Post-estimation, wald tests after using Mlogit for MI Estimate

[Mike Holland]

Hello Statalist community,

I am using Stata version 13.
I want to run a mlogit model (multinomial logistic regression) after MI estimate (for imputed data), and svyset (clustering schools and regions).

Here is my model:

mi estimate, rr: svy: mlogit summerlowses_group intervaldays i.grade ///
i.ethnicityx childage2012 parentage2012 ///
i.Male i.born_in_canada siblings ///
i.time_on_homeworkx i.Readx Financial_plan participate resources aspirations i.participate_in_org ///
Spring_2012_PMB_DRA Average_Grade, ///
base(0)

Since my dependent variable is 3 categories ("other summer learners", summer loss, and summer gains), I wanted to use a Wald test to look at the differences between summer loss and gains ("other summer learners" = base category for the mlogit model). However, I cannot find a way to do this.

I would like to use a wald test, but currently not sure how to set this up after MI estimate.

Prior to using MI estimate, (I used ICE), I was simply able to conduct a Wald test using the "test command". For example:

test [Low_SES_Summer_Gain]2.time_on_homework = [Low_SES_Summer_Loss]2.time_on_homework

This command produces an f-test. Because it was significant, it informed me that time spent on homework was a better predictor of summer gain than summer loss.

But with MI estimate, I do not know how to recreate this.


I was experimenting with mim2, (not using svy) such as:

xi:mim2: mlogit summerlowses_group intervaldays i.grade ///
i.ethnicityx childage2012 parentage2012 ///
i.Male i.born_in_canada siblings ///
i.time_on_homeworkx i.Readx Financial_plan participate resources aspirations i.participate_in_org ///
Spring_2012_PMB_DRA Average_Grade, ///
base(0)

and then using:

mim2: testparm

However, mim2 and testparm do not allow for options for me to specify equations. I also do not know how to specify outcome group.
My goal is to look at how the independent variables may or may not be better predictors of summer learning loss vs summer learning gains. Currently, my model allows me to compare the effects of these factors on the likelihood of summer learning gains and summer learning losses compared to the base category ("other summer learning).


Any advice would be appreciated.

Thanks,

Mike