Estimating dyad association probability under imperfect and heterogeneous detection: a Bayesian approach

In social studies, individual association indices estimate the proportion of time two individuals (i.e. a dyad) spend together. However, traditional association indices rely on the assumptions that individual detection probabilities (p) are either one (at least approximately) or independent of the association states. Based on marked individuals, we develop a model to estimate the probability a dyad is in associated state () accounting for detectability less than one and varying according to association states. Our model allows for both individual and dyad missing observation and can easily be extended to incorporate covariate information for modeling detectability and dyad association probability. Parameter estimates are obtained as posterior means via Monte Carlo Markov Chain. A simulation study showed that our model-based approach yield unbiased estimates, even for low and heterogeneous detection probabilities, while, in contrast, standard indices showed moderate to strong biases.