Subgroup analysis on multiply imputed complex survey data

[Esme Lilly]

Hi all,

I am trying to compare coefficients between two groups using multiply imputed complex survey data. Usually I would use the suest command and then test the two coefficients, however, suest isn't supported by mi estimate.

I followed the advice here: https://www.stata.com/support/faqs/s...ata/index.html however, this suest equivalent programme does not work with svyset data.

The model is as follows:

mi svyset psu [pweight = xweight], strata(strata) singleunit(scaled)

mi estimate: svy: reg labrank hhrank edu yphfm14 ypeatn14 if sex==0
eststo male

mi estimate: svy: reg labrank hhrank edu yphfm14 ypeatn14 if sex==1
eststo female

Is there any other way to test the coefficients between the two groups, that works for both multiply imputed and svyset data?

Thanks.

[Andrew Musau]

There is no need for suest here. The coefficient on the interaction term below will give you the difference that you want.

mi svyset psu [pweight = xweight], strata(strata) singleunit(scaled)
mi estimate: svy: reg labrank i.sex##(c.hhrank c.edu c.yphfm14 c.ypeatn14)

or else, the regression with all coefficients (suest equivalent) is given by:

mi svyset psu [pweight = xweight], strata(strata) singleunit(scaled)
gen cons=1
mi estimate: svy: reg labrank i.sex#(c.hhrank c.edu c.yphfm14 c.ypeatn14 c.cons), nocons

[daniel klein]

Did you impute missing values separately for the two groups? If not, did you include all those interactions during imputation? If not, then your test, however conducted, will likely be biased towards zero.

[Esme Lilly]

Thank you both for your advice, I will follow both.