Abstract:

 Capture–recapture methods are statistical procedures based on a sequence of trapping experiments originally used to estimate the size of wildlife populations. The rationale behind this method is to account for unobserved individuals by using observed individual trapping histories. In human populations, capture-recapture methods can be utilized to estimate the demographic characteristics of interest using information from two or more incomplete but overlapping lists of cases from different sources.  In literature the methods are also referred to as multiple system, multiple-records system and multiple-records system methods . Here, each list is viewed as a capture sample, the identification number (or name) is treated as a mark (or tag) and the statement "being captured in sample i" is replaced by "being observed in list i". Data are usually arranged in a 2^S contingency table (where S is the number of available lists), with one missing cell corresponding to absence in all lists and data are analyzed by the use of log-linear models. Modelling dependence between lists is one of the major issue in multiple system framework. Dependence among lists may be due to both list dependence (inclusion in a list has a direct causal influence on inclusion to another list), or heterogeneity between individuals (differences of behavior results in heterogeneous inclusion probabilities and may cause indirect dependence between lists). To model dependence, we propose the use of the log-linear multidimensional Rasch (MR) model. In the model, heterogeneity of capture probabilities is taken into account, and lists are viewed as dichotomously scored indicators of one or more latent variables that can account for correlations among lists. In particular, under this assumption, the probability of a generic capture profile can be easily expressed in a log-linear form, using an extension of the Dutch Identity for the multidimensional partial credit model. The resulting model can be used either in the case with or without a stratifying variable available.