Some of my variable shows weak collinear relationship on my dependent variable.. how should I address this?


closed as too broad by Giskard, Fizz, Bayesian, emeryville, BB King Apr 28 at 22:31

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ Did you mean between independent variables? $\endgroup$ – Adam Bailey Apr 22 at 9:04

Your question contains very little information so these are my two cents. Assuming your model is correctly specified and there is no underlying problem with the data you are using, there's really little you can do about this. In my experience, this is often the case where some variables are included based on the literature and the relationship found in the regression model does not necessarily reflect the hypothesized relationships. Now, assuming you are studying a relationship based on some theoretical work and that your sample isn't small (so as to have a negligible effect of insignificant variables on p-values), you should not drop variables (this is the equivalent as specifying a theoretical relationship based on the results of a regression model and it is, from a purely academic perspective, the wrong direction to work in). This is generally the case where these variables have a hypothesized relationship as predictors or serve as controls. When deciding whether the addition of these variables may create a problem in the OLS estimation you may try to run different regressions adding and excluding those variables to see if there are any important changes (e.g. in the variance explained by your model or in the overall fit). Now, as I said, there is little information provided, so I am assuming many things here.

  • $\begingroup$ Ok. Thank you. From what I understand, I can still fit my variables into the regression even If the correlation isn't significant since in the first place I have a theoretical model to back up for it. However, upon running it I should modify my variables as to fit the model....Actually, this is exactly what I did so I drop one of my variables to address multicollinearity problems. However, I was just pondering if I could drop a variable based on pearson correlation... $\endgroup$ – Ry G Apr 21 at 23:45
  • $\begingroup$ since disregarding those variables which are insignificant have made my p- values for other variables more significant unlike before which only one of it is significant. $\endgroup$ – Ry G Apr 21 at 23:50
  • $\begingroup$ If all your variables are justified from a theoretical perspective then you should not drop them just because doing so makes the remaining variables more significant. You can, however, report on this and possibly think of reasons as to why your regression model is reflecting this. I suggest you also look at the empirical literature on your specified relationship to see if others have had similar results. $\endgroup$ – user20105 Apr 22 at 10:39

Not the answer you're looking for? Browse other questions tagged or ask your own question.