A comparison of association indices

Indices used to measure the frequency of association between individuals in fission/fusion societies are frequently borrowed from ecological studies of species association without adequate justification. In this paper several such indices, under specific conditions likely to be encountered in field studies of animal behaviour, are examined. Each of three indices commonly found in the literature is shown through simulations to be accurate over only a narrow range of possible sampling biases. As an alternative approach, examples are given of the derivation of a maximum-likelihood estimator based on two simple models. Both models assume that a constant proportion of existing subgroups is located. In the first model all combinations of subgroups are equally likely to be located, while in the second model the probability of locating one subgroup is allowed to differ from that of the others. The maximum-likelihood estimators are shown to be less biased and to have lower variance than the other three indices under the assumptions of the models.