On the Population Size Estimation from Dual-record System: Profile-Likelihood Approaches

TL;DR

This paper investigates profile-likelihood methods for estimating human population size from dual-record systems, focusing on behavioral dependence models and proposing an adjusted approach, with simulations and real data illustrations.

Contribution

It introduces an adjusted pseudo-likelihood method for the behavioral dependence model Mtb in capture-recapture studies, enhancing estimation accuracy.

Findings

Proposed method outperforms traditional profile likelihood in simulations.

Adjusted approach shows robustness under model mis-specification.

Real data examples demonstrate practical applicability.

Abstract

Motivated by various applications, we consider the problem of homogeneous human population size (N) estimation from Dual-record system (DRS) (equivalently, two-sample capture-recapture experiment). The likelihood estimate from the independent capture-recapture model Mt is widely used in this context though appropriateness of the behavioral dependence model Mtb is unanimously acknowledged. Our primary aim is to investigate the use of several relevant pseudo-likelihood methods profiling N, explicitly for model Mtb. An adjustment over profile likelihood is proposed. Simulation studies are carried out to evaluate the performance of the proposed method compared with Bayes estimate suggested for general capture-recapture experiment by Lee et al. (Statistica Sinica, 2003, vol. 13). We also analyse the effect of possible model mis-specification, due to the use of model Mt, in terms of efficiency and robustness. Finally two real life examples with different characteristics are presented for illustration of the methodologies discussed.