All:
I'm estimating logit models using MI data (m=20) using Stata 12. I want to test an interaction term: race(black/white) X treatment. From everything I've been reading, I understand that I cannot simply use the parameter estimate for the interaction term to determine whether the interaction is significant. Instead, it appears that I need to use marginal estimates to examine this question.
My questions to the listserv are:
1) Is it correct that examining margins is the best/appropriate way to look at interactions in nonlinear, logit models?
2) Since I'm using MI data, can I use Stata's margin command, or do I need to use another procedure (I've just downloaded the user-written program mimrgns, for instance)?
Any advice/help/suggestions is appreciated.
Scott Holupka
Johns Hopkins University
Institute for Health and Social Policy
I'm estimating logit models using MI data (m=20) using Stata 12. I want to test an interaction term: race(black/white) X treatment. From everything I've been reading, I understand that I cannot simply use the parameter estimate for the interaction term to determine whether the interaction is significant. Instead, it appears that I need to use marginal estimates to examine this question.
My questions to the listserv are:
1) Is it correct that examining margins is the best/appropriate way to look at interactions in nonlinear, logit models?
2) Since I'm using MI data, can I use Stata's margin command, or do I need to use another procedure (I've just downloaded the user-written program mimrgns, for instance)?
Any advice/help/suggestions is appreciated.
Scott Holupka
Johns Hopkins University
Institute for Health and Social Policy