# Workshop on ERGM at ION6

** What?**

This is a one-day workshop on ERGMs (p* models) at ION6. The morning session (10:30-13:00) will provide an intuitive, yet thorough understanding of the logic behind ERGMs (p* models). The afternoon session (14:00-17:00) focuses on running models and the interpretation of results/output. The approach taken is that in order to understand – and to be able to interpret – the results of an ERGM, one needs some basic knowledge of the logic behind ERGMs. Once this logic becomes clear, the meaning of specific local configurations included in an ERGM become much easier to grasp.

** Why?**

In many cases we are interested in knowing whether our observed network is the result of specific forces or processes (such as a tendency for popular people to attract more ties, a tendency towards closure in triads, or homophily). These forces can be captured by specific local properties in the network (e.g., high centralization, many closed triads, or many homophilous ties).

However, when can we say (with some confidence) that we have more closed triads or more homophilous ties in our network than we could expect by chance? And can we be sure that closure is not simply the result of homophily? Or vice versa?

Hence, the aim of these statistical tests is to answer questions such as:

– Is my network more centralized than chance… ?

– Does my network exhibit a force towards triadic closure… ?

– Is there a tendency towards homophily in my network… ?

… **given** the other forces in the network (i.e. taking these into account, or controlling for them).

** How?**

In order to fully understand the logic underlying Exponential Random Graph Models, we will spend time in the morning session looking at: 1) the (correct, but impossible) maximum likelihood approach, 2) the original pseudolikelihood (p*) approximation, and 3) the MCMC approach to approximate the maximum likelihood. Although superseded, the pseudolikelihood approach makes it far easier to understand the meaning of specific local configurations and enables us to understand the MCMC estimation procedure (which are used in the current programs) from an intuitive perspective. A good understanding of the MCMC procedure is crucial in order to get models to convergence.

While the morning session focuses on acquiring the principles of ERGM, the afternoon session will be applied and hands-on. We will spend time on: 1) running ERGM with PNet, and 2) especially on interpreting the results of ERGMs.

Link to ION6: https://sites.google.com/site/ion6conference/home

Key references:

•Lusher, D., Koskinen, J., Robins, G. (eds.) 2013. Exponential Random Graph Models for Social Networks. Structural Analysis in the Social Sciences. New York: Cambridge University Press.

**Please note: **MPNet.exe is build for Windows. However, if you have a Mac notebook you might want to ask your IT support how to run Windows (.exe) files on your Mac (if you are unfamiliar with this).