Accurate confidence intervals from recapture experiments

Abstract

Confidence intervals for estimates of population size are difficult in practice because of the high skewness and bias of the estimator under realistic sampling conditions. In this paper I investigate construction of accurate confidence intervals for the population size using pivotals (i) based on the estimated number of unseen individuals as suggested by Chao(1989), (ii) the profile likelihood function, (iii) the approximate score function of Darroch(1958). Bias adjustments are available and produce more accurate one-sided intervals while skewness does not seem a major problem in practice. Although only the simplest sampling model is considered, the method should extend to models with time a..