Résumé
Interpreting interaction terms in econometric modelling is fussy. McFadden (2001) explains the complexity of consumers' choice faced with complex attributes of good. He emphasizes the difficulty of econometric models to reveal the psychologist process which is most of the time non directly observable, but induces behavioral effects. The models logit and probit are the more used modelling consumers' choice. However, they omit the interaction effects among the explanatory variables during the process of choice. These blended effects are however declared in the modifications of the consumption decision. The difficulty is then to interpret the coefficients associated with these interaction variables. Aï Norton (2003) have demonstrated, in the case of non linear models, that these coefficients could not be appreciate with a simple t-test. To solve this problem, we propose a decision rule enforceable whatever the nature of the estimators and for each kind of models, linear and non-linear. We build a simple rule decision, which offers an easy lecture and then becomes very convenient. We carry out an application of this decision rule to the choices of the wine consumers confronted with increasingly sophisticated products: which arbitrations to realize between the color, the degree or the type of wine when these various attributes combine on the labels and on increasingly broad rays? How to take into account these combinations of attributes for better understanding the processes of choice of the consumers? Using data from ONIVINS - Recensement exhaustif des premiers contrats de vente non AOC de 1987 `a 2003 (fctv03) - we give an experimental result describing the arbitration of wine consumers between color, degree and type of French wine.