Test the equality of coefficients (mi estimate)

[Mi Choi]

Hello All,

I want to test the equality of a coefficient (x1) in the two models using the "mi estimate" command.
The variables are the same in the two models, but the two models use different samples (model 1 is only for male, and model 2 is only for female).

Here is my command:

mi estimate (diff: [est1_depress]x1 - [est2_depress]x1): mysuest "regress depress x1 x3 x4 x5 if x2==1" "regress depress x1 x3 x4 x5 if x2==2"

x2 is sex with 1=male, 2=female.

Is it okay to use this command for two different sample sets (male and female)?

I appreciate your reply!




My Stata version is Stata/MP 16.1

[Richard Williams]

I assume mysuest is a command you wrote yourself?

My first impulse would be to compute interactions with x2, estimate a pooled model with the interactions, and then use mi test to test whether the interaction terms are jointly significant. This seems like a more conventional approach. Your way may work, but I don't really know especially since I don't know what mysuest is.

[Mi Choi]

Thank you, Richard.

I appreciate your suggestion of using the interaction term.

I got the code, "mysuest", from this source: https://www.stata.com/support/faqs/s...-imputed-data/
The example at the bottom of the linked page tests the coefficient on a variable from the two logistic models. In the example, two models use the same sample but different numbers of variables.

For my study, I also compare the coefficient of a variable between two models. However, my two models include the same variables, but use different samples. I wonder if my code above can use for models with different samples.