2
$\begingroup$

I have two continuous predictor variables to predict a dichotomous variable. In addition i have constructed two (interaction) models, based on domain knowledge which use both variables to predict the third.

So now i want to compare these predictors, using R. I get these result on the wilcox test: W = 36655, p-value = 3.896e-09 (single predictor) W = 29680.5, p-value < 2.2e-16 (model)

But i'm wondering if the area under the ROC curve would not be a good (better?) measure then the wilcox or the t-test. To get a general idea of how well the continuous predictor separates the two groups i plot the two histograms together. But the differences are small and besides, i'd like a 'grade' to tell me which is better and by how much.

$\endgroup$
2
  • $\begingroup$ Do you mean the Wilcoxon rank-sum test? $\endgroup$ Aug 10 '12 at 16:36
  • $\begingroup$ @Michael Chernick Yes. But i could also use the t test as the sample is large enough. $\endgroup$
    – Ivana
    Aug 11 '12 at 9:24
1
$\begingroup$

While the t test and the the Wilcoxon tell you about the statistical significance of the covariate's parameter value, the ROC curve gives a different picture, telling you how well cases are classified as a threshold changes. So from the viewpoint of classification the AUC of the ROC might be more interesting for you.

$\endgroup$

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.