I have a response variable y which is binary, several feature variables xi which may be float or integer and a grouping variable g which is categorical. I would like to know which variables Xi i=1:n effect y and if this effect is group dependent. To make the story even more complicated, each measurement is associated with a subject. Anova classically assume response to be continuous and independent variable to be categorical but here I have the opposite situation. Is it legitimate to use Anova anyway, treating Xi as the dependent variable? Notice that g has a marginal effect on y which is not interesting. So what I am looking at is actually the effect of Xi on y and the product effect of Xi and g, taking the direct effect of g out. I know there is Anova but I am not sure it's applicable. Maybe logistic regression?
$\begingroup$
$\endgroup$
2
-
1$\begingroup$ I think you should investigate logistic regression as you suggest. $\endgroup$– mdeweyCommented Jan 9, 2017 at 11:58
-
$\begingroup$ Try to rewrite the title to be more consistent with the question! $\endgroup$– kjetil b halvorsen ♦Commented Jan 31, 2021 at 6:20
Add a comment
|