| Social dominance is an important feature of many groups of social animals. We aim to explore a new network-based method, Percolation and Conductance, for constructing dominance rankings. The Percolation and Conductance method leverages direct and indirect network pathways to clarify ambiguous dominance relationships caused by sparse or missing observations. Randomly sampling without replacement at 10% increments (i.e. 10%, 20%,...,90% ), a data set comprised of 8655 observed instances of aggression in a single group of captive rhesus macaques (Macaca mulatta) was used to address: 1) what effect using "partial networks" constructed from a random subset has on global and individual network measures, 2) how changes in data density affect measures of dominance certainty, and 3) how much variation in rank order to expect given an estimation of initial data density. We find the network measures follow similar patterns as described in Wey et al. (2008). Although variation in the outputs of Percolation and Conductance increases as data density decreases, dominance rankings are relatively robust to decreases in data density as similar rank orders are produced across data reductions of up to 50%. However, there is more variation in middle ranked individuals then in high and low ranked individuals. We also provide a linear model to allow users an estimation of the amount of variation in rank order given an initial data density in terms of hours per event sampled and aggressive events per individual. |
