Les liaisons fallacieuses : quasi-colinéarité et « suppresseur classique », aide au développement et croissance

This paper shows that a multiple regression with two highly correlated explanatory variables, both of them with a near zero correlation with the dependent variable may correspond to a spurious regression or to a homeostatic model, with estimates highly sensible to outliers. The regression method does not allow how to decide which one of the two models is relevant. Statistical significance of the (very high) parameters is easily obtained, as shown doing Monte Carlo simulations. An example is provided by the Burnside and Dollar [2000] article on aid, policies and growth. Classification JEL : C12, C18, C52, F35, O47<np pagenum="558"/>