1
$\begingroup$

Have a certain (high) number of rows corresponding to people. Have a bunch of independent variables which are group memberships and a dependent variable which is the outcome of a treatment. All are binary. Now, the group memberships are not exclusive -- a person can be a member of multiple groups, and the group memberships have some degree of collinearity. Feel like this is violating some assumptions of the $\chi^2$ test, namely that the rows and columns must sum to the number of observations.

What's a better test to check if the treatment of groups is different? Regression analysis (while looking at VIF?)?

Also the number of groups is too high (around 10) to enumerate all the possible combinations as new variables with distinct observation counts.

$\endgroup$

1 Answer 1

1
$\begingroup$

When I had a similar issue, I was recommended the Cochran's Q test for multiple non-exclusive groups, and McNemar for 2 non-exclusive groups.

$\endgroup$
1
  • $\begingroup$ Do you have a reference for the non-exclusive groups version of Cochran's Q test? $\endgroup$
    – levesque
    Mar 25, 2021 at 13:55

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.