1

I ran a chi squared test on multiple features & also used these features to build a binary classifier using logistic regression. The feature which had the least p value (~0.1) had a low coefficient (=0) whereas the feature which had a higher p value (~0.3) had a high coefficient (~2.9). How do I interpret this?

Is it possible for a feature to have low p value but have zero coefficient?

user16584277
  • 149
  • 1
  • 1
  • 9
  • 1
    It will matter what the standard errors on those coefficients are. If your estimate is $6$ but with a standard error of $12$, that would not be significant at the usual levels like $0.05$. If your estimate is $0.1$ but with a standard error of $0.01$, that will be significant at the usual levels. // Another issue could be if your variables are correlated. // Screening predictors based on individual relationship with the outcome is discouraged. – Dave Aug 18 '21 at 17:04

1 Answers1

1

The magnitude of the coefficients is not to be considered while finalizing a model. The magnitude of the coefficient should be completely ignored while choosing them.

Also, the significant coefficients will not be zero, they can be low numbers, if you want to see higher numbers transform the same variable by doing *100 across the whole variable, the coefficient will come out to be a higher magnitude number.

Low p-value variables are significant even if their coefficient is low.

srishtigarg
  • 131
  • 3