Print

Print


***** To join INSNA, visit http://www.insna.org *****

Dear all,

First, thank you all for the comments (by Jacob, Filip, Tom, Carter, Dave & Johan), they have been quite helpful both in terms of better understanding the analytic logic(s) of ERGM’s and how I could (maybe alternatively) go about my research.

The distinction between *holistic* and *element-wise* conceptions of an ERGM that Carter elaborated has been most informative (I’m not familiar with classical mechanics, but in my past life I did some ecosystem energy-flow modeling, so I do relate to the notion of a series of equations stochastically describing the potential function of a system).á Apart from section 3 of the paper on ERGM terms Dave suggested I don’t believe I had come across any explicit statement of why the usual dependent/independent variable interpretation can be misleading.á I believe a published elaboration of these notions would be a very helpful and welcome addition to the ERGM literature, which to this point has been mostly of a technical nature –for understandable reasons.

Based on the above, I think I now have a better idea of when it would be advisable to employ an ERGM on my own research.á There are a couple of my research interests that would indeed benefit from an ERGM since I do want to model the whole network.á The particular question I had in mind when I wrote my post to the forum, however, I now believe would be better addressed by some form of a network autocorrelation model (I’ll have to give it some thought –some of the node attributes in which I’m interest are dichotomous, some continuous. And being an anthropologist, I do have ethnographic information ;-) ).

Again, thanks to all for the help

 

Jorge.

 

 

From: Social Networks Discussion Forum [mailto:[log in to unmask]] On Behalf Of Johan Koskinen
Sent: Tuesday, May 03, 2011 4:03 AM
To: [log in to unmask]
Subject: Re: [SOCNET] ergm question

 

***** To join INSNA, visit http://www.insna.org *****

Jorge,

To further add to previous comments, it may be good to keep in mind two things. Firstly, the method should mirror the substantive issue. If you are interested in explaining nodal attributes you should use a model for nodal attributes and not a model, like ERGM, for tie-variables. This is much in the same way that if you are interested in income differences for men and women you would regress income on gender and not the other way around (i.e. regressing gender on income). Secondly, if you want to explain attributes, you model attributes of, say, n individuals. If you want to explain tie-variables, you are modelling (in the undirected case) n times n-1 tie-variables. Consequently, if you draw inference about your covariates based on an ERG model for tie-variables, in a sense you are using your actor attributes n-1 times. You may for example have 25 homophilous tied pairs of actors for a binary attribute (for or against) but only 10 individuals in total. An ERGM would treat the homophilous (and heterophilous, naturally) ties as observations.

In addition to the techniques mentioned previously (QAP and network effects/autocorrelation models), an attribute model analogous to the ERGM for tie-variables is derived in Robins, G., Pattison, P., & Elliot, P. (2001). Network models for social influence processes. Psychometrika, 66 , 161–190. (A related model is described and elaborated in a chapter of a book on ERGM that is forthcoming)

j.

 

The email [log in to unmask] WILL EXPIRE on 30 April 2011

************************************************************
Social Statistics Discipline Area
Humanities Bridgeford Street
University of Manchester
MANCHESTER M13 9PL
+44 (0)161 30 66 953 (ManU, Kantorovich Building)
johan.koskinen[at]manchester.ac.uk


From: Social Networks Discussion Forum [[log in to unmask]] on behalf of Garry Robins [[log in to unmask]]
Sent: 03 May 2011 04:57
To: [log in to unmask]
Subject: [SOCNET] FW: ergm question

***** To join INSNA, visit http://www.insna.org *****

Jorge,

 

There is nothing wrong with what you are doing. But just as with inferring causality from a standard regression, you can’t simply infer that the “pattern of social interactions influences the type of opinion individuals have”.  What you can say, though, is that there is an association between the attribute (opinions) and network structure. This is all we can do with cross-sectional data. It may or may not be enough for your research purposes. Of course, if you are specifically investigating influence hypotheses, then it makes sense to use an influence model rather than a selection model – and Tom gives some good suggestions. I’m opposed to using simple regressions here, Tom is right to point to the issue of dependency.

 

BTW there is an ERGM-based influence-type model in the pnet suite of programs, akin to a logistic regression regression but taking into account the network dependencies. It needs a binary attribute, however.

 

Good luck with the research

 

Garry

 

 

Dr Garry Robins

Associate Professor and Reader
Psychological Sciences
University of Melbourne
Victoria 3010
Australia

Tel: 61 3 8344 4454
Fax: 61 3 9347 6618
Web: http://www.psych.unimelb.edu.au/people/staff/RobinsG.html
Melnet website: http://www.sna.unimelb.edu.au/



 

 


From: Social Networks Discussion Forum [mailto:[log in to unmask]] On Behalf Of Thomas Valente
Sent: Tuesday, 3 May 2011 4:32 AM
To: [log in to unmask]
Subject: Re: ergm question

 

All

I think Filip’s suggestions are quite helpful and at the same time Jorge’s frustrations understandable.  The problem is some people will be quite concerned about network dependencies and the challenges with specifying network effects whereas others will be content to focus on behaviors and policy interpretations of these types of influences.  (Somewhat like the balance between Type I and Type II error.)  And the researcher wishes to “do the right thing.”  We/I have opted to do both, node level regressions using somewhat standard multi-level (or hierarchical) regression techniques and, when possible, build models using STATNET and SIENA.  So far the results are fairly consistent with different approaches yielding different but not contradictory insights.  An example is our paper on the adoption/diffusion of the FCTC:

Wipfli, H., Fujimoto, K., & Valente, T. W. (2010).  Global tobacco control diffusion: The case of the framework convention on tobacco control. American Journal of Public Health, 100, 1260-1266.

So there is no right answer and no single approach that covers all situations.  The researcher will need to pursue at least 2 and sometimes multiple modeling approaches to be confident the findings are consistent across multiple specifications.

-Tom

 

Thomas W. Valente, PhD

Current:  ╔cole des hautes Útudes en santÚ publique (Rennes/Paris, France)

Usual:

Director, Master of Public Health Program      http://www.usc.edu/medicine/mph/

Professor, Department of Preventive Medicine

Keck School of Medicine

University of Southern California

1000 S. Fremont Ave., #8

Building A Room 5133                     

Alhambra CA 91803              

phone: (626) 457-4139; cell: (626) 429-4123

email: [log in to unmask]

 

Social Networks and Health: Models, Methods, and Applications:

       http://www.oup.com/us (promo code: 28569)

Evaluating Health Promotion Programs: www.oup-usa.org/

Network Models of the Diffusion of Innovations: www.hamptonpress.com

My personal webpage: http://www-hsc.usc.edu/~tvalente/

The Empirical Networks Project: http://ipr1.hsc.usc.edu/networks/

You Tube video on Diffusion of Innovations: http://www.youtube.com/watch?v=ZG9dAIBd4xQ

 

From: Social Networks Discussion Forum [mailto:[log in to unmask]] On Behalf Of F.Agneessens
Sent: Monday, May 02, 2011 7:58 PM
To: [log in to unmask]
Subject: Re: ergm question

 

***** To join INSNA, visit http://www.insna.org ***** Dear Jorge,

If you are interested in explaining individual outcome, why not use a regression model with these opinions as dependent variables and network structure as independent?
This has been done quite extensively: e.g.,
- Brass, D.J. 1984. Being in the right place. A structural analysis of individual influence in an organization. Administrative Science Quarterly 29: 518-539.
- Sparrowe, R.T, Liden, R.C., Wayne, S.J., Kraimer, M.L. 2001. Social networks and the performance of individuals and groups. Academy of Management Journal: 316-325.

If you do not want to rely on the significance test of the 'normal' regression (which traditionally relies on a random sampling from a population), you could even make use of a QAP-like node level regression. For example in UCINET you can do this with the following command: Tools>Testing hypotheses>Node Level>Regression.

However, if you think a person's opinion is influenced by the opinions of his connections you can use an 'autocorrelation'-type of regression. In that case, see for example:
-Doreian, P. 1981. Estimating linear models with spatially distributed data. Sociological Methodology 12:359-388.
-Doreian, P., K. Teuter, C. Wang. 1984. Network autocorrelation models: Some Monte Carlo results. Sociological Methods & Research 13: 155-200.
-Marsden, P.V., Friedkin, N.E. 1993. Network studies of social influence. Sociological Methods & Research 22, 127-151.
-Leenders, R.Th.A.J. 2002. Modelling social influence through network autocorrelation: Constructing the weight matrix. Social Networks 24: 21-47.

Hope this helps,
Filip


On 02-05-11, Jorge M Rocha <[log in to unmask]> wrote:

***** To join INSNA, visit http://www.insna.org *****

Dear all,

I have a conceptual question that is probably somewhat na´ve regarding the analytic logic behind ERGM’s (I’ve been using statnet and its suite of tools).  If someone in the community could point me in the right direction literature wise, I would appreciate it.  Now let me try to state my question/doubt as briefly and clearly as I can:

From what I understand, the modeling framework of an ERGM treats the observed network as the dependent variable and the specified structural configurations and covariate information about nodes/edges as independent variables whose presence/absence increases the log-likelihood for the model.  In other words, what we want to achieve in these models is to gain a better understanding of the types of processes that might have gone into generating a network similar to the observed one; hypotheses are about this or that configuration or node/edge attribute and their effect on the pattern of social interactions represented in the network. If so, how would I go about testing hypotheses concerning the effects of social interactions on nodal attributes (I’m particularly thinking of different types of opinions individuals might tend to have based on their interactions)?  

What I’ve done so far is kind of arguing backwards (or at least that is how it feels to me –maybe because I’m coming from a traditional linear/logit regression background, and probably part of my problem is thinking in terms of DV and IV).  If the coefficients are significant for some node covariate of my interest in the ERGM estimation, then I’ve been interpreting this as evidence that it is not unreasonable to argue that the pattern of social interactions influences the type of opinion individuals have (I know that if I had longitudinal data this issue of reverse causality would not be so much of an issue, but so far all I have is cross-sectional data).  How wrong am I on this (the backward-arguing)?

Finally, part of my worry is that I’ll be sending the research to a consumer behavior/marketing journal where the familiarity with social network stuff –let alone ERGM’s—is limited.  Any suggestions on how to best explain these issues to a non-specialized audience?

In advance, thanks,

 

Jorge M Rocha

EGADE Business School

 

 

_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.


--

 

Dr. Filip Agneessens

Department of Sociology/ICS

University of Groningen

Grote Rozenstraat 31

9712 TG Groningen

The Netherlands

email: [log in to unmask]

http://filipagneessens.wordpress.com/

_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.

_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message. _____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.

_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.

_____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.