La regressione logistica e i modelli log-lineari nella ricerca sociale

The log-linear models and logistic regression assume two different representations of the relationships between variables: the firsts are suited to analyze structures of symmetrical relationships between variables and the seconds are used to analyze relations of dependence, therefore they requires the researcher to distinguish between dependent and independent variables. Starting from this general difference, the book describes the specificity of the two techniques, especially trying to highlight the circumstances under which a technique is preferable to the other. Besides this, the comparison between the log-linear models and logistic regression is also useful to show the potential and the limitations they have in common. I refer in particular to the difficulties which arise in interpreting the results of log-linear models and logistic regression when the two techniques are used to analyze relationships between polytomous categorical variables. These difficulties are not due to the formal aspects of the two techniques, but rather to the high semantic autonomy of categorical variables, which has important consequences in the analysis of the data, regardless of the technique used to analyze them.