Anova based on Linear Mixed Models (LMMs) fitted to contributions in Stage 2, using ‘subject nested in group’ as a random effect.

The model estimates are given in the Supporting Information (S1 Table). The analysis includes a comparison of (i) Uninformed Assortment and Uninformed Random grouping to study the effect of assortment and (ii) a comparison of Informed Assortment and Uninformed Assortment to study the effect of information. The factors “Assortment” and “Information” compare treatments. ‘Round’ is the round number (1–15) to account for time trends in cooperation levels in Stage 2 (Fig 2) and ‘Cooperative tier’ is a factor reflecting individuals’ tier rank (1, 2, 3 or 4) based on cooperation in Stage 1. The significant interaction between ‘Assortment’ and ‘Cooperative tier’ results from larger differences between individuals with the highest and lowest cooperation tendencies in assortment compared to the random grouping treatment (i.e., assortment amplifies existing cooperation tendencies). Surprisingly, the effect of ‘Information’ is not significant. In each treatment, there were 80 subjects (five sessions with 16 subjects in each session).