Figure 4.

Flowcharts of two approaches for regression analysis in animal social networks. In the first (A), a network is generated that attempts to adjust for confounding effects (through e.g. GAIs) which is then used to derive the response. In the second (B), the original network is used to derive the response variable, with confounds instead being incorporated as covariates in the inferential model. In both methods, inference is based on post-network permutations (such as MRQAP or node-label permutations).