ABSTRACT
A general model is suggested for latent structure analysis of categorical data. The data is viewed as a contingency table and the connection to other approaches in contingency table analysis is briefly discussed. The suggested model assumes that category weights, needed to estimate a one-dimensional latent variable, are known, although some suggestions are made for tentative estimates of the weights. It is then shown that all the elements of well-known inference methods for the Rasch model applies, for example, population free estimates of item parameters and estimation in and check of the latent population distribution based on the observed score distribution. The theory is applied to one set of data from depression rating and one set from consumer complaint behavior.
