Figure 1.

Example of the mechanism by which datastream permutations may lead to false positives in linear regression. In the original network, there is variation in strength among individuals driven by differences in gregariousness (represented by node size in the social networks). Individuals are assigned a trait value (represented by colour in the social network) unrelated to their network position. By chance, there is a slight negative relationship between network strength and trait value in the observed network. After several permutations, there is a reduction in the variance in the strength of individuals in the permuted network, and thus the magnitude of the relationship is reduced. The bottom histogram shows the distribution of null coefficients after 10,000 permutations (black), and the coefficient from the original linear model (red).