Consistency of dominance rank order: a comparison of David's Scores with I&SI and Bayesian methods in macaques

In nonhuman primate social groups, dominance ranks are usually assigned to individuals based on outcomes of dyadic agonistic encounters. Multiple approaches have been used, but currently there is no consensus. One approach, David's Scores (DS), offers dual advantages of yielding cardinal scores that may in turn be used to compute hierarchical steepness. Here we correlate rank orders yielded by DS with those yielded by both the traditionally used I&SI approach and the recently proposed parametric Bayesian approach. We use six datasets for female macaques (three despotic and three tolerant groups), and 90 artificially generated datasets modeling macaque groups. We also use the artificial datasets to determine the impact of three characteristics (group size, interaction frequency, and directional asymmetry of aggression) on the magnitude of correlation coefficients, and assess the relative utility of two indices used to compute DS: Dij versus Pij. DS-based rank orders were strongly positively correlated with those yielded by the other two approaches for five out of the six macaque datasets, and for the majority of artificial datasets. Magnitudes of correlation coefficients were unrelated to group size or interaction frequency, but increased with directional asymmetry, suggesting methodological inconsistencies were more likely when dyads had more frequent reversals in directions of aggression. Finally, rank orders calculated using the Dij and Pij indices were similarly consistent with orders from other methods. We conclude that DS offers consistent estimates of rank orders, except perhaps in groups with very low levels of aggression asymmetry. In such "tolerant" groups, we suggest that the relatively greater methodological variability in rank orders may reflect behavioral characteristics of tolerant groups rather than computational inconsistencies between methods. We hypothesize that this quality may be quantified using posterior probability scores of Bayesian rank orders and may also index macaque social styles.