ABSTRACT
Capture-recapture (CR) methods provide a natural way to estimate the unknown size of a partially observed population, through samples derived using some identification mechanism. This chapter reviews work on the CR method using the multinomial (conditional) logit model (MCML). MCML treats continuous covariates in their original measurement scale to estimate the size of a population of interest. The chapter also reviews the conditional likelihood approach that allows for the modeling of dependence between sources for models incorporating continuous (and categorical) covariates. These models can be fitted with available software by exploiting the similarity of the likelihood with that of the stratified proportional hazards model. The chapter focuses on observed heterogeneity whereby capture probabilities are allowed to vary with auxiliary variables. Other models treat heterogeneity as a latent feature, without using covariates. These models are important when time auxiliary covariates are not available or uninformative.
