2
$\begingroup$

I ran a logistic regression with 1 predictor for dichotomous outcome, and the predictor was not statistically significant. I ran an ROC curve with the same variable, and got AUC around .70 which is fairly good predictive ability. How do I explain these discrepant results with the same variables?

$\endgroup$
2
  • $\begingroup$ Can you provide more details about data and the model. $\endgroup$ – Michael R. Chernick May 9 '17 at 0:50
  • 1
    $\begingroup$ Note that the AUC does not really measure "predictive ability". It measures the degree to which the predicted probabilities tend to sort the 1s & 0s. $\endgroup$ – gung - Reinstate Monica May 9 '17 at 11:56
4
$\begingroup$

The most likely reason is that your sample is small.

Statistical significance is not the same as practical importance. With a large enough N, even a tiny AUC would be significant. With a small enough N, even an AUC of 0.9 would not be significant.

$\endgroup$

Your Answer

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

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