogenous in the time-frame you have. I?m > betting most of your small components are single family companies without > external board memberships. Those create small stars in the bipartiate > network (cliques in the projection). So that would imply: > > a) a hard-constraint on target degree. You could just fix that as a > constraint. Again, not elegant (Carter?s cutting at joints and all), but > likely true. > > b) a size mixing logic. Family-only/small-board cliques are isolated, > leaving big-with-big, so there?s effectively a two-mode degree > assortativity here. If you can?t induce this by an attribute (family > name/ownership), then use assortativity on degree. > > - Cheating a little, but you could make component membership at attribute > and hard-code mixing within/between. That means you can?t model what drives > membership in the largest components vs. the small fractions, but, again, > this is such a weird case (from a graph expectation sense), as anything > that had even a little random noise in it would link across those small > components, so the restriction here is almost certainly a legal/possibility > restriction that should be treated as exogenous. > > - that?s, of course, just the crudest version of Daniel?s idea ? find a > structural pattern that implies high/low probability of mixing across modes > and hard-code it. I.e. do some old-fashioned inductive modeling of your > network before the ERGM to generate classes of cases based on your best > effort to induce the (to you) invisible restrictions patterning the ties, > then add those back into the model as appropriate node/edge attributes. > > > > PTs > > Jim > > > > > > > > *From:* statnet_help *On > Behalf Of *Carter T. Butts > *Sent:* Friday, December 8, 2023 4:53 AM > *To:* statnet_help@u.washington.edu > *Subject:* Re: [statnet_help] fragmented bipartite network... > > > > Hi, Daniel - > > Most of the cases to which I believe you are referring deal with > differential mixing; the "blocks" here are what are sometimes called > "density" blocks, which are quantitative relaxations of the complete/null > blocks. I don't think anyone doubts that differential mixing exists, but > that is very far from e.g. nontrivial global automorphism orbits or the > like. Indeed, John Boyd had a running bet for some years, in which he > offered to pay a sum of money (I forget how much) to anyone who could show > a statistically significant regular equivalence pattern (above and beyond > SE - he also had some other boundary conditions that ruled out "easy" > cases). My vague recollection was that Steve Borgatti claimed to have one, > and they then haggled over John's way of calculating "significance," but my > memory on the subject is hazy and doubtless untrustworthy; I never did buy > John's extreme conjecture, but it is true that he was not exactly > overwhelmed with claimants. At any rate, models for differential mixing > with discrete group structure are well-trod. As far as other kinds of > generalized blocks (moving away from complete/null blocks), you can fit > models with strict versions of e.g. regular, row/column dominant, and > row/column functional blocks with clever use of constraints (in ergm, the > bd() constraint term). The most obvious path to soft versions of those > block types is to create statistics that count violations of the block > pattern. Some can be implemented using the degrange() term, together with > appropriate use of the optional attribute arguments. (Obviously, these are > all "confirmatory" models, in the sense that one has to specify the block > structure one wants to impose/parameterize. But that is not without its > virtues.) > > Vis a vis dependence, I'm not sure that it is very helpful to think in > terms of "violating assumptions." It is probably more useful to think of > H-C and friends as giving you a "recipe" for the statistics you need to > implement particular kinds of dependence conditions (should you want to do > so). So, e.g., if you want edges to depend on each other when they share > endpoints, then you will want (in the unvalued case) indicators for each > edge variable, and indicators for each mutual dyad. If you also want the > corresponding effects to be homogeneous, then this reduces to the edge > count and the count of mutuals. Adding e.g. a 2-outstar term to a model > with edges and mutuals is not violating any particular assumption imposed > by the latter - it's just that this new model will now belong to a > different (and broader) dependence class than the original one. (It will, > in particular, have a form of Markov graph dependence.) Nothing says that > your model has to belong to *any* particular dependence class - unless > you want to impose such a condition. Of course, if you *do *want to > restrict your dependence to a particular class, then you will indeed need > to ensure that your statistics are a subset of those admitted by that class > (which, for H-C, can be determined from the cliques of the conditional > dependence graph). In my experience, this is rarely a useful way to > proceed; however, it sometimes can be handy to know the type of dependence > class to which your terms belong. Likewise, it can sometimes be handy to > start by positing a form of dependence that makes sense in a specific > situation, and then deriving the statistics that result. Pip, in > particular, has done a great deal to elucidate these sorts of connections. > > As far as long-range dependence, there's again nothing ruling it out. > (Pip and Tom, IIRC, have a very nice typology working out statistics for > dependence classes at different distances.) For instance, k-cycles can be > long-range, for large k. The various component and bridging statistics can > be arbitrarily long-range. The statistics that arise from density and dyad > census mixtures do them one better by being completely global (i.e., they > create conditional dependence between edge variables irrespective of > whether there is even a path of any length between their endpoints). All > of these lead to well-defined models - those models just happen not to > belong e.g. to the Markov graphs (or the social circuit graphs, the > Bernoulli graphs, the u|man family, etc.). If there is a reason that you > need your model to belong to such a family, then you would not want to use > terms that are not within the class specifying that family. But otherwise, > such restrictions are arbitrary, and may get in the way of specifying > important mechanisms. > > Hope that helps, > > -Carter > > > > On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: > > Hello Carter, > > i agree that stricter types oft equivalence are very rare and I would > personally also look at either generalized blockmodeling or actually just > measures of structural or positional similarity - but indeed not only local > ones (which are already included in ergm of course). I did mention them > here because most results of the relevance of more global equivalence > structures I know have been found in especially kinship research and > organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional > fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some > recent research in foreign trade and political conflicts that indicate that > block structures might matter (Guler et al. 2002, Zhou & Park 2012, > Olivella et al. 2022). I am curious though which tools you are thinking > about for implementing aspects oft generalized block structures? > > Regarding hammersley-clifford I mostly wanted to be careful here, but I > did think that H-C and extensions like social circuit dependency (which > allows partial depensence) did matter to ensure some (conditional) > independence assumption with a few parameters (one for each clique of the > dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block > er al. 2019). I thought dependencies (far) beyond the local neigborhood > might violate these properties. This is probably beyond Harald's concerns > but I would be happy if you could indicate any literature to alleviate my > misunderstanding. > > Best Regards > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, > Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > ------------------------------ > > *Von:* statnet_help > im Auftrag von Carter > T. Butts > *Gesendet:* Freitag, 8. Dezember 2023 07:06:46 > *An:* statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > > > > Local automorphism orbits and their associations with covariates can be > modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial > *global* automorphisms are extremely rare in typical social networks, so > such terms would be unlikely to be useful - what one might call the "strong > algebraic paradigm" of network analysis (the idea that we could explain > most social network structure in terms of small numbers of roles, as > defined through algebraic equivalences) was a very compelling idea that > didn't really work out, and I don't think many folks are pushing in that > direction right now. (See also compositional factorization, as famously > illustrated by the semigroup on the cover of Wasserman and Faust (1994). > Beautiful idea with some lovely technical results, but one with few if any > real-world success stories. Sometimes, things just don't work out.) I > think there could be some potential uses for terms for adherence to > (confirmatory) generalized blockmodel structure (in the > Doreian/Ferligoj/Batagelj tradition), though some of this can already be > emulated using existing tools; there has also been a relative dearth of > empirical cases in which complex block types have been shown to be > important for capturing network structure. If such cases were to become > more often encountered, this would naturally motivate more work to model > them. > > With respect to your second comment, I am not sure what you mean by > "violating" Hammersley-Clifford. H-C provides one way of establishing an > equivalence between sets of network statistics and associated dependence > conditions; Pip Pattison, Gary Robbins, and others have obtained various > refinements to the original result (allowing for more subtle conditions to > be treated). H-C and friends simply say (effectively) that certain classes > of statistics implement certain kinds of dependence. These are important > results for constructing and interpreting statistics, but they are not > rules that can be violated. > > Hope that clarifies things, > > -Carter > > On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: > > Dear Harald, > > after Martinas very insightful message and considering that you have > kinship and business ties but not so many node covariates, I am wondering > if you need or should think of structural equivalance as a driving factor. > With White and others there is a strong tradition of focussing on this for > kinship networks and DiMaggio and Burt have studied the importance oft > business roles and structural position. In your case that probably means > non-local forms of equivalence (automorphic, role, etc) that might matter > directly in the network behavior or could represent unmeasured node > attributes. Feature and embedding based measures are more scalable and now > allow to measure those concepts better in larger networks. > > To the best of my knowledge this is not considered offen in generative > network models and i don't think that we can include those less-localized > mechanisms directly (yet). Plesae let me know if this is a direction that > makes sense for you from a theoretical point of view and also something > that could be identified in your data. I am currently working on this in > the context oft actor-oriented models but am interested in the potential of > ergms in this regard as well. At least as exogenous covariates this might > be possible but otherwise we might violate conditional independence > (Hammersley-Clifford theorem). I am curious to hear about the thoughts of > experienced ergm modelers on this, though. > > Best Regards, > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, > Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > ------------------------------ > > *Von:* Martina Morris > *Gesendet:* Donnerstag, 7. Dezember 2023 23:45:59 > *An:* Harald Waxenecker > *Cc:* Gotthardt, Daniel; statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > > > > Hi Harald, > > > > You do have a complicated analysis here, and I'm a bit under-equipped to > help you Dx what is going on, as I don't have much experience with either > bipartite or multi-level nets (let alone both together!). > > > > What I can say, though, is that factor and covariate effects on the nodes > are, in the non-multilevel context, one of the most important brakes on the > feedback effects caused by dyad-dependent terms, making them more > well-behaved and more likely to produce the kinds of networks we actually > observe (caveat: sometimes those dependent effects are needed, see Carter's > work on amyloid fibrils). > > > > In this case, it seems like you don't have many attributes to work with -- > indeed, only on one of the modes. For gender, I would fit as a factor btw, > not a quantitative covariate, tho if there are only 2 levels this will not > have much impact. But when I think about the goals of board composition in > non-profits (the closest I get to your world), it's clear that gender is > not the only attribute that influences board member invitations -- and I > would expect the same would be true here. You might try adding family > name as a bxnodefactor (will pick up both family size and family activity > level differentials), or sociality for either (or both) modes (to condition > on the degree of each node). Your additional terms can then be interpreted > as effects operating beyond these differences in degree. Degree > distributions definitely influence component size distributions, up to a > point, so if your model is not getting these right, you can start there. > > > > Thinking about the orgs, it seems there must be org attributes that > influence the size and composition of the board. Org size, sector, > geographic location, age, specialization, etc. -- I can imagine all of > these would influence board memberships. Properties these nodes show in > the other nets you have might be able to be represented on the cheap here > as nodal attributes in this network. If these effects are at work -- and if > you're not including them in the model, it is a form of mis-specification > that compromises all of the other model estimates. > > > > Then there's homophily, which works differently in bip nets -- for one, > it's a dyad-dependent term. But it's also more complicated to think > about. Perhaps families might choose to specialize in an org sector, or > maybe the opposite, they aim to integrate across sectors. Orgs might want > diversity (on some measure) for members, which would show up as > anti-homophily in bip two-paths. Again though, this would require more > measured attributes for both orgs and persons. > > > > Adding model terms like components is different. In my modeling world, we > want our (parsimonious) models to represent the mechanistic effects that > may actually generate the ties in the network. For us, component size > distributions are an *output* of a network formation process, not the > generating mechanism (people aren't creating ties with the explicit intent > of structuring the network component size distributions, with one key > exception, and that we do model). We instead use the component size > distribution as a goodness-of-fit indicator -- to test whether the > mechanistic terms we included in our model reproduce these higher order > excluded network stats. > > > > But your context may be different. When an org board is formed, if there > is an explicit strategy to create specific component structures in the > overall network then those intentions should be included as model terms. I > can imagine that bridging structural holes might be one of those > strategies. But again, not my area of expertise. > > > > I'm not sure how much any of this helps your specific issues. But when > models don't fit the data properly, it's worth thinking about specification > from first principles. So I hope this helps. > > > > best, > > Martina > > > > On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: > > Dear Tom, Martina, Carter and Daniel > > Thank you for your supportive answers. > > > > First, I will try to address some of your questions. The dependent network > is a bipartite business network (6902 persons x 5178 companies), based > exclusively on interlocking directorates. This dependent bipartite network > represents the business ties of elite members in their home country. We > include two covariates for the first node set (persons): *traditional > surname* and *gender*. Isolates in this network represent elite members > without any business ties. We belief that isolated nodes are meaningful in > this network; e.g., women are often constrained to ?reproduction? rather > than participating in ?production? (businesses). However, in different > network layers they contribute to elite cohesion. > > > > Regarding these different layers: we have six more networks. The first is > a one-mode kinship network (6902x6902), and the others are bipartite > networks (based on interlocks), where persons form the first node set and > entities the second. Hence, all matrices share a consistent number of rows > (n = 6902), while the number of columns varies according to the number of > entities in each network layer: offshore companies in Panama (n = 1537), > business associations (n = 128), non-profit organizations (n = 236), > political parties (n = 55), and public entities (n = 431). > > > > We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) > https://doi.org/10.1017/S0143814X18000181 > , > to test whether a common property (?homophily?) of the nodes in the first > node set, such as a shared attribute (gender, traditional surname), a > direct tie (kinship relation), or a mutual membership in other bipartite > layers (offshore companies, business associations, etc.) contribute to the > probability of two individuals forming ties with the same company in the > dependent network. > > > > Regarding the modeling process, it?s true that the model we shared relies > only on dyad-dependent terms. We always ?come back? to this model > specification because all our attempts, which certainly were also based > primarily on dyad-dependent terms, did not produce better results. We > explored various options, including nodematch to control for component > membership to split the network into smaller fragments. Then we > incorporated component membership of the nodes as constraint to induce > network fragmentation. While this partially improved network fragmentation, > problems with goodness-of-fit persisted. Additionally, we encountered some > computational limitations while running these options. > > > > Now, we have incorporated several of your recommendations, introducing > dyad-independent terms and utilizing components() from the ergm.components > package. Please find the new outcomes (model 0) attached. We've also > attached summary files and component distribution for a comparative > analysis between the observed network and the simulated network. > > > > We also tried to include the terms compsizesum() and dimers() into the > model; however, we observe degeneracy issues. In addition, we still could > not get results with bridges(), because it seems to be very time consuming > and/or needs much computational capacity. > > > > I think this bridges-term relates somehow to your question @Martina about > cross-group ties in the simulated data. Or maybe I am wrong. Please, could > you explain that in more detail? Thanks. > > > > Thank you again for your support. Looking very forward to read your > thoughts and advice. > > > > Kind regards, > > Harald > > > > > > > > > > > > > > > > > > El 1/12/23, 21:53, "[NOMBRE]" escribi?: > > Hello Harald, > > > > if I understand you correctly you have a within-mode network as well as > > a bipartite network. James Hollway et al. (2017) has described an > > approach to handle these kinds of combined networks as multilevel social > > spaces with stochastic actor-oriented models: > > > https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 > > > - There are also some tricks to transform these types of networks into > > an extended multimodal network matrix, exemplified e.g. in Knoke et al. > > (2021): > > > https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 > > > > > I personally don't know of any ergm model that can handle this kind of > > co-evolution of one-mode and two-mode networks but some kind of > > multilevel ergms (see Wang et al. (2013) > > https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051 > ) > > > might be the way to go: - I'm sure others here know more about the > > capabilities of ergm.multi though. > > > > If these kinship structures explain the fragmentation of the bipartite > > network, you might need to include them either directly with the > > approaches above or construct some corresponding dyadic or monadic > > covariates to represent the kinship structure in your single level network. > > > > Best Regards, > > > > Daniel > > > > Am 01.12.2023 um 02:13 schrieb Martina Morris: > > > > > > Hi Harald, > > > > > > I'm looking for some clarification here, which I think Tom Kraft might > > > also have wondered about. > > > > > > You say: > > >> > > >> Our research focuses on tie formation and elite cohesion, specifically > > >> examining interlocking directorates and kinship relations. The > > >> dependent bipartite business network comprises 6,902 individuals and > > >> 5,178 companies, exhibiting sparsity (density = 0.00012) and > > >> fragmentation with 4,455 components, including 3,850 isolates in the > > >> first mode (persons) > > >> > > > For a bipartite network ties are allowed only between modes (persons, > > > companies), not within. It's clear how interlocking directorates would > > > meet that criteria. But kinship relations would be among persons, so > > > within-mode, not between, and this would not be a bipartite network. > > > > > > Is the model you've sent us for the interlocking directorships only? > > > And by isolates in the person mode, do you mean persons who are not > > > affiliated with any of the companies? If so, then it's a bit odd to > > > include them in the bipartite network. > > > > > > I'm wondering if this problem is better posed as a multilevel network > > > (not my area of expertise). > > > > > > thanks, > > > Martina > > > > > > > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > > > wrote: > > > > > > __ > > > > > > Hi, Harald - > > > > > > Coexistence of large complex components does not generally occur > > > unless something drives the fragmentation, and this is what your > > > models are telling you: the terms you are currently using do not > > > include the forces that are sufficient to reproduce your component > > > size distribution. That means that you need to think about why your > > > network is split into fragments, and include terms that capture the > > > relevant social forces. Thinking about likely mechanisms is step > > > zero, so do that before anything else! Guided by your substantive > > > knowledge of what is likely going on, you will next (as others have > > > said) want to look at covariate effects relating to differential > > > mixing, since those are your most obvious and most important sources > > > of heterogeneity. If you find that there is still more > > > fragmentation that can be explained by other means, you may need to > > > consider model terms relating directly to component count or size. > > > These are still somewhat experimental, and are currently sequestered > > > in an add-on package called ergm.components > > > (https://github.com/statnet/ergm.components > > > > < > https://urldefense.com/v3/__https://github.com/statnet/ergm.components__;!!K-Hz7m0Vt54!iKts-XLv39sY0gvmpW6MWLIxNMCNKjKQKOhJszIbp3PIy_J5mdLCs0MytfHsBu-cjnQjk997tCRX0MMs6LDW$ > >). > However, this package can be installed from github (see the github page), > and the terms will work automagically with ergm() and friends once the > package is loaded. Depending on your situation, you may need or want to > examine the components() or compsizesum() terms, both of which are > documented within the package. > > > > > > Hope that helps, > > > > > > -Carter > > > > > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > >> > > >> Dear ?statnet community?,____ > > >> > > >> __ __ > > >> > > >> Our research focuses on tie formation and elite cohesion, > > >> specifically examining interlocking directorates and kinship > > >> relations. The dependent bipartite business network comprises > > >> 6,902 individuals and 5,178 companies, exhibiting sparsity > > >> (density = 0.00012) and fragmentation with 4,455 components, > > >> including 3,850 isolates in the first mode (persons). The attached > > >> documents contain descriptives and the component size distribution > > >> from the observed network.____ > > >> > > >> ____ > > >> > > >> The fragmented structure is important, as other network layers, > > >> like kinship relations, are expected to contribute to the cohesion > > >> of this business network. We apply ERGM to model these processes, > > >> but we struggle to capture the fragmented structure of the > > >> observed network. The component size distribution of the simulated > > >> network differs significantly. In addition, the goodness-of-fit > > >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) > > >> shows significant results. All these results are also attached.____ > > >> > > >> ____ > > >> > > >> We've explored various options, including constraints, MCMC > > >> propositions, and simulated annealing, but haven't achieved > > >> success. Please, we would like to ask for your help to improve our > > >> model. Thank you!____ > > >> > > >> __ __ > > >> > > >> Kind regards,____ > > >> > > >> Harald____ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> --- ____ > > >> > > >> *Harald Waxenecker > > >> > > >> *____ > > >> > > >> *Masaryk University | Faculty of social studies* > > >> Department of Environment Studies > > >> A: Jostova 10 | 602 00 Brno | Czech Republic > > >> E: waxenecker@fss.muni.cz ____ > > >> > > >> __ __ > > >> > > >> > > >> _______________________________________________ > > >> statnet_help mailing list > > >> statnet_help@u.washington.edu statnet_help@u.washington.edu> > > >> > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > < > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > > > > > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > -- > > > > Daniel Gotthardt, M.A. > > > > Wissenschaftlicher Mitarbeiter / Research Associate > > > > Universit?t Hamburg > > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > > Business, Economics and Social Sciences > > Fachbereich Sozialwissenschaften / Department of Social Sciences > > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > > Social Science > > > > Max-Brauer-Allee 60 > > 22765 Hamburg > > www.uni-hamburg.de > > > > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > _______________________________________________ > > statnet_help mailing list > > statnet_help@u.washington.edu > > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From dhunter at stat.psu.edu Fri Dec 8 18:44:17 2023 From: dhunter at stat.psu.edu (Hunter, David) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Message-ID: Following up on Martina?s observations among others? In case it helps, the b1nodematch and b2nodematch terms in the ergm package do not merely provide a census of 2-paths with matching end-nodes. They do provide this census, but merely as one end of a spectrum (two spectra, actually) of statistics created in the same spirit as the geometrically weighted statistics (GWESP, GWD, etc.) pioneered by Snijders et al back in 2006 (?New Specifications for Exponential Random Graph Models?). The full spectra entail a more flexible way to capture homophily in a bipartite network. We?ve just submitted a manuscript on this, and coincidentally we use a bipartite network of interlocking directorates to illustrate the method in this article. I?ll try to get it up on arXiv soon, but if anyone wants a copy please send me an email individually. Best, Dave From: statnet_help on behalf of Martina Morris Date: Friday, December 8, 2023 at 3:47?PM To: James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Subject: Re: [statnet_help] fragmented bipartite network... This is a great conversation; many thanks to the contributors. As I read through the proposed stats, though, I keep stumbling on the bipartite bit: how would some of these translate into bip net terms? I appreciate Jim's effort to bring this back to practical advice. So, some really basic thoughts here. There are two general types of blocks: those based on exogenous attributes, and those based on endogenous processes. I think the reason we're circling around the idea of blocks is that these depictions tend to capture the clustering observed in real world networks, and that blocking can help explain why dyad-dependent effects operate locally, rather than globally across a network. The exogenous type of block is captured by nodemix and nodematch type terms in ergm (which have a number of different specifications). In the bip net context these terms become more complicated as they no longer represent the crosstabulation of pairwise nodal attributes, but instead a crosstab of the terminal node attributes of a 2-mode triad. What's interesting about the bip net version of these terms is that this 2-path configuration is also a building block of equivalence. More on this below. The endogenous type of block is captured as latent block structures in hergms (for the ergm framework, other frameworks are out there). HERGMs are an interesting approach to identifying observed or latent neighborhoods of dependence (https://www.jstatsoft.org/article/view/v085i01), but I don't know if the package (or the models) can handle bipartite nets. I've added Michael Schweinberger to this email in case he would like to comment. Back to the exogenous blocking then. Family name could be a powerful blocking effect (e.g. Jim's example of Tata), showing up in this bip net as org board memberships shared by people with the same family name. Ignoring the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner (NSP) statistics. If two people shared all of their org memberships, they are structurally equivalent (whether they share an exogenous attribute or not) -- and more generally, the more NSPs, the higher the equivalence. And if the nodal name attribute is not driving these 2 paths, these high value NSPs are indicators of latent structure. The 2-paths can also be used to examine the org equivalence pattern in the same way. And my intuition would be that, conditioned on density, NSP distributions with higher means or longer tails would lead to fragmentation in the network. So, that makes me think perhaps the place to start is with EDA -- look at the NSP distributions, for both persons and orgs. Compare these to the expected distributions under a simple null random graph. If the distributions differ significantly, then start to look for exogenous effects that help to explain the deviation from the null (using the bip homophily terms with some more attributes on the nodes of both modes). And look into whether endogenously defined blocks (a la HERGM) can be used for bip nets. For me, the ideal would be to identify the latent blocks, and then explain almost all of that blocking in terms of exogenous/observed attributes. The blocks capture the structure. The explicit exogenous effects "explain" it. best, mm On Fri, Dec 8, 2023 at 6:28?AM James Moody > wrote: Fun discussion, thanks for sharing, always learn something in these sorts of posts. As to this this application per se; a couple of pragmatic (i.e. may not be elegant!) ideas: - theory should be able to inform some unlikely mixing that one could specify using a mixingmatrix term or two, no? So family, private/public, industry, etc. - For many business group applications, the actual family name is embedded in many of the subsidiaries (Tata group, tata inc, tata industries, etc.) so a name-similarity score could help (if you have nodenames) - The interlock limit will be size of the boards. While its possible to change the size of each board in a company, its not trivial, and I think you can justifiably take that as exogenous in the time-frame you have. I?m betting most of your small components are single family companies without external board memberships. Those create small stars in the bipartiate network (cliques in the projection). So that would imply: a) a hard-constraint on target degree. You could just fix that as a constraint. Again, not elegant (Carter?s cutting at joints and all), but likely true. b) a size mixing logic. Family-only/small-board cliques are isolated, leaving big-with-big, so there?s effectively a two-mode degree assortativity here. If you can?t induce this by an attribute (family name/ownership), then use assortativity on degree. - Cheating a little, but you could make component membership at attribute and hard-code mixing within/between. That means you can?t model what drives membership in the largest components vs. the small fractions, but, again, this is such a weird case (from a graph expectation sense), as anything that had even a little random noise in it would link across those small components, so the restriction here is almost certainly a legal/possibility restriction that should be treated as exogenous. - that?s, of course, just the crudest version of Daniel?s idea ? find a structural pattern that implies high/low probability of mixing across modes and hard-code it. I.e. do some old-fashioned inductive modeling of your network before the ERGM to generate classes of cases based on your best effort to induce the (to you) invisible restrictions patterning the ties, then add those back into the model as appropriate node/edge attributes. PTs Jim From: statnet_help > On Behalf Of Carter T. Butts Sent: Friday, December 8, 2023 4:53 AM To: statnet_help@u.washington.edu Subject: Re: [statnet_help] fragmented bipartite network... Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks. I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like. Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases). My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants. At any rate, models for differential mixing with discrete group structure are well-trod. As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term). The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern. Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments. (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize. But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions." It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so). So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad. If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals. Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one. (It will, in particular, have a form of Markov graph dependence.) Nothing says that your model has to belong to any particular dependence class - unless you want to impose such a condition. Of course, if you do want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph). In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong. Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result. Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out. (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.) For instance, k-cycles can be long-range, for large k. The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints). All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.). If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family. But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From steffentriebel at icloud.com Sat Dec 9 04:08:01 2023 From: steffentriebel at icloud.com (steffentriebel@icloud.com) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Message-ID: Dear Harald, I?ll also chime in, albeit with a less statistically profound lens than the others. First, I?ll encourage you to take a look at the manuscript David will share on arXiv; it may prove helpful and will hopefully allow you to capture theoretical considerations better. Second, maybe it makes sense to ?dumb down? your model a bit and take an iterative approach to refiner your theory. You write that there are many different types of ties to the second mode, ranging from off-shore companies to businesses or non-profits. It is probably safe to assume that all of these will follow different theoretical logics ? e.g., for businesses, we know that geographical proximity plays a major role in business networks as well as sectors (in less regulated economies, at least), but this will likely not be true for off-shore affiliations, which will perhaps be facilitated through the same broker organizing these off-shore affiliations? That would imply a different mechanism leading to the fragmented components you?re observing. These different institutional logics will be difficult to capture. Remember, the components you observe are a function of these (social) mechanisms ? at least typically ? and not a driving force. So, I think obtaining clarity on which mechanisms theory (and prior research) suggests to be especially pertinent will help obtain a clearer picture of what?s happening. I?m sure you did your due diligence here, but with networks as complex as this, it might make sense first to understand the different micro-processes underpinning them better, refine your theory, and then tackle the ?full network?. Perhaps you could model the bipartite affiliation per organizational type in the second mode and include dyadic covariates for ?on the same non-profit?, ?on the same company board?, .. depending on which network you are modelling? I assume this could help with honing in on the solution. Best wishes & best of luck Steffen Von: statnet_help im Auftrag von Hunter, David Datum: Samstag, 9. Dezember 2023 um 03:45 An: Martina Morris , James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Betreff: Re: [statnet_help] fragmented bipartite network... Following up on Martina?s observations among others? In case it helps, the b1nodematch and b2nodematch terms in the ergm package do not merely provide a census of 2-paths with matching end-nodes. They do provide this census, but merely as one end of a spectrum (two spectra, actually) of statistics created in the same spirit as the geometrically weighted statistics (GWESP, GWD, etc.) pioneered by Snijders et al back in 2006 (?New Specifications for Exponential Random Graph Models?). The full spectra entail a more flexible way to capture homophily in a bipartite network. We?ve just submitted a manuscript on this, and coincidentally we use a bipartite network of interlocking directorates to illustrate the method in this article. I?ll try to get it up on arXiv soon, but if anyone wants a copy please send me an email individually. Best, Dave From: statnet_help on behalf of Martina Morris Date: Friday, December 8, 2023 at 3:47?PM To: James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Subject: Re: [statnet_help] fragmented bipartite network... This is a great conversation; many thanks to the contributors. As I read through the proposed stats, though, I keep stumbling on the bipartite bit: how would some of these translate into bip net terms? I appreciate Jim's effort to bring this back to practical advice. So, some really basic thoughts here. There are two general types of blocks: those based on exogenous attributes, and those based on endogenous processes. I think the reason we're circling around the idea of blocks is that these depictions tend to capture the clustering observed in real world networks, and that blocking can help explain why dyad-dependent effects operate locally, rather than globally across a network. The exogenous type of block is captured by nodemix and nodematch type terms in ergm (which have a number of different specifications). In the bip net context these terms become more complicated as they no longer represent the crosstabulation of pairwise nodal attributes, but instead a crosstab of the terminal node attributes of a 2-mode triad. What's interesting about the bip net version of these terms is that this 2-path configuration is also a building block of equivalence. More on this below. The endogenous type of block is captured as latent block structures in hergms (for the ergm framework, other frameworks are out there). HERGMs are an interesting approach to identifying observed or latent neighborhoods of dependence (https://www.jstatsoft.org/article/view/v085i01), but I don't know if the package (or the models) can handle bipartite nets. I've added Michael Schweinberger to this email in case he would like to comment. Back to the exogenous blocking then. Family name could be a powerful blocking effect (e.g. Jim's example of Tata), showing up in this bip net as org board memberships shared by people with the same family name. Ignoring the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner (NSP) statistics. If two people shared all of their org memberships, they are structurally equivalent (whether they share an exogenous attribute or not) -- and more generally, the more NSPs, the higher the equivalence. And if the nodal name attribute is not driving these 2 paths, these high value NSPs are indicators of latent structure. The 2-paths can also be used to examine the org equivalence pattern in the same way. And my intuition would be that, conditioned on density, NSP distributions with higher means or longer tails would lead to fragmentation in the network. So, that makes me think perhaps the place to start is with EDA -- look at the NSP distributions, for both persons and orgs. Compare these to the expected distributions under a simple null random graph. If the distributions differ significantly, then start to look for exogenous effects that help to explain the deviation from the null (using the bip homophily terms with some more attributes on the nodes of both modes). And look into whether endogenously defined blocks (a la HERGM) can be used for bip nets. For me, the ideal would be to identify the latent blocks, and then explain almost all of that blocking in terms of exogenous/observed attributes. The blocks capture the structure. The explicit exogenous effects "explain" it. best, mm On Fri, Dec 8, 2023 at 6:28?AM James Moody > wrote: Fun discussion, thanks for sharing, always learn something in these sorts of posts. As to this this application per se; a couple of pragmatic (i.e. may not be elegant!) ideas: - theory should be able to inform some unlikely mixing that one could specify using a mixingmatrix term or two, no? So family, private/public, industry, etc. - For many business group applications, the actual family name is embedded in many of the subsidiaries (Tata group, tata inc, tata industries, etc.) so a name-similarity score could help (if you have nodenames) - The interlock limit will be size of the boards. While its possible to change the size of each board in a company, its not trivial, and I think you can justifiably take that as exogenous in the time-frame you have. I?m betting most of your small components are single family companies without external board memberships. Those create small stars in the bipartiate network (cliques in the projection). So that would imply: a) a hard-constraint on target degree. You could just fix that as a constraint. Again, not elegant (Carter?s cutting at joints and all), but likely true. b) a size mixing logic. Family-only/small-board cliques are isolated, leaving big-with-big, so there?s effectively a two-mode degree assortativity here. If you can?t induce this by an attribute (family name/ownership), then use assortativity on degree. - Cheating a little, but you could make component membership at attribute and hard-code mixing within/between. That means you can?t model what drives membership in the largest components vs. the small fractions, but, again, this is such a weird case (from a graph expectation sense), as anything that had even a little random noise in it would link across those small components, so the restriction here is almost certainly a legal/possibility restriction that should be treated as exogenous. - that?s, of course, just the crudest version of Daniel?s idea ? find a structural pattern that implies high/low probability of mixing across modes and hard-code it. I.e. do some old-fashioned inductive modeling of your network before the ERGM to generate classes of cases based on your best effort to induce the (to you) invisible restrictions patterning the ties, then add those back into the model as appropriate node/edge attributes. PTs Jim From: statnet_help > On Behalf Of Carter T. Butts Sent: Friday, December 8, 2023 4:53 AM To: statnet_help@u.washington.edu Subject: Re: [statnet_help] fragmented bipartite network... Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks. I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like. Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases). My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants. At any rate, models for differential mixing with discrete group structure are well-trod. As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term). The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern. Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments. (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize. But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions." It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so). So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad. If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals. Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one. (It will, in particular, have a form of Markov graph dependence.) Nothing says that your model has to belong to any particular dependence class - unless you want to impose such a condition. Of course, if you do want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph). In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong. Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result. Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out. (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.) For instance, k-cycles can be long-range, for large k. The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints). All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.). If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family. But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From morrism at uw.edu Sat Dec 9 10:54:55 2023 From: morrism at uw.edu (Martina Morris) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> Message-ID: Those are all great suggestions Steffen -- thx for posting :) On Sat, Dec 9, 2023 at 4:08?AM steffentriebel@icloud.com < steffentriebel@icloud.com> wrote: > Dear Harald, I?ll also chime in, albeit with a less statistically profound > lens than the others. First, I?ll encourage you to take a look at the > manuscript David will share on arXiv; it may prove helpful and will > hopefully allow you to capture > ZjQcmQRYFpfptBannerStart > This Message Is From an Untrusted Sender > You have not previously corresponded with this sender. > See https://itconnect.uw.edu/email-tags for additional information. > Please contact the UW-IT Service Center, help@uw.edu 206.221.5000, for > assistance. > > ZjQcmQRYFpfptBannerEnd > > Dear Harald, > > > > I?ll also chime in, albeit with a less statistically profound lens than > the others. > > > > First, I?ll encourage you to take a look at the manuscript David will > share on arXiv; it may prove helpful and will hopefully allow you to > capture theoretical considerations better. > > > > Second, maybe it makes sense to ?dumb down? your model a bit and take an > iterative approach to refiner your theory. You write that there are many > different types of ties to the second mode, ranging from off-shore > companies to businesses or non-profits. It is probably safe to assume that > all of these will follow different theoretical logics ? e.g., for > businesses, we know that geographical proximity plays a major role in > business networks as well as sectors (in less regulated economies, at > least), but this will likely not be true for off-shore affiliations, which > will perhaps be facilitated through the same broker organizing these > off-shore affiliations? That would imply a different mechanism leading to > the fragmented components you?re observing. These different institutional > logics will be difficult to capture. > > > > Remember, the components you observe are a function of these (social) > mechanisms ? at least typically ? and not a driving force. So, I think > obtaining clarity on which mechanisms theory (and prior research) suggests > to be especially pertinent will help obtain a clearer picture of what?s > happening. I?m sure you did your due diligence here, but with networks as > complex as this, it might make sense first to understand the different > micro-processes underpinning them better, refine your theory, and then > tackle the ?full network?. Perhaps you could model the bipartite > affiliation per organizational type in the second mode and include dyadic > covariates for ?on the same non-profit?, ?on the same company board?, .. > depending on which network you are modelling? I assume this could help with > honing in on the solution. > > > > Best wishes & best of luck > > Steffen > > > > *Von: *statnet_help im > Auftrag von Hunter, David > *Datum: *Samstag, 9. Dezember 2023 um 03:45 > *An: *Martina Morris , James Moody > *Cc: *statnet_help@u.washington.edu , > Schweinberger, Michael > *Betreff: *Re: [statnet_help] fragmented bipartite network... > > Following up on Martina?s observations among others? > > > > In case it helps, the b1nodematch and b2nodematch terms in the ergm > package do not merely provide a census of 2-paths with matching end-nodes. > They do provide this census, but merely as one end of a spectrum (two > spectra, actually) of statistics created in the same spirit as the > geometrically weighted statistics (GWESP, GWD, etc.) pioneered by Snijders > et al back in 2006 (?New Specifications for Exponential Random Graph > Models?). The full spectra entail a more flexible way to capture homophily > in a bipartite network. > > > > We?ve just submitted a manuscript on this, and coincidentally we use a > bipartite network of interlocking directorates to illustrate the method in > this article. I?ll try to get it up on arXiv soon, but if anyone wants a > copy please send me an email individually. > > > > Best, > > Dave > > > > *From: *statnet_help on > behalf of Martina Morris > *Date: *Friday, December 8, 2023 at 3:47?PM > *To: *James Moody > *Cc: *statnet_help@u.washington.edu , > Schweinberger, Michael > *Subject: *Re: [statnet_help] fragmented bipartite network... > > This is a great conversation; many thanks to the contributors. > > > > As I read through the proposed stats, though, I keep stumbling on the > bipartite bit: how would some of these translate into bip net terms? I > appreciate Jim's effort to bring this back to practical advice. > > > > So, some really basic thoughts here. There are two general types of > blocks: those based on exogenous attributes, and those based on endogenous > processes. I think the reason we're circling around the idea of blocks is > that these depictions tend to capture the clustering observed in real world > networks, and that blocking can help explain why dyad-dependent effects > operate locally, rather than globally across a network. > > > > The exogenous type of block is captured by nodemix and nodematch type > terms in ergm (which have a number of different specifications). In the > bip net context these terms become more complicated as they no longer > represent the crosstabulation of pairwise nodal attributes, but instead a > crosstab of the terminal node attributes of a 2-mode triad. What's > interesting about the bip net version of these terms is that this 2-path > configuration is also a building block of equivalence. More on this below. > > > > The endogenous type of block is captured as latent block structures in > hergms (for the ergm framework, other frameworks are out there). HERGMs > are an interesting approach to identifying observed or latent neighborhoods > of dependence (https://www.jstatsoft.org/article/view/v085i01 > ), > but I don't know if the package (or the models) can handle bipartite nets. > > > > I've added Michael Schweinberger to this email in case he would like to > comment. > > > > Back to the exogenous blocking then. Family name could be a powerful > blocking effect (e.g. Jim's example of Tata), showing up in this bip net as > org board memberships shared by people with the same family name. Ignoring > the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner > (NSP) statistics. If two people shared all of their org memberships, they > are structurally equivalent (whether they share an exogenous attribute or > not) -- and more generally, the more NSPs, the higher the equivalence. And > if the nodal name attribute is not driving these 2 paths, these high value > NSPs are indicators of latent structure. > > > > The 2-paths can also be used to examine the org equivalence pattern in the > same way. > > > > And my intuition would be that, conditioned on density, NSP distributions > with higher means or longer tails would lead to fragmentation in the > network. > > > > So, that makes me think perhaps the place to start is with EDA -- look at > the NSP distributions, for both persons and orgs. Compare these to the > expected distributions under a simple null random graph. If the > distributions differ significantly, then start to look for exogenous > effects that help to explain the deviation from the null (using the bip > homophily terms with some more attributes on the nodes of both modes). And > look into whether endogenously defined blocks (a la HERGM) can be used for > bip nets. For me, the ideal would be to identify the latent blocks, and > then explain almost all of that blocking in terms of exogenous/observed > attributes. The blocks capture the structure. The explicit exogenous > effects "explain" it. > > > > best, > > mm > > > > > > On Fri, Dec 8, 2023 at 6:28?AM James Moody wrote: > > Fun discussion, thanks for sharing, always learn something in these sorts > of posts. > > > > As to this this application per se; a couple of pragmatic (i.e. may not be > elegant!) ideas: > > > > - theory should be able to inform some unlikely mixing that one could > specify using a mixingmatrix term or two, no? So family, private/public, > industry, etc. > > - For many business group applications, the actual family name is > embedded in many of the subsidiaries (Tata group, tata inc, tata > industries, etc.) so a name-similarity score could help (if you have > nodenames) > > - The interlock limit will be size of the boards. While its possible to > change the size of each board in a company, its not trivial, and I think > you can justifiably take that as exogenous in the time-frame you have. I?m > betting most of your small components are single family companies without > external board memberships. Those create small stars in the bipartiate > network (cliques in the projection). So that would imply: > > a) a hard-constraint on target degree. You could just fix that as a > constraint. Again, not elegant (Carter?s cutting at joints and all), but > likely true. > > b) a size mixing logic. Family-only/small-board cliques are isolated, > leaving big-with-big, so there?s effectively a two-mode degree > assortativity here. If you can?t induce this by an attribute (family > name/ownership), then use assortativity on degree. > > - Cheating a little, but you could make component membership at attribute > and hard-code mixing within/between. That means you can?t model what drives > membership in the largest components vs. the small fractions, but, again, > this is such a weird case (from a graph expectation sense), as anything > that had even a little random noise in it would link across those small > components, so the restriction here is almost certainly a legal/possibility > restriction that should be treated as exogenous. > > - that?s, of course, just the crudest version of Daniel?s idea ? find a > structural pattern that implies high/low probability of mixing across modes > and hard-code it. I.e. do some old-fashioned inductive modeling of your > network before the ERGM to generate classes of cases based on your best > effort to induce the (to you) invisible restrictions patterning the ties, > then add those back into the model as appropriate node/edge attributes. > > > > PTs > > Jim > > > > > > > > *From:* statnet_help *On > Behalf Of *Carter T. Butts > *Sent:* Friday, December 8, 2023 4:53 AM > *To:* statnet_help@u.washington.edu > *Subject:* Re: [statnet_help] fragmented bipartite network... > > > > Hi, Daniel - > > Most of the cases to which I believe you are referring deal with > differential mixing; the "blocks" here are what are sometimes called > "density" blocks, which are quantitative relaxations of the complete/null > blocks. I don't think anyone doubts that differential mixing exists, but > that is very far from e.g. nontrivial global automorphism orbits or the > like. Indeed, John Boyd had a running bet for some years, in which he > offered to pay a sum of money (I forget how much) to anyone who could show > a statistically significant regular equivalence pattern (above and beyond > SE - he also had some other boundary conditions that ruled out "easy" > cases). My vague recollection was that Steve Borgatti claimed to have one, > and they then haggled over John's way of calculating "significance," but my > memory on the subject is hazy and doubtless untrustworthy; I never did buy > John's extreme conjecture, but it is true that he was not exactly > overwhelmed with claimants. At any rate, models for differential mixing > with discrete group structure are well-trod. As far as other kinds of > generalized blocks (moving away from complete/null blocks), you can fit > models with strict versions of e.g. regular, row/column dominant, and > row/column functional blocks with clever use of constraints (in ergm, the > bd() constraint term). The most obvious path to soft versions of those > block types is to create statistics that count violations of the block > pattern. Some can be implemented using the degrange() term, together with > appropriate use of the optional attribute arguments. (Obviously, these are > all "confirmatory" models, in the sense that one has to specify the block > structure one wants to impose/parameterize. But that is not without its > virtues.) > > Vis a vis dependence, I'm not sure that it is very helpful to think in > terms of "violating assumptions." It is probably more useful to think of > H-C and friends as giving you a "recipe" for the statistics you need to > implement particular kinds of dependence conditions (should you want to do > so). So, e.g., if you want edges to depend on each other when they share > endpoints, then you will want (in the unvalued case) indicators for each > edge variable, and indicators for each mutual dyad. If you also want the > corresponding effects to be homogeneous, then this reduces to the edge > count and the count of mutuals. Adding e.g. a 2-outstar term to a model > with edges and mutuals is not violating any particular assumption imposed > by the latter - it's just that this new model will now belong to a > different (and broader) dependence class than the original one. (It will, > in particular, have a form of Markov graph dependence.) Nothing says that > your model has to belong to *any* particular dependence class - unless > you want to impose such a condition. Of course, if you *do *want to > restrict your dependence to a particular class, then you will indeed need > to ensure that your statistics are a subset of those admitted by that class > (which, for H-C, can be determined from the cliques of the conditional > dependence graph). In my experience, this is rarely a useful way to > proceed; however, it sometimes can be handy to know the type of dependence > class to which your terms belong. Likewise, it can sometimes be handy to > start by positing a form of dependence that makes sense in a specific > situation, and then deriving the statistics that result. Pip, in > particular, has done a great deal to elucidate these sorts of connections. > > As far as long-range dependence, there's again nothing ruling it out. > (Pip and Tom, IIRC, have a very nice typology working out statistics for > dependence classes at different distances.) For instance, k-cycles can be > long-range, for large k. The various component and bridging statistics can > be arbitrarily long-range. The statistics that arise from density and dyad > census mixtures do them one better by being completely global (i.e., they > create conditional dependence between edge variables irrespective of > whether there is even a path of any length between their endpoints). All > of these lead to well-defined models - those models just happen not to > belong e.g. to the Markov graphs (or the social circuit graphs, the > Bernoulli graphs, the u|man family, etc.). If there is a reason that you > need your model to belong to such a family, then you would not want to use > terms that are not within the class specifying that family. But otherwise, > such restrictions are arbitrary, and may get in the way of specifying > important mechanisms. > > Hope that helps, > > -Carter > > > > On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: > > Hello Carter, > > i agree that stricter types oft equivalence are very rare and I would > personally also look at either generalized blockmodeling or actually just > measures of structural or positional similarity - but indeed not only local > ones (which are already included in ergm of course). I did mention them > here because most results of the relevance of more global equivalence > structures I know have been found in especially kinship research and > organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional > fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some > recent research in foreign trade and political conflicts that indicate that > block structures might matter (Guler et al. 2002, Zhou & Park 2012, > Olivella et al. 2022). I am curious though which tools you are thinking > about for implementing aspects oft generalized block structures? > > Regarding hammersley-clifford I mostly wanted to be careful here, but I > did think that H-C and extensions like social circuit dependency (which > allows partial depensence) did matter to ensure some (conditional) > independence assumption with a few parameters (one for each clique of the > dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block > er al. 2019). I thought dependencies (far) beyond the local neigborhood > might violate these properties. This is probably beyond Harald's concerns > but I would be happy if you could indicate any literature to alleviate my > misunderstanding. > > Best Regards > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, > Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > ------------------------------ > > *Von:* statnet_help > im Auftrag von Carter > T. Butts > *Gesendet:* Freitag, 8. Dezember 2023 07:06:46 > *An:* statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > > > > Local automorphism orbits and their associations with covariates can be > modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial > *global* automorphisms are extremely rare in typical social networks, so > such terms would be unlikely to be useful - what one might call the "strong > algebraic paradigm" of network analysis (the idea that we could explain > most social network structure in terms of small numbers of roles, as > defined through algebraic equivalences) was a very compelling idea that > didn't really work out, and I don't think many folks are pushing in that > direction right now. (See also compositional factorization, as famously > illustrated by the semigroup on the cover of Wasserman and Faust (1994). > Beautiful idea with some lovely technical results, but one with few if any > real-world success stories. Sometimes, things just don't work out.) I > think there could be some potential uses for terms for adherence to > (confirmatory) generalized blockmodel structure (in the > Doreian/Ferligoj/Batagelj tradition), though some of this can already be > emulated using existing tools; there has also been a relative dearth of > empirical cases in which complex block types have been shown to be > important for capturing network structure. If such cases were to become > more often encountered, this would naturally motivate more work to model > them. > > With respect to your second comment, I am not sure what you mean by > "violating" Hammersley-Clifford. H-C provides one way of establishing an > equivalence between sets of network statistics and associated dependence > conditions; Pip Pattison, Gary Robbins, and others have obtained various > refinements to the original result (allowing for more subtle conditions to > be treated). H-C and friends simply say (effectively) that certain classes > of statistics implement certain kinds of dependence. These are important > results for constructing and interpreting statistics, but they are not > rules that can be violated. > > Hope that clarifies things, > > -Carter > > On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: > > Dear Harald, > > after Martinas very insightful message and considering that you have > kinship and business ties but not so many node covariates, I am wondering > if you need or should think of structural equivalance as a driving factor. > With White and others there is a strong tradition of focussing on this for > kinship networks and DiMaggio and Burt have studied the importance oft > business roles and structural position. In your case that probably means > non-local forms of equivalence (automorphic, role, etc) that might matter > directly in the network behavior or could represent unmeasured node > attributes. Feature and embedding based measures are more scalable and now > allow to measure those concepts better in larger networks. > > To the best of my knowledge this is not considered offen in generative > network models and i don't think that we can include those less-localized > mechanisms directly (yet). Plesae let me know if this is a direction that > makes sense for you from a theoretical point of view and also something > that could be identified in your data. I am currently working on this in > the context oft actor-oriented models but am interested in the potential of > ergms in this regard as well. At least as exogenous covariates this might > be possible but otherwise we might violate conditional independence > (Hammersley-Clifford theorem). I am curious to hear about the thoughts of > experienced ergm modelers on this, though. > > Best Regards, > Daniel > > -- > Daniel Gotthardt, M.A. > > Wissenschaftlicher Mitarbeiter / Research Associate > > Universit?t Hamburg > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, > Economics and Social Sciences > Fachbereich Sozialwissenschaften / Department of Social Sciences > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > Social Science > > Max-Brauer-Allee 60 > 22765 Hamburg > www.uni-hamburg.de > > ------------------------------ > > *Von:* Martina Morris > *Gesendet:* Donnerstag, 7. Dezember 2023 23:45:59 > *An:* Harald Waxenecker > *Cc:* Gotthardt, Daniel; statnet_help@u.washington.edu > *Betreff:* Re: [statnet_help] fragmented bipartite network... > > > > Hi Harald, > > > > You do have a complicated analysis here, and I'm a bit under-equipped to > help you Dx what is going on, as I don't have much experience with either > bipartite or multi-level nets (let alone both together!). > > > > What I can say, though, is that factor and covariate effects on the nodes > are, in the non-multilevel context, one of the most important brakes on the > feedback effects caused by dyad-dependent terms, making them more > well-behaved and more likely to produce the kinds of networks we actually > observe (caveat: sometimes those dependent effects are needed, see Carter's > work on amyloid fibrils). > > > > In this case, it seems like you don't have many attributes to work with -- > indeed, only on one of the modes. For gender, I would fit as a factor btw, > not a quantitative covariate, tho if there are only 2 levels this will not > have much impact. But when I think about the goals of board composition in > non-profits (the closest I get to your world), it's clear that gender is > not the only attribute that influences board member invitations -- and I > would expect the same would be true here. You might try adding family > name as a bxnodefactor (will pick up both family size and family activity > level differentials), or sociality for either (or both) modes (to condition > on the degree of each node). Your additional terms can then be interpreted > as effects operating beyond these differences in degree. Degree > distributions definitely influence component size distributions, up to a > point, so if your model is not getting these right, you can start there. > > > > Thinking about the orgs, it seems there must be org attributes that > influence the size and composition of the board. Org size, sector, > geographic location, age, specialization, etc. -- I can imagine all of > these would influence board memberships. Properties these nodes show in > the other nets you have might be able to be represented on the cheap here > as nodal attributes in this network. If these effects are at work -- and if > you're not including them in the model, it is a form of mis-specification > that compromises all of the other model estimates. > > > > Then there's homophily, which works differently in bip nets -- for one, > it's a dyad-dependent term. But it's also more complicated to think > about. Perhaps families might choose to specialize in an org sector, or > maybe the opposite, they aim to integrate across sectors. Orgs might want > diversity (on some measure) for members, which would show up as > anti-homophily in bip two-paths. Again though, this would require more > measured attributes for both orgs and persons. > > > > Adding model terms like components is different. In my modeling world, we > want our (parsimonious) models to represent the mechanistic effects that > may actually generate the ties in the network. For us, component size > distributions are an *output* of a network formation process, not the > generating mechanism (people aren't creating ties with the explicit intent > of structuring the network component size distributions, with one key > exception, and that we do model). We instead use the component size > distribution as a goodness-of-fit indicator -- to test whether the > mechanistic terms we included in our model reproduce these higher order > excluded network stats. > > > > But your context may be different. When an org board is formed, if there > is an explicit strategy to create specific component structures in the > overall network then those intentions should be included as model terms. I > can imagine that bridging structural holes might be one of those > strategies. But again, not my area of expertise. > > > > I'm not sure how much any of this helps your specific issues. But when > models don't fit the data properly, it's worth thinking about specification > from first principles. So I hope this helps. > > > > best, > > Martina > > > > On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: > > Dear Tom, Martina, Carter and Daniel > > Thank you for your supportive answers. > > > > First, I will try to address some of your questions. The dependent network > is a bipartite business network (6902 persons x 5178 companies), based > exclusively on interlocking directorates. This dependent bipartite network > represents the business ties of elite members in their home country. We > include two covariates for the first node set (persons): *traditional > surname* and *gender*. Isolates in this network represent elite members > without any business ties. We belief that isolated nodes are meaningful in > this network; e.g., women are often constrained to ?reproduction? rather > than participating in ?production? (businesses). However, in different > network layers they contribute to elite cohesion. > > > > Regarding these different layers: we have six more networks. The first is > a one-mode kinship network (6902x6902), and the others are bipartite > networks (based on interlocks), where persons form the first node set and > entities the second. Hence, all matrices share a consistent number of rows > (n = 6902), while the number of columns varies according to the number of > entities in each network layer: offshore companies in Panama (n = 1537), > business associations (n = 128), non-profit organizations (n = 236), > political parties (n = 55), and public entities (n = 431). > > > > We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) > https://doi.org/10.1017/S0143814X18000181 > , > to test whether a common property (?homophily?) of the nodes in the first > node set, such as a shared attribute (gender, traditional surname), a > direct tie (kinship relation), or a mutual membership in other bipartite > layers (offshore companies, business associations, etc.) contribute to the > probability of two individuals forming ties with the same company in the > dependent network. > > > > Regarding the modeling process, it?s true that the model we shared relies > only on dyad-dependent terms. We always ?come back? to this model > specification because all our attempts, which certainly were also based > primarily on dyad-dependent terms, did not produce better results. We > explored various options, including nodematch to control for component > membership to split the network into smaller fragments. Then we > incorporated component membership of the nodes as constraint to induce > network fragmentation. While this partially improved network fragmentation, > problems with goodness-of-fit persisted. Additionally, we encountered some > computational limitations while running these options. > > > > Now, we have incorporated several of your recommendations, introducing > dyad-independent terms and utilizing components() from the ergm.components > package. Please find the new outcomes (model 0) attached. We've also > attached summary files and component distribution for a comparative > analysis between the observed network and the simulated network. > > > > We also tried to include the terms compsizesum() and dimers() into the > model; however, we observe degeneracy issues. In addition, we still could > not get results with bridges(), because it seems to be very time consuming > and/or needs much computational capacity. > > > > I think this bridges-term relates somehow to your question @Martina about > cross-group ties in the simulated data. Or maybe I am wrong. Please, could > you explain that in more detail? Thanks. > > > > Thank you again for your support. Looking very forward to read your > thoughts and advice. > > > > Kind regards, > > Harald > > > > > > > > > > > > > > > > > > El 1/12/23, 21:53, "[NOMBRE]" escribi?: > > Hello Harald, > > > > if I understand you correctly you have a within-mode network as well as > > a bipartite network. James Hollway et al. (2017) has described an > > approach to handle these kinds of combined networks as multilevel social > > spaces with stochastic actor-oriented models: > > > https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 > > > - There are also some tricks to transform these types of networks into > > an extended multimodal network matrix, exemplified e.g. in Knoke et al. > > (2021): > > > https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 > > > > > I personally don't know of any ergm model that can handle this kind of > > co-evolution of one-mode and two-mode networks but some kind of > > multilevel ergms (see Wang et al. (2013) > > https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051 > ) > > > might be the way to go: - I'm sure others here know more about the > > capabilities of ergm.multi though. > > > > If these kinship structures explain the fragmentation of the bipartite > > network, you might need to include them either directly with the > > approaches above or construct some corresponding dyadic or monadic > > covariates to represent the kinship structure in your single level network. > > > > Best Regards, > > > > Daniel > > > > Am 01.12.2023 um 02:13 schrieb Martina Morris: > > > > > > Hi Harald, > > > > > > I'm looking for some clarification here, which I think Tom Kraft might > > > also have wondered about. > > > > > > You say: > > >> > > >> Our research focuses on tie formation and elite cohesion, specifically > > >> examining interlocking directorates and kinship relations. The > > >> dependent bipartite business network comprises 6,902 individuals and > > >> 5,178 companies, exhibiting sparsity (density = 0.00012) and > > >> fragmentation with 4,455 components, including 3,850 isolates in the > > >> first mode (persons) > > >> > > > For a bipartite network ties are allowed only between modes (persons, > > > companies), not within. It's clear how interlocking directorates would > > > meet that criteria. But kinship relations would be among persons, so > > > within-mode, not between, and this would not be a bipartite network. > > > > > > Is the model you've sent us for the interlocking directorships only? > > > And by isolates in the person mode, do you mean persons who are not > > > affiliated with any of the companies? If so, then it's a bit odd to > > > include them in the bipartite network. > > > > > > I'm wondering if this problem is better posed as a multilevel network > > > (not my area of expertise). > > > > > > thanks, > > > Martina > > > > > > > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > > > wrote: > > > > > > __ > > > > > > Hi, Harald - > > > > > > Coexistence of large complex components does not generally occur > > > unless something drives the fragmentation, and this is what your > > > models are telling you: the terms you are currently using do not > > > include the forces that are sufficient to reproduce your component > > > size distribution. That means that you need to think about why your > > > network is split into fragments, and include terms that capture the > > > relevant social forces. Thinking about likely mechanisms is step > > > zero, so do that before anything else! Guided by your substantive > > > knowledge of what is likely going on, you will next (as others have > > > said) want to look at covariate effects relating to differential > > > mixing, since those are your most obvious and most important sources > > > of heterogeneity. If you find that there is still more > > > fragmentation that can be explained by other means, you may need to > > > consider model terms relating directly to component count or size. > > > These are still somewhat experimental, and are currently sequestered > > > in an add-on package called ergm.components > > > (https://github.com/statnet/ergm.components > > > > < > https://urldefense.com/v3/__https://github.com/statnet/ergm.components__;!!K-Hz7m0Vt54!iKts-XLv39sY0gvmpW6MWLIxNMCNKjKQKOhJszIbp3PIy_J5mdLCs0MytfHsBu-cjnQjk997tCRX0MMs6LDW$ > >). > However, this package can be installed from github (see the github page), > and the terms will work automagically with ergm() and friends once the > package is loaded. Depending on your situation, you may need or want to > examine the components() or compsizesum() terms, both of which are > documented within the package. > > > > > > Hope that helps, > > > > > > -Carter > > > > > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: > > >> > > >> Dear ?statnet community?,____ > > >> > > >> __ __ > > >> > > >> Our research focuses on tie formation and elite cohesion, > > >> specifically examining interlocking directorates and kinship > > >> relations. The dependent bipartite business network comprises > > >> 6,902 individuals and 5,178 companies, exhibiting sparsity > > >> (density = 0.00012) and fragmentation with 4,455 components, > > >> including 3,850 isolates in the first mode (persons). The attached > > >> documents contain descriptives and the component size distribution > > >> from the observed network.____ > > >> > > >> ____ > > >> > > >> The fragmented structure is important, as other network layers, > > >> like kinship relations, are expected to contribute to the cohesion > > >> of this business network. We apply ERGM to model these processes, > > >> but we struggle to capture the fragmented structure of the > > >> observed network. The component size distribution of the simulated > > >> network differs significantly. In addition, the goodness-of-fit > > >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) > > >> shows significant results. All these results are also attached.____ > > >> > > >> ____ > > >> > > >> We've explored various options, including constraints, MCMC > > >> propositions, and simulated annealing, but haven't achieved > > >> success. Please, we would like to ask for your help to improve our > > >> model. Thank you!____ > > >> > > >> __ __ > > >> > > >> Kind regards,____ > > >> > > >> Harald____ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> __ __ > > >> > > >> --- ____ > > >> > > >> *Harald Waxenecker > > >> > > >> *____ > > >> > > >> *Masaryk University | Faculty of social studies* > > >> Department of Environment Studies > > >> A: Jostova 10 | 602 00 Brno | Czech Republic > > >> E: waxenecker@fss.muni.cz ____ > > >> > > >> __ __ > > >> > > >> > > >> _______________________________________________ > > >> statnet_help mailing list > > >> statnet_help@u.washington.edu statnet_help@u.washington.edu> > > >> > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > < > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > > > > > > > > > > _______________________________________________ > > > statnet_help mailing list > > > statnet_help@u.washington.edu > > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > -- > > > > Daniel Gotthardt, M.A. > > > > Wissenschaftlicher Mitarbeiter / Research Associate > > > > Universit?t Hamburg > > Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of > > Business, Economics and Social Sciences > > Fachbereich Sozialwissenschaften / Department of Social Sciences > > Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital > > Social Science > > > > Max-Brauer-Allee 60 > > 22765 Hamburg > > www.uni-hamburg.de > > > > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > > _______________________________________________ > > statnet_help mailing list > > statnet_help@u.washington.edu > > https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > -------------- next part -------------- An HTML attachment was scrubbed... URL: From daniel.gotthardt at uni-hamburg.de Sat Dec 9 19:50:29 2023 From: daniel.gotthardt at uni-hamburg.de (Gotthardt, Daniel) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... In-Reply-To: References: <33f3df53-2f2c-4748-be94-334cfa85e66c@uci.edu> <4c743a78-f197-4c96-bc81-e2cc3c4d5875@uci.edu> , Message-ID: <739c56f21fe245458aa8e19b55b93524@uni-hamburg.de> Dear Harald, to add onto Steffen's points: organizational forms tend to differ in their behavior but they are also offen connected to other forms - that's one oft the fundamental ideas oft the organizational field (DiMaggio & Powell, 1986). So I think it does make sense to keep them together but you might want to try to identify how different their behavior is (or the behavior of the persons connected to them). First, you can take a look at the subsets as Steffen indicates but they could be even more fragmented. If there is serious heterogeneity you might be able capture that with with some care in the larger model while still allowing interactions between forms. Best Regards, Daniel ________________________________ Von: statnet_help im Auftrag von steffentriebel@icloud.com Gesendet: Samstag, 9. Dezember 2023 13:08:01 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Dear Harald, I?ll also chime in, albeit with a less statistically profound lens than the others. First, I?ll encourage you to take a look at the manuscript David will share on arXiv; it may prove helpful and will hopefully allow you to capture theoretical considerations better. Second, maybe it makes sense to ?dumb down? your model a bit and take an iterative approach to refiner your theory. You write that there are many different types of ties to the second mode, ranging from off-shore companies to businesses or non-profits. It is probably safe to assume that all of these will follow different theoretical logics ? e.g., for businesses, we know that geographical proximity plays a major role in business networks as well as sectors (in less regulated economies, at least), but this will likely not be true for off-shore affiliations, which will perhaps be facilitated through the same broker organizing these off-shore affiliations? That would imply a different mechanism leading to the fragmented components you?re observing. These different institutional logics will be difficult to capture. Remember, the components you observe are a function of these (social) mechanisms ? at least typically ? and not a driving force. So, I think obtaining clarity on which mechanisms theory (and prior research) suggests to be especially pertinent will help obtain a clearer picture of what?s happening. I?m sure you did your due diligence here, but with networks as complex as this, it might make sense first to understand the different micro-processes underpinning them better, refine your theory, and then tackle the ?full network?. Perhaps you could model the bipartite affiliation per organizational type in the second mode and include dyadic covariates for ?on the same non-profit?, ?on the same company board?, .. depending on which network you are modelling? I assume this could help with honing in on the solution. Best wishes & best of luck Steffen Von: statnet_help im Auftrag von Hunter, David Datum: Samstag, 9. Dezember 2023 um 03:45 An: Martina Morris , James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Betreff: Re: [statnet_help] fragmented bipartite network... Following up on Martina?s observations among others? In case it helps, the b1nodematch and b2nodematch terms in the ergm package do not merely provide a census of 2-paths with matching end-nodes. They do provide this census, but merely as one end of a spectrum (two spectra, actually) of statistics created in the same spirit as the geometrically weighted statistics (GWESP, GWD, etc.) pioneered by Snijders et al back in 2006 (?New Specifications for Exponential Random Graph Models?). The full spectra entail a more flexible way to capture homophily in a bipartite network. We?ve just submitted a manuscript on this, and coincidentally we use a bipartite network of interlocking directorates to illustrate the method in this article. I?ll try to get it up on arXiv soon, but if anyone wants a copy please send me an email individually. Best, Dave From: statnet_help on behalf of Martina Morris Date: Friday, December 8, 2023 at 3:47?PM To: James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Subject: Re: [statnet_help] fragmented bipartite network... This is a great conversation; many thanks to the contributors. As I read through the proposed stats, though, I keep stumbling on the bipartite bit: how would some of these translate into bip net terms? I appreciate Jim's effort to bring this back to practical advice. So, some really basic thoughts here. There are two general types of blocks: those based on exogenous attributes, and those based on endogenous processes. I think the reason we're circling around the idea of blocks is that these depictions tend to capture the clustering observed in real world networks, and that blocking can help explain why dyad-dependent effects operate locally, rather than globally across a network. The exogenous type of block is captured by nodemix and nodematch type terms in ergm (which have a number of different specifications). In the bip net context these terms become more complicated as they no longer represent the crosstabulation of pairwise nodal attributes, but instead a crosstab of the terminal node attributes of a 2-mode triad. What's interesting about the bip net version of these terms is that this 2-path configuration is also a building block of equivalence. More on this below. The endogenous type of block is captured as latent block structures in hergms (for the ergm framework, other frameworks are out there). HERGMs are an interesting approach to identifying observed or latent neighborhoods of dependence (https://www.jstatsoft.org/article/view/v085i01), but I don't know if the package (or the models) can handle bipartite nets. I've added Michael Schweinberger to this email in case he would like to comment. Back to the exogenous blocking then. Family name could be a powerful blocking effect (e.g. Jim's example of Tata), showing up in this bip net as org board memberships shared by people with the same family name. Ignoring the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner (NSP) statistics. If two people shared all of their org memberships, they are structurally equivalent (whether they share an exogenous attribute or not) -- and more generally, the more NSPs, the higher the equivalence. And if the nodal name attribute is not driving these 2 paths, these high value NSPs are indicators of latent structure. The 2-paths can also be used to examine the org equivalence pattern in the same way. And my intuition would be that, conditioned on density, NSP distributions with higher means or longer tails would lead to fragmentation in the network. So, that makes me think perhaps the place to start is with EDA -- look at the NSP distributions, for both persons and orgs. Compare these to the expected distributions under a simple null random graph. If the distributions differ significantly, then start to look for exogenous effects that help to explain the deviation from the null (using the bip homophily terms with some more attributes on the nodes of both modes). And look into whether endogenously defined blocks (a la HERGM) can be used for bip nets. For me, the ideal would be to identify the latent blocks, and then explain almost all of that blocking in terms of exogenous/observed attributes. The blocks capture the structure. The explicit exogenous effects "explain" it. best, mm On Fri, Dec 8, 2023 at 6:28?AM James Moody > wrote: Fun discussion, thanks for sharing, always learn something in these sorts of posts. As to this this application per se; a couple of pragmatic (i.e. may not be elegant!) ideas: - theory should be able to inform some unlikely mixing that one could specify using a mixingmatrix term or two, no? So family, private/public, industry, etc. - For many business group applications, the actual family name is embedded in many of the subsidiaries (Tata group, tata inc, tata industries, etc.) so a name-similarity score could help (if you have nodenames) - The interlock limit will be size of the boards. While its possible to change the size of each board in a company, its not trivial, and I think you can justifiably take that as exogenous in the time-frame you have. I?m betting most of your small components are single family companies without external board memberships. Those create small stars in the bipartiate network (cliques in the projection). So that would imply: a) a hard-constraint on target degree. You could just fix that as a constraint. Again, not elegant (Carter?s cutting at joints and all), but likely true. b) a size mixing logic. Family-only/small-board cliques are isolated, leaving big-with-big, so there?s effectively a two-mode degree assortativity here. If you can?t induce this by an attribute (family name/ownership), then use assortativity on degree. - Cheating a little, but you could make component membership at attribute and hard-code mixing within/between. That means you can?t model what drives membership in the largest components vs. the small fractions, but, again, this is such a weird case (from a graph expectation sense), as anything that had even a little random noise in it would link across those small components, so the restriction here is almost certainly a legal/possibility restriction that should be treated as exogenous. - that?s, of course, just the crudest version of Daniel?s idea ? find a structural pattern that implies high/low probability of mixing across modes and hard-code it. I.e. do some old-fashioned inductive modeling of your network before the ERGM to generate classes of cases based on your best effort to induce the (to you) invisible restrictions patterning the ties, then add those back into the model as appropriate node/edge attributes. PTs Jim From: statnet_help > On Behalf Of Carter T. Butts Sent: Friday, December 8, 2023 4:53 AM To: statnet_help@u.washington.edu Subject: Re: [statnet_help] fragmented bipartite network... Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks. I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like. Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases). My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants. At any rate, models for differential mixing with discrete group structure are well-trod. As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term). The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern. Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments. (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize. But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions." It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so). So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad. If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals. Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one. (It will, in particular, have a form of Markov graph dependence.) Nothing says that your model has to belong to any particular dependence class - unless you want to impose such a condition. Of course, if you do want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph). In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong. Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result. Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out. (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.) For instance, k-cycles can be long-range, for large k. The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints). All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.). If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family. But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: From waxenecker at fss.muni.cz Tue Dec 12 10:32:34 2023 From: waxenecker at fss.muni.cz (Harald Waxenecker) Date: Mon Mar 25 10:47:53 2024 Subject: [statnet_help] fragmented bipartite network... Message-ID: Dear all, Big thanks for the insightful discussion and all your valuable inputs. We're going to explore all the suggestions you've shared, and come back to the group with updates and, very likely also with more specific questions. But that will certainly take some time... Thanks again and best regards, Harald De: Gotthardt, Daniel Fecha: domingo, 10 de diciembre de 2023, 04:50 Para: steffentriebel@icloud.com , statnet_help@u.washington.edu Asunto: Re: [statnet_help] fragmented bipartite network... Dear Harald, to add onto Steffen's points: organizational forms tend to differ in their behavior but they are also offen connected to other forms - that's one oft the fundamental ideas oft the organizational field (DiMaggio & Powell, 1986). So I think it does make sense to keep them together but you might want to try to identify how different their behavior is (or the behavior of the persons connected to them). First, you can take a look at the subsets as Steffen indicates but they could be even more fragmented. If there is serious heterogeneity you might be able capture that with with some care in the larger model while still allowing interactions between forms. Best Regards, Daniel ________________________________ Von: statnet_help im Auftrag von steffentriebel@icloud.com Gesendet: Samstag, 9. Dezember 2023 13:08:01 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Dear Harald, I?ll also chime in, albeit with a less statistically profound lens than the others. First, I?ll encourage you to take a look at the manuscript David will share on arXiv; it may prove helpful and will hopefully allow you to capture theoretical considerations better. Second, maybe it makes sense to ?dumb down? your model a bit and take an iterative approach to refiner your theory. You write that there are many different types of ties to the second mode, ranging from off-shore companies to businesses or non-profits. It is probably safe to assume that all of these will follow different theoretical logics ? e.g., for businesses, we know that geographical proximity plays a major role in business networks as well as sectors (in less regulated economies, at least), but this will likely not be true for off-shore affiliations, which will perhaps be facilitated through the same broker organizing these off-shore affiliations? That would imply a different mechanism leading to the fragmented components you?re observing. These different institutional logics will be difficult to capture. Remember, the components you observe are a function of these (social) mechanisms ? at least typically ? and not a driving force. So, I think obtaining clarity on which mechanisms theory (and prior research) suggests to be especially pertinent will help obtain a clearer picture of what?s happening. I?m sure you did your due diligence here, but with networks as complex as this, it might make sense first to understand the different micro-processes underpinning them better, refine your theory, and then tackle the ?full network?. Perhaps you could model the bipartite affiliation per organizational type in the second mode and include dyadic covariates for ?on the same non-profit?, ?on the same company board?, .. depending on which network you are modelling? I assume this could help with honing in on the solution. Best wishes & best of luck Steffen Von: statnet_help im Auftrag von Hunter, David Datum: Samstag, 9. Dezember 2023 um 03:45 An: Martina Morris , James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Betreff: Re: [statnet_help] fragmented bipartite network... Following up on Martina?s observations among others? In case it helps, the b1nodematch and b2nodematch terms in the ergm package do not merely provide a census of 2-paths with matching end-nodes. They do provide this census, but merely as one end of a spectrum (two spectra, actually) of statistics created in the same spirit as the geometrically weighted statistics (GWESP, GWD, etc.) pioneered by Snijders et al back in 2006 (?New Specifications for Exponential Random Graph Models?). The full spectra entail a more flexible way to capture homophily in a bipartite network. We?ve just submitted a manuscript on this, and coincidentally we use a bipartite network of interlocking directorates to illustrate the method in this article. I?ll try to get it up on arXiv soon, but if anyone wants a copy please send me an email individually. Best, Dave From: statnet_help on behalf of Martina Morris Date: Friday, December 8, 2023 at 3:47?PM To: James Moody Cc: statnet_help@u.washington.edu , Schweinberger, Michael Subject: Re: [statnet_help] fragmented bipartite network... This is a great conversation; many thanks to the contributors. As I read through the proposed stats, though, I keep stumbling on the bipartite bit: how would some of these translate into bip net terms? I appreciate Jim's effort to bring this back to practical advice. So, some really basic thoughts here. There are two general types of blocks: those based on exogenous attributes, and those based on endogenous processes. I think the reason we're circling around the idea of blocks is that these depictions tend to capture the clustering observed in real world networks, and that blocking can help explain why dyad-dependent effects operate locally, rather than globally across a network. The exogenous type of block is captured by nodemix and nodematch type terms in ergm (which have a number of different specifications). In the bip net context these terms become more complicated as they no longer represent the crosstabulation of pairwise nodal attributes, but instead a crosstab of the terminal node attributes of a 2-mode triad. What's interesting about the bip net version of these terms is that this 2-path configuration is also a building block of equivalence. More on this below. The endogenous type of block is captured as latent block structures in hergms (for the ergm framework, other frameworks are out there). HERGMs are an interesting approach to identifying observed or latent neighborhoods of dependence (https://www.jstatsoft.org/article/view/v085i01), but I don't know if the package (or the models) can handle bipartite nets. I've added Michael Schweinberger to this email in case he would like to comment. Back to the exogenous blocking then. Family name could be a powerful blocking effect (e.g. Jim's example of Tata), showing up in this bip net as org board memberships shared by people with the same family name. Ignoring the modes, these 2paths would be Nullwise (or non-edgewise) Shared Partner (NSP) statistics. If two people shared all of their org memberships, they are structurally equivalent (whether they share an exogenous attribute or not) -- and more generally, the more NSPs, the higher the equivalence. And if the nodal name attribute is not driving these 2 paths, these high value NSPs are indicators of latent structure. The 2-paths can also be used to examine the org equivalence pattern in the same way. And my intuition would be that, conditioned on density, NSP distributions with higher means or longer tails would lead to fragmentation in the network. So, that makes me think perhaps the place to start is with EDA -- look at the NSP distributions, for both persons and orgs. Compare these to the expected distributions under a simple null random graph. If the distributions differ significantly, then start to look for exogenous effects that help to explain the deviation from the null (using the bip homophily terms with some more attributes on the nodes of both modes). And look into whether endogenously defined blocks (a la HERGM) can be used for bip nets. For me, the ideal would be to identify the latent blocks, and then explain almost all of that blocking in terms of exogenous/observed attributes. The blocks capture the structure. The explicit exogenous effects "explain" it. best, mm On Fri, Dec 8, 2023 at 6:28?AM James Moody > wrote: Fun discussion, thanks for sharing, always learn something in these sorts of posts. As to this this application per se; a couple of pragmatic (i.e. may not be elegant!) ideas: - theory should be able to inform some unlikely mixing that one could specify using a mixingmatrix term or two, no? So family, private/public, industry, etc. - For many business group applications, the actual family name is embedded in many of the subsidiaries (Tata group, tata inc, tata industries, etc.) so a name-similarity score could help (if you have nodenames) - The interlock limit will be size of the boards. While its possible to change the size of each board in a company, its not trivial, and I think you can justifiably take that as exogenous in the time-frame you have. I?m betting most of your small components are single family companies without external board memberships. Those create small stars in the bipartiate network (cliques in the projection). So that would imply: a) a hard-constraint on target degree. You could just fix that as a constraint. Again, not elegant (Carter?s cutting at joints and all), but likely true. b) a size mixing logic. Family-only/small-board cliques are isolated, leaving big-with-big, so there?s effectively a two-mode degree assortativity here. If you can?t induce this by an attribute (family name/ownership), then use assortativity on degree. - Cheating a little, but you could make component membership at attribute and hard-code mixing within/between. That means you can?t model what drives membership in the largest components vs. the small fractions, but, again, this is such a weird case (from a graph expectation sense), as anything that had even a little random noise in it would link across those small components, so the restriction here is almost certainly a legal/possibility restriction that should be treated as exogenous. - that?s, of course, just the crudest version of Daniel?s idea ? find a structural pattern that implies high/low probability of mixing across modes and hard-code it. I.e. do some old-fashioned inductive modeling of your network before the ERGM to generate classes of cases based on your best effort to induce the (to you) invisible restrictions patterning the ties, then add those back into the model as appropriate node/edge attributes. PTs Jim From: statnet_help > On Behalf Of Carter T. Butts Sent: Friday, December 8, 2023 4:53 AM To: statnet_help@u.washington.edu Subject: Re: [statnet_help] fragmented bipartite network... Hi, Daniel - Most of the cases to which I believe you are referring deal with differential mixing; the "blocks" here are what are sometimes called "density" blocks, which are quantitative relaxations of the complete/null blocks. I don't think anyone doubts that differential mixing exists, but that is very far from e.g. nontrivial global automorphism orbits or the like. Indeed, John Boyd had a running bet for some years, in which he offered to pay a sum of money (I forget how much) to anyone who could show a statistically significant regular equivalence pattern (above and beyond SE - he also had some other boundary conditions that ruled out "easy" cases). My vague recollection was that Steve Borgatti claimed to have one, and they then haggled over John's way of calculating "significance," but my memory on the subject is hazy and doubtless untrustworthy; I never did buy John's extreme conjecture, but it is true that he was not exactly overwhelmed with claimants. At any rate, models for differential mixing with discrete group structure are well-trod. As far as other kinds of generalized blocks (moving away from complete/null blocks), you can fit models with strict versions of e.g. regular, row/column dominant, and row/column functional blocks with clever use of constraints (in ergm, the bd() constraint term). The most obvious path to soft versions of those block types is to create statistics that count violations of the block pattern. Some can be implemented using the degrange() term, together with appropriate use of the optional attribute arguments. (Obviously, these are all "confirmatory" models, in the sense that one has to specify the block structure one wants to impose/parameterize. But that is not without its virtues.) Vis a vis dependence, I'm not sure that it is very helpful to think in terms of "violating assumptions." It is probably more useful to think of H-C and friends as giving you a "recipe" for the statistics you need to implement particular kinds of dependence conditions (should you want to do so). So, e.g., if you want edges to depend on each other when they share endpoints, then you will want (in the unvalued case) indicators for each edge variable, and indicators for each mutual dyad. If you also want the corresponding effects to be homogeneous, then this reduces to the edge count and the count of mutuals. Adding e.g. a 2-outstar term to a model with edges and mutuals is not violating any particular assumption imposed by the latter - it's just that this new model will now belong to a different (and broader) dependence class than the original one. (It will, in particular, have a form of Markov graph dependence.) Nothing says that your model has to belong to any particular dependence class - unless you want to impose such a condition. Of course, if you do want to restrict your dependence to a particular class, then you will indeed need to ensure that your statistics are a subset of those admitted by that class (which, for H-C, can be determined from the cliques of the conditional dependence graph). In my experience, this is rarely a useful way to proceed; however, it sometimes can be handy to know the type of dependence class to which your terms belong. Likewise, it can sometimes be handy to start by positing a form of dependence that makes sense in a specific situation, and then deriving the statistics that result. Pip, in particular, has done a great deal to elucidate these sorts of connections. As far as long-range dependence, there's again nothing ruling it out. (Pip and Tom, IIRC, have a very nice typology working out statistics for dependence classes at different distances.) For instance, k-cycles can be long-range, for large k. The various component and bridging statistics can be arbitrarily long-range. The statistics that arise from density and dyad census mixtures do them one better by being completely global (i.e., they create conditional dependence between edge variables irrespective of whether there is even a path of any length between their endpoints). All of these lead to well-defined models - those models just happen not to belong e.g. to the Markov graphs (or the social circuit graphs, the Bernoulli graphs, the u|man family, etc.). If there is a reason that you need your model to belong to such a family, then you would not want to use terms that are not within the class specifying that family. But otherwise, such restrictions are arbitrary, and may get in the way of specifying important mechanisms. Hope that helps, -Carter On 12/7/23 11:02 PM, Gotthardt, Daniel wrote: Hello Carter, i agree that stricter types oft equivalence are very rare and I would personally also look at either generalized blockmodeling or actually just measures of structural or positional similarity - but indeed not only local ones (which are already included in ergm of course). I did mention them here because most results of the relevance of more global equivalence structures I know have been found in especially kinship research and organisational science (Krackhardt & Porter 1986 and e.g. in insitutuional fields DiMaggio 1996 and Alsaas & Taamneh 2019). There has also been some recent research in foreign trade and political conflicts that indicate that block structures might matter (Guler et al. 2002, Zhou & Park 2012, Olivella et al. 2022). I am curious though which tools you are thinking about for implementing aspects oft generalized block structures? Regarding hammersley-clifford I mostly wanted to be careful here, but I did think that H-C and extensions like social circuit dependency (which allows partial depensence) did matter to ensure some (conditional) independence assumption with a few parameters (one for each clique of the dependence graph) in ergms (see e.g. Koskinen & Daraganova 2012 and Block er al. 2019). I thought dependencies (far) beyond the local neigborhood might violate these properties. This is probably beyond Harald's concerns but I would be happy if you could indicate any literature to alleviate my misunderstanding. Best Regards Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: statnet_help im Auftrag von Carter T. Butts Gesendet: Freitag, 8. Dezember 2023 07:06:46 An: statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Local automorphism orbits and their associations with covariates can be modeled using graphlet statistics; see e.g. ergm.graphlets. Nontrivial global automorphisms are extremely rare in typical social networks, so such terms would be unlikely to be useful - what one might call the "strong algebraic paradigm" of network analysis (the idea that we could explain most social network structure in terms of small numbers of roles, as defined through algebraic equivalences) was a very compelling idea that didn't really work out, and I don't think many folks are pushing in that direction right now. (See also compositional factorization, as famously illustrated by the semigroup on the cover of Wasserman and Faust (1994). Beautiful idea with some lovely technical results, but one with few if any real-world success stories. Sometimes, things just don't work out.) I think there could be some potential uses for terms for adherence to (confirmatory) generalized blockmodel structure (in the Doreian/Ferligoj/Batagelj tradition), though some of this can already be emulated using existing tools; there has also been a relative dearth of empirical cases in which complex block types have been shown to be important for capturing network structure. If such cases were to become more often encountered, this would naturally motivate more work to model them. With respect to your second comment, I am not sure what you mean by "violating" Hammersley-Clifford. H-C provides one way of establishing an equivalence between sets of network statistics and associated dependence conditions; Pip Pattison, Gary Robbins, and others have obtained various refinements to the original result (allowing for more subtle conditions to be treated). H-C and friends simply say (effectively) that certain classes of statistics implement certain kinds of dependence. These are important results for constructing and interpreting statistics, but they are not rules that can be violated. Hope that clarifies things, -Carter On 12/7/23 8:52 PM, Gotthardt, Daniel wrote: Dear Harald, after Martinas very insightful message and considering that you have kinship and business ties but not so many node covariates, I am wondering if you need or should think of structural equivalance as a driving factor. With White and others there is a strong tradition of focussing on this for kinship networks and DiMaggio and Burt have studied the importance oft business roles and structural position. In your case that probably means non-local forms of equivalence (automorphic, role, etc) that might matter directly in the network behavior or could represent unmeasured node attributes. Feature and embedding based measures are more scalable and now allow to measure those concepts better in larger networks. To the best of my knowledge this is not considered offen in generative network models and i don't think that we can include those less-localized mechanisms directly (yet). Plesae let me know if this is a direction that makes sense for you from a theoretical point of view and also something that could be identified in your data. I am currently working on this in the context oft actor-oriented models but am interested in the potential of ergms in this regard as well. At least as exogenous covariates this might be possible but otherwise we might violate conditional independence (Hammersley-Clifford theorem). I am curious to hear about the thoughts of experienced ergm modelers on this, though. Best Regards, Daniel -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de ________________________________ Von: Martina Morris Gesendet: Donnerstag, 7. Dezember 2023 23:45:59 An: Harald Waxenecker Cc: Gotthardt, Daniel; statnet_help@u.washington.edu Betreff: Re: [statnet_help] fragmented bipartite network... Hi Harald, You do have a complicated analysis here, and I'm a bit under-equipped to help you Dx what is going on, as I don't have much experience with either bipartite or multi-level nets (let alone both together!). What I can say, though, is that factor and covariate effects on the nodes are, in the non-multilevel context, one of the most important brakes on the feedback effects caused by dyad-dependent terms, making them more well-behaved and more likely to produce the kinds of networks we actually observe (caveat: sometimes those dependent effects are needed, see Carter's work on amyloid fibrils). In this case, it seems like you don't have many attributes to work with -- indeed, only on one of the modes. For gender, I would fit as a factor btw, not a quantitative covariate, tho if there are only 2 levels this will not have much impact. But when I think about the goals of board composition in non-profits (the closest I get to your world), it's clear that gender is not the only attribute that influences board member invitations -- and I would expect the same would be true here. You might try adding family name as a bxnodefactor (will pick up both family size and family activity level differentials), or sociality for either (or both) modes (to condition on the degree of each node). Your additional terms can then be interpreted as effects operating beyond these differences in degree. Degree distributions definitely influence component size distributions, up to a point, so if your model is not getting these right, you can start there. Thinking about the orgs, it seems there must be org attributes that influence the size and composition of the board. Org size, sector, geographic location, age, specialization, etc. -- I can imagine all of these would influence board memberships. Properties these nodes show in the other nets you have might be able to be represented on the cheap here as nodal attributes in this network. If these effects are at work -- and if you're not including them in the model, it is a form of mis-specification that compromises all of the other model estimates. Then there's homophily, which works differently in bip nets -- for one, it's a dyad-dependent term. But it's also more complicated to think about. Perhaps families might choose to specialize in an org sector, or maybe the opposite, they aim to integrate across sectors. Orgs might want diversity (on some measure) for members, which would show up as anti-homophily in bip two-paths. Again though, this would require more measured attributes for both orgs and persons. Adding model terms like components is different. In my modeling world, we want our (parsimonious) models to represent the mechanistic effects that may actually generate the ties in the network. For us, component size distributions are an *output* of a network formation process, not the generating mechanism (people aren't creating ties with the explicit intent of structuring the network component size distributions, with one key exception, and that we do model). We instead use the component size distribution as a goodness-of-fit indicator -- to test whether the mechanistic terms we included in our model reproduce these higher order excluded network stats. But your context may be different. When an org board is formed, if there is an explicit strategy to create specific component structures in the overall network then those intentions should be included as model terms. I can imagine that bridging structural holes might be one of those strategies. But again, not my area of expertise. I'm not sure how much any of this helps your specific issues. But when models don't fit the data properly, it's worth thinking about specification from first principles. So I hope this helps. best, Martina On Mon, Dec 4, 2023 at 12:28?AM Harald Waxenecker > wrote: Dear Tom, Martina, Carter and Daniel Thank you for your supportive answers. First, I will try to address some of your questions. The dependent network is a bipartite business network (6902 persons x 5178 companies), based exclusively on interlocking directorates. This dependent bipartite network represents the business ties of elite members in their home country. We include two covariates for the first node set (persons): traditional surname and gender. Isolates in this network represent elite members without any business ties. We belief that isolated nodes are meaningful in this network; e.g., women are often constrained to ?reproduction? rather than participating in ?production? (businesses). However, in different network layers they contribute to elite cohesion. Regarding these different layers: we have six more networks. The first is a one-mode kinship network (6902x6902), and the others are bipartite networks (based on interlocks), where persons form the first node set and entities the second. Hence, all matrices share a consistent number of rows (n = 6902), while the number of columns varies according to the number of entities in each network layer: offshore companies in Panama (n = 1537), business associations (n = 128), non-profit organizations (n = 236), political parties (n = 55), and public entities (n = 431). We employ ?bipartite homophily terms?, as proposed by Metz et al. (2018) https://doi.org/10.1017/S0143814X18000181, to test whether a common property (?homophily?) of the nodes in the first node set, such as a shared attribute (gender, traditional surname), a direct tie (kinship relation), or a mutual membership in other bipartite layers (offshore companies, business associations, etc.) contribute to the probability of two individuals forming ties with the same company in the dependent network. Regarding the modeling process, it?s true that the model we shared relies only on dyad-dependent terms. We always ?come back? to this model specification because all our attempts, which certainly were also based primarily on dyad-dependent terms, did not produce better results. We explored various options, including nodematch to control for component membership to split the network into smaller fragments. Then we incorporated component membership of the nodes as constraint to induce network fragmentation. While this partially improved network fragmentation, problems with goodness-of-fit persisted. Additionally, we encountered some computational limitations while running these options. Now, we have incorporated several of your recommendations, introducing dyad-independent terms and utilizing components() from the ergm.components package. Please find the new outcomes (model 0) attached. We've also attached summary files and component distribution for a comparative analysis between the observed network and the simulated network. We also tried to include the terms compsizesum() and dimers() into the model; however, we observe degeneracy issues. In addition, we still could not get results with bridges(), because it seems to be very time consuming and/or needs much computational capacity. I think this bridges-term relates somehow to your question @Martina about cross-group ties in the simulated data. Or maybe I am wrong. Please, could you explain that in more detail? Thanks. Thank you again for your support. Looking very forward to read your thoughts and advice. Kind regards, Harald El 1/12/23, 21:53, "[NOMBRE]" > escribi?: Hello Harald, if I understand you correctly you have a within-mode network as well as a bipartite network. James Hollway et al. (2017) has described an approach to handle these kinds of combined networks as multilevel social spaces with stochastic actor-oriented models: https://www.cambridge.org/core/journals/network-science/article/abs/multilevel-social-spaces-the-network-dynamics-of-organizational-fields/602BB810A44497EBDE2A111A6C2771A3 - There are also some tricks to transform these types of networks into an extended multimodal network matrix, exemplified e.g. in Knoke et al. (2021): https://www.cambridge.org/core/books/abs/multimodal-political-networks/agency-influence-power/57CB185C6E9429B34A9DE181C37EADF3 I personally don't know of any ergm model that can handle this kind of co-evolution of one-mode and two-mode networks but some kind of multilevel ergms (see Wang et al. (2013) https://www.sciencedirect.com/science/article/abs/pii/S0378873313000051) might be the way to go: - I'm sure others here know more about the capabilities of ergm.multi though. If these kinship structures explain the fragmentation of the bipartite network, you might need to include them either directly with the approaches above or construct some corresponding dyadic or monadic covariates to represent the kinship structure in your single level network. Best Regards, Daniel Am 01.12.2023 um 02:13 schrieb Martina Morris: > > Hi Harald, > > I'm looking for some clarification here, which I think Tom Kraft might > also have wondered about. > > You say: >> >> Our research focuses on tie formation and elite cohesion, specifically >> examining interlocking directorates and kinship relations. The >> dependent bipartite business network comprises 6,902 individuals and >> 5,178 companies, exhibiting sparsity (density = 0.00012) and >> fragmentation with 4,455 components, including 3,850 isolates in the >> first mode (persons) >> > For a bipartite network ties are allowed only between modes (persons, > companies), not within. It's clear how interlocking directorates would > meet that criteria. But kinship relations would be among persons, so > within-mode, not between, and this would not be a bipartite network. > > Is the model you've sent us for the interlocking directorships only? > And by isolates in the person mode, do you mean persons who are not > affiliated with any of the companies? If so, then it's a bit odd to > include them in the bipartite network. > > I'm wondering if this problem is better posed as a multilevel network > (not my area of expertise). > > thanks, > Martina > > > On Thu, Nov 30, 2023 at 4:33?PM Carter T. Butts > >> wrote: > > __ > > Hi, Harald - > > Coexistence of large complex components does not generally occur > unless something drives the fragmentation, and this is what your > models are telling you: the terms you are currently using do not > include the forces that are sufficient to reproduce your component > size distribution. That means that you need to think about why your > network is split into fragments, and include terms that capture the > relevant social forces. Thinking about likely mechanisms is step > zero, so do that before anything else! Guided by your substantive > knowledge of what is likely going on, you will next (as others have > said) want to look at covariate effects relating to differential > mixing, since those are your most obvious and most important sources > of heterogeneity. If you find that there is still more > fragmentation that can be explained by other means, you may need to > consider model terms relating directly to component count or size. > These are still somewhat experimental, and are currently sequestered > in an add-on package called ergm.components > (https://github.com/statnet/ergm.components > >). However, this package can be installed from github (see the github page), and the terms will work automagically with ergm() and friends once the package is loaded. Depending on your situation, you may need or want to examine the components() or compsizesum() terms, both of which are documented within the package. > > Hope that helps, > > -Carter > > On 11/30/23 9:58 AM, Harald Waxenecker wrote: >> >> Dear ?statnet community?,____ >> >> __ __ >> >> Our research focuses on tie formation and elite cohesion, >> specifically examining interlocking directorates and kinship >> relations. The dependent bipartite business network comprises >> 6,902 individuals and 5,178 companies, exhibiting sparsity >> (density = 0.00012) and fragmentation with 4,455 components, >> including 3,850 isolates in the first mode (persons). The attached >> documents contain descriptives and the component size distribution >> from the observed network.____ >> >> ____ >> >> The fragmented structure is important, as other network layers, >> like kinship relations, are expected to contribute to the cohesion >> of this business network. We apply ERGM to model these processes, >> but we struggle to capture the fragmented structure of the >> observed network. The component size distribution of the simulated >> network differs significantly. In addition, the goodness-of-fit >> (GOF) for k-stars (in both modes) and geodesic distances (Inf) >> shows significant results. All these results are also attached.____ >> >> ____ >> >> We've explored various options, including constraints, MCMC >> propositions, and simulated annealing, but haven't achieved >> success. Please, we would like to ask for your help to improve our >> model. Thank you!____ >> >> __ __ >> >> Kind regards,____ >> >> Harald____ >> >> __ __ >> >> __ __ >> >> __ __ >> >> --- ____ >> >> *Harald Waxenecker >> >> *____ >> >> *Masaryk University | Faculty of social studies* >> Department of Environment Studies >> A: Jostova 10 | 602 00 Brno | Czech Republic >> E: waxenecker@fss.muni.cz >____ >> >> __ __ >> >> >> _______________________________________________ >> statnet_help mailing list >> statnet_help@u.washington.edu > >> https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!KK5UcPVRvb25ILHn7wJt4TEsP-Ic39L133WdzimKJv-378bLqah-hO8Gm9Yd_qoWgV_tbzbT6swweifmS5mRRQ$ > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help > > > > _______________________________________________ > statnet_help mailing list > statnet_help@u.washington.edu > http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -- Daniel Gotthardt, M.A. Wissenschaftlicher Mitarbeiter / Research Associate Universit?t Hamburg Fakult?t f?r Wirtschafts- und Sozialwissenschaften / Faculty of Business, Economics and Social Sciences Fachbereich Sozialwissenschaften / Department of Social Sciences Soziologie, insb. Digitale Sozialwissenschaft / Sociology, esp. Digital Social Science Max-Brauer-Allee 60 22765 Hamburg www.uni-hamburg.de _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu https://urldefense.com/v3/__http://mailman13.u.washington.edu/mailman/listinfo/statnet_help__;!!CzAuKJ42GuquVTTmVmPViYEvSg!Jy0dmFtPSz9FGZILsxIzHWpAcAK5wDvLWuQ2s4hKJdX0uaJX7imnKxe9w1W52yrNrJRKiI-YzcF0M4kcXbfma0JgQ7mPF8AH$ _______________________________________________ statnet_help mailing list statnet_help@u.washington.edu http://mailman13.u.washington.edu/mailman/listinfo/statnet_help -------------- next part -------------- An HTML attachment was scrubbed... URL: