Accurate inference for recapture experiments with behavioural response

Abstract

For capture-recapture experiments, confidence intervals for the population size N have been based on asymptotic normality of the maximum likelihood estimator. Such intervals are known to have poor coverage accuracy for practical sample sizes and may violate the lower bound on the population size implied by the data. We investigate five different methods of constructing confidence intervals, under the behavioural response model. We develop bias adjustments and examine interval coverage using simulation both with and without this adjustment.