Purely Structural Effects
Initially, we fitted a model that included six structural effects: arc, reciprocity, popularity spread and activity spread, path closure, and multiple connectivity. Although the model converged, GOF statistics suggested that our model failed to capture the skewness of the in- and out-degree distributions: it did not account for the fact that some actors were very central in the leadership network (had high in-degrees), and that some actors did not choose any other actors as leaders, whereas others chose many (high variance in the out-degree).
Following Robins, Pattison, and Wang (2009), we fitted a second model, which included the above six parameters together with sink, source, and isolates effects. The model converged, and GOF statistics suggested that the inclusion of these three parameters successfully controlled for the skewed degree distributions.
We use three basic attribute effect types: sender, receiver, and heterophily (Chapter 8). These three effects were fit for each of the four continuous attributes that we collected: general cognitive ability, performance on physical tasks, nonphysical performance (military-related, specific cognitive abilities), and optimism. The full model therefore includes twenty- one parameters: nine parameters for purely structural effects, and twelve parameters for actor-relation effects. We are especially interested in the receiver effects because they represent emergent leadership.