***** To join INSNA, visit *****
Dear SOCNETters,

I think I am not alone in not fully appreciating why there is a need for special statistical models for networks (ERGMs, SIENA, etc.). I can accept that there is such a need, but I'd like to understand what dependencies between observations do to traditional models or obvious approaches. 

E.g., in a March 16th email, Elly Power had asked about using centrality as a covariate; there were helpful replies from Philip Leifeld, Tom Snijders, and Phillip Bonacich, but aside from there being more suitable alternatives, the effects of dependencies are still opaque to me. 

Does anybody know of literature or resources that explain this?

In what I've found so far, the network autocorrelation literature (and 'Galton's problem') from Dow and colleagues in the 80s helps a lot for understanding problems with regression on node attributes, as does Shalizi & Thomas (2011) for C&F's 'network effects' approach. Still, I'd like to see similar explorations for other obvious approaches including using centrality as a covariate, or doing a logistic regression on the presence/absence of individual edges (not pairs of edges as in p1), or perhaps doing a regression with network-level measures of ego networks. What would we be missing? How misleading could results be?

As a related point, does anybody know of introductions that gather up and compare all approaches to network statistics? E.g., I don't think I've seen any introduction that has even just ERGMs and SIENA side-by-side, let alone network autocorrelation with those two, even if just to say what the differences are in modeling goals and assumptions (or, for the former two, to note the range of Tom Snijders' contributions!). 

Thank you!

Partial list of what I've been searching through:

_____________________________________________________________________ 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.