Sample size affects network measures in a species with high fission-fusion dynamics (#843)
Social structure can be defined using network metrics such as degree and strength. Recently, there have been warnings of the tendency for commonly used association indices (such as half-weight indices) to overestimate specific social preferences between dyads in situations with large individual variation in gregariousness. Most authors set a minimum number of observations per individual so that rarely seen animals do not have a disproportionate effect on results. However, the possible effect of individual sample size on number of associates and network measures based on pair-wise association indices has not been examined. To investigate this effect, we computed Pearson correlations for 153 marked eastern grey kangaroos Macropus giganteus for which we had a minimum of 10 or 40 observations. Eastern grey kangaroos live in fission-fusion societies with mean half-weight indices around 0.01 at high desity. Of the six measures of sociability that we investigated, all but the clustering coefficient were highly and positively correlated with sample size. When a minimum of 40 observations was used, the clustering coefficient also became highly related to sample size. The number of associates (degree), strength, centrality, reach and the clustering coefficient all approximately doubled with an increase from 50 to 100 observations, whereas affinity increased by 20%. Using these individual network measures could be problematic in further correlation analyses with phenotypic characteristics (e.g. sex, size, parasite load and personality) or reproductive success, unless controlling for observational sample size of individuals. We suggest that sample size effects are more likely to occur in large populations where association strengths are generally low and individuals have many weak associations. We further recommend that researchers verify whether sample size affects individual network measures in other species with strong fission-fusion dynamics.