Modèle Linéaire contre modèle logistique en régression sur données qualitatives

Abstract

The Linear Model Versus The Logistic Model in Regression on Qualitative Data. Regression on qualitative data is usually done by using a logistic model. By examining data where "everything is otherwise equal", one can show that the linear model is quite compatible with this type of data. Results of an in-deep analysis of cross-tabulation data (called tabular analysis ) , and of regression using linear and logistic models, are quite similar (in terms of percentage effects). The theoretical question conceming the possibility of a linear model remains to be examined: it seems that the specific situation of a reference category introduces constraints which make the linear model possible.