Probabilistic record linkage with less than three matching variables

Probabilistic record linkage based on Fellegi-Sunter theory is a methodology for integrating data collected in different sources when a unique common identifier is not available. It requires at least three matching variables are available to identify the probability model. In official statistics, it is emerging the need to join archives even with less than three common variables, this is the case for instance of addresses and business archives of poor quality. For this problem, we compare available common variables by means of string comparators and propose mixtures of continuous and categorical distributions rather than usual the latent class models to estimate linkage probabilities