As one of the preliminary steps in my data analysis project, I am looking at a correlation table I made of my potential predictors and one output variable. There is no significant information overlap between the predictors. (There are very weak correlations of 0 with a deviation of about +-0.05) However, two of these predictors are quite correlated with the output variable, their values are about 0.6.

I learned in class that usually you remove/combine predictors that are fairly correlated with each other, but how does this relate to a predictor correlating with the output variable? My instinct tells me that this might be quite a useful predictor.

Should I have not included the output variable in my correlation matrix in the first place? How do I treat these predictors?

  • $\begingroup$ Isn't the whole point to find variables that are strongly correlated with what you want to predict? Why would you possibly want not to use those variables? $\endgroup$
    – whuber
    Apr 13, 2016 at 20:59
  • $\begingroup$ but always remember "correlation does not equal causation" $\endgroup$
    – bdeonovic
    Apr 13, 2016 at 21:14

1 Answer 1


You ought to be quite happy that some predictors are highly correlated with the response, that gives reason to expect successful modelling. That some predictors have correlation about 0.6 gives no reason for concern, possible problems ("multicollinearity") would only appear at much higher correlation levels.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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