LCCR: An R Package for Inference on Latent Class Models for Capture-Recapture Data with Covariates

A detailed description of the R package LCCR for the analysis of capture-recapture data relating to a closed population is provided. The data that can be analyzed consist of the full capture history for each sample unit and may possibly include individual covariates. The package allows the specification and estimation of latent class models in which the distribution of the capture history conditional on the latent class and covariates follows either a log-linear model in which bivariate interactions are allowed, or a logit model for the conditional probability of capture at each occasion given the previous capture history. Alternatively, the conditional distribution of each capture occasion can be formulated by a logit model that may account for the effect of previous capture occasions. Apart from the conditional distribution of the capture history, the covariates can also affect the distribution of the class weights. Estimation is based on the unconditional maximum likelihood method, suitably extended to account for the presence of covariates by including unit-specific weights, which are commonly used in empirical likelihood methods. The package also allows simulation of capture-recapture data from a specified model and the computation of the profile confidence interval for the population size. For illustration, we use a data set about meningitis cases in an Italian region.