R code for "Nepotism masks reciprocity in cooperation networks"

Nepotism and reciprocity are not mutually exclusive, because kin cooperation might be based on kinship cues, reciprocal help, or both. Observational evidence for each strategy can therefore be difficult to disentangle. We developed a resampling procedure for inferring power to detect evidence of nepotism and reciprocity. We first simulated datasets of reciprocal helping with varying degrees of nepotism and varying sample sizes of observations. We then assessed how nepotism and number of observations influence the probability of detecting reciprocity. We show that nepotism consistently masks evidence for reciprocity. Even when reciprocal help is a better predictor than kinship, kinship effects are often detected faster and appear larger, especially with fewer observations. This occurs because sampling fewer events does not affect kinship estimates, but it does decrease the accuracy of estimates of help given and received. We confirmed evidence for this power bias in three real datasets: food sharing between vampire bats, and allogrooming between mandrills and Japanese macaques. Our resampling procedure highlights the need to estimate the relative reliability of estimates of kinship and helping rates before comparing their relative importance. We provide R scripts to assess the stability of kinship and reciprocal help estimates in samples of cooperative behavior.