The analyses use the social selection models of Chapter 8. These models assume that attributes are fixed and that social ties may vary (i.e., given the actor attributes, where do we expect the ties to be?).
ERGM estimation was done using the PNet program (Wang, Robins, & Pattison, 2009). The estimation procedure successfully converged (see Chapter 12) for all parameters presented for all models in this chapter. (The practice of only presenting results for converging estimations is followed in all subsequent chapters.)
Goodness of Fit
Goodness of fit (GOF) statistics were calculated for the second (final) model for both schoolboys and footballers networks. The GOF statistics (see Chapter 12) were less than 2.0 (in absolute value) for features that were not explicitly modeled, suggesting that the models manage to reproduce and fit the data. However, the t-statistic for the skewness of the out- degree distribution (for the schoolboys) was larger (2.71). Such a result is not unusual and indicates that there was some difficulty in capturing the out-degree distribution in that there are some highly active actors who are not completely represented by the model. Attempts to improve the skew on the out-degree distribution by including additional parameters were unsuccessful; however, in all other respects, we may consider the fit of the models excellent.