Multiple regression?


I'm wondering if it is a bad violation to try to conduct a multiple regression analysis if some of the variables are non-parametric?

My dependent variable is olfactory test and I am trying to see if I can establish variables which are accounting for variance on this test. Some of the variables such as IQ, age etc are normally distributed but some of the test scores that I wish to put into the regression i.e. short-term memory scores, language production scores, are not.

I have used a combination of parametric and non-parametric stats (Kruskal-Wallis) to look at differences across groups already.

Can I use the multiple regression and what would the justification be?



TS Contributor
There really isn't a good method of multiple regression for nonparametric variables. Try transforming the non-normally distributed variables using natural log or other (square root, etc.) and run the multiple regression the regular way.
ROC Curves

Hi thanks for the help, -another question!

I am carrying out ROC curve on SPSS to try to establish the sensitivity and specificity of three measures in my thesis.

I'm wondering do I have to meet specific assumptions to use this curve?

The data I will will using is 'normally distributed'...

I'm just inserting the informtion into the ROC command and trying to interpret the best area under the curve i.e. best sensitivity and specificity of test. Is this ok? Any advice?


TS Contributor
Tests that use contingency tables or assess sensitivity or specificity are usually considered nonparametric, so it doesn't matter what the underlying distribution is.