*****  To join INSNA, visit  *****

I know of some partial solutions, but no satisfactory ones.
You can do several things:
(1) Calculate robust estimates, also called (I think), sandwich 
estimates or Huber-White.  These widen the confidence intervals reducing 
likelihood of type I error. (Although you never feel you've eliminated 
the problem.)
(2) Convert the data to a dyadic dataset (each case is the person and 
his/her contact) and use hierarchical linear model (of multi-level 
modeling) to control for clustering on the individual.  This has the 
advantage of allowing you to test relational hypothesis (such as an 
interaction based on dyadic characteristics).  It has the disadvantage 
of not allowing the calculation of network exposure values.
(3) Hope the statisticians (Handcock, Morris, Snijders, Robbins, 
Pattison) have something suitable in ERGM or StocNet.


Michael Haenlein wrote:

>*****  To join INSNA, visit  *****
>Dear colleagues,
>I have a social network consisting of several thousand actors. Besides
>knowing something about the social relationships between these actors, I
>also have information about a (ratio-scaled) variable for each of them. For
>example, assume that I know everyone's income or the amount of chocolate
>everyone consumes per month or anything in this spirit.
>Based on this data, I want to analyze whether the value of this variable for
>a given actor is influenced by the value of this variable among people that
>are friends of this actor. For example, is there a relationship between my
>income and the income of my friends? The type of question I'd like to answer
>is: Are people whose friends are primarily high income more likely to be
>high income themselves that people whose friends are low income?
>My first idea to do this type of analysis was to run a set of regressions.
>For every actor, I could for example run a regression which relates his/ her
>income to the average income of his/ her friends. If I have n actors in
>total, I would need to run n such regressions. However, the problem is that
>these regressions are likely to suffer from an endogeneity bias as the same
>variable can appear as a dependent variable in one regression and as an
>independent in another. Take the following example: Assume a network that
>consists of three actors (Ben, Adam and Chris) that are all related to each
>other (i.e. Ben, Adam and Chris form a 3-clique). The three regressions I
>would have to run would be:
>Income_Adam = Alpha + Beta [(Income_Ben + Income_Chris)/2]
>Income_Ben = Alpha + Beta [(Income_Adam + Income_Chris)/2]
>Income_Chris = Alpha + Beta [(Income_Adam + Income_Ben)/2]
>Adam's income, for example, is a dependent variable in Regression 1 and an
>independent one in Regressions 2 and 3. This endogeneity is likely to result
>in biased parameter estimates in my regressions.
>Does anyone know which approach I can use to overcome this bias?
>Are there any software tools and/ or articles that address this issue?
>I'm familiar with R, so if anyone could recommend an R Package to do this
>type of analysis, that would be great!
>Thanks very much for your help,
>Michael Haenlein
>Assistant Professor of Marketing
>ESCP-EAP European School of Management
>79, Avenue de la République | 75011 Paris | France
>SOCNET is a service of INSNA, the professional association for social
>network researchers ( To unsubscribe, send
>an email message to [log in to unmask] containing the line
>UNSUBSCRIBE SOCNET in the body of the message.

Evaluating Health Promotion Programs (Oxford U. Press):

My personal webpage:	

The Empirical Networks Project

Thomas W. Valente, PhD
Director, Master of Public Health Program
Department of  Preventive Medicine
School of Medicine
University of Southern California
1000 S. Fremont Ave.
Building A Room 5133
Alhambra CA 91803
phone: (626) 457-6678
fax: (626) 457-6699
email: [log in to unmask]

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