Abstract

Multicollinearity refers to linear dependence among the explanatory variables in multiple regression. Most textbooks discuss the phenomenon as one of the problems that may hinder useful multiple regression analysis. This paper argues, in contrast, that multicollinearity is one of the main reasons why we would want to do a multiple regression analysis in the first place, especially in a context of multicausality. High multicollinearity does not preclude reliable estimation of regression models at all. When the estimated slopes of correlating explanatory variables are not reliable, then probably the model was estimated with ill-conditioned data, or maybe the theoretical model is wrong, or maybe some concepts were not operationalised well. Hence, remedies for unreliable estimation should be sought in data resources or in a conceptual cleanup, or both. Other tips and tricks are unlikely to be of any help to sociologists.

How to Cite:

Van Bavel, J., (2006) “Multicausaliteit en multicollineariteit bij meervoudige regressie”, Tijdschrift voor Sociologie 27(4), 351–375. doi: https://doi.org/10.21825/sociologos.86666