Consequences de la sélection de variables sur l'interprétation des résultats en régression linéaire multiple

Abstract

Consequences of variable selection on the interpretation of the results in multiple linear regression. A priori or a posteriori variable selection is a common practise in multiple linear regression. The user is however not always aware of the consequences on the results due to this variable selection. In this note, the presence of omission bias and selection bias is explained by means of a Monte Carlo experiment. The consequences of variable selection on the regression coefficients and on the predicted values are then analysed. The user’s attention is drawn to the risk of misinterpretation of the regression coefficients, specially after variable selection. On the other hand, the consequences of variable selection on the predicted values of the response variable are rather limited, at least for the given example.

Keywords

• Regression
• variable selection
• omission bias
• selection bias
• simulation
• statistical method