Log-linear models for assessing persons' response structures

An interpretation of log-linear models for the analysis of twodimensional and multidimensional tables is presented, in order to assess persons' response structures. The structures characteristic of persons' response cross-classifications are represented by models specified by independence and conditional independence probability forms respectively for two and for more than two persons. Controlling the fit of the models and comparing the valid model in order to find the most parsimonious one, in terms of number of parameters, allow to define the best model which represents the structure of the persons' responses in a group. An illustrative, experimental example is made in order to control the fit of threedimensional log-linear models. Results showed that the hypothesis of mutual independence was demonstrated in three out of five cases analysed and the hypothesis of conditional independence was demonstrated in eight out of 11 cases.