[Gretl-users] RES: RES: Same model, different position of variables, different results

Dear all, Thank you very much for your replies (Rodrigo, Sven and Henrique). This was part of an exogeneity test. When the variable yhat1 was excluded, the test does not work. But when one rearrange the model in a way that yhat1 is not placed at the end, the test is possible (i.e. gretl excludes logarea instead of yhat1, due to the collinearity problem). Best, Bruno De: gretl-users-bounces(a)lists.wfu.edu [mailto:gretl-users-bounces@lists.wfu.edu] Em nome de Rodrigo Alfaro Arancibia Enviada em: sexta-feira, 5 de outubro de 2012 10:12 Para: Gretl list Assunto: Re: [Gretl-users] RES: Same model, different position of variables, different results Dear Bruno, It seems that you are running a regresion with boht logarea and yhat1, however, gretl drops one of the variable because they collinear. That is right given that correlation is 99.93% abertura logrendpc logarea yhat1 1.0000 0.1402 -0.6693 0.6698 abertura 1.0000 -0.1715 0.2093 logrendpc 1.0000 -0.9993 logarea 1.0000 yhat1 Now, your question is why you have a different dropped variable if you change the order of the regressors, right? I am not sure why that is the case, but it happens also in other packages. I remember similar issue in Stata, when you have a different dropped variable if you use a different command, but forcing to apply the same procedure. For example comparing OLS con IV (but with trivial instruments). Although, I don't have the question I think is not a serious problem. If 2 variables are highly correlated then any of those can be used. Indeed, many times you could cut down your regressor into principal components to avoid collinearity. Moreover, yhat1 is predicted variable from a previous model and it is likely that logarea (or some similar variable) is a relevant in that model. Best, Rodrigo.