An Improved Removal Method for Estimating Animal Abundance

Abstract

The problem of estimating animal populations from the numbers in successive catches has been addressed by many authors. Most have assumed asymptotic normality of maximum likelihood estimates to construct confidence intervals. This assumption can be shown to fail in many common practical situations. We construct intervals based on the profile likelihood ratio and show, by stimulation, that their true confidence is much closer to the nominal value than that of previous methods.