I ran a logistic model to predict a YES/NO (equivalent to 0 or 1, that is a binary variable) through a stepwise selection (I have either categorical and numerical variables).

I was then asked to perform a "correlation/association" analysis between the dependent binary variable and the independent variables.

I used Cramer's V and Spearman correlation indexes and my result was that there is a low correlation (from to 0.1 to 0.4 maximum) between binary target variable and every independent variables.

Does it make sense to have such a low measure of correlation??

I thought I would have found high indexes of correlation :(

Thanks :)

  • $\begingroup$ Yes, low correlation also makes sense. It can be possible that attributes are not that correlated with target variable. But, sometimes the data can be wrong and that can lead to misleading results, so i would suggest you to first make sure the data you are using is not bad data (don't have wrong entries or outliers in it). For association, it is better to build decision tree and look for nodes/decision rules which can lead to majority of one class. Also, Why you thought those features will have higher correlations ? $\endgroup$ – Harshit Mehta Jan 24 '19 at 21:45
  • $\begingroup$ Is your independent variable categorical or numerical? $\endgroup$ – ColorStatistics Jan 24 '19 at 23:51
  • $\begingroup$ The independent variables (features) are either numeric and categorical. I tried to bin the numeric ones by binning procedure so that outliers were grouped into one class and do not “hurt” the results. Through a decision tree l found that the same variables found in the stepwise selection are discriminant. Regarding the high correlation expected to me, l remember that the variables got After stepwise selection should have a high correlation with the binary target variabile..am l wrong? $\endgroup$ – LUm-1 Jan 25 '19 at 7:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.