Pooling Multimputation Variable Outputs

[Maya Muenzer]

Hello all!

I am working with a complex survey dataset and have 1 variable, povcat, that needs to be imputed. Due to the complex survey design, I need to use *svy* before my commands, and due to the multimputation (with 6 implicates), I need to use *mi* before commands as well.

Is there a way in Stata 17 to pool the tabulated data for an imputed variable (in my dataset, povcat_i) that comes from the *mi xeq* command? I have been doing that by hand by averaging the values from the 6 implicates, however this is very time consuming.

Here is an example of the code for unpooled tabulate of the variable which runs well and outputs proportions:

mi xeq: svy: tab povcat_i

I am also wondering if there is a way to get the estimated weighted frequencies (rather than proportions) of an imputed variable like povcat_i?

Any advice on working with a complex survey design while simultaneously working with multimputation would be great!

Thank you!
Maya

[Weiwen Ng]

We’re you aware that you can use svyset within mi?

[Maya Muenzer]

When I tried using svyset within mi, I received this error:

mi est: svyset: tabulate povcat_i, count format (%12.3gc)
command prefix svyset: not allowed
r(198);

Also, the dataex command gave me this error:

datasex
input statement exceeds linesize limit. Try specifying fewer variables
r(1000);

My dataset has >400 variables and >100,000 observations. Is there a separate command or option that would enable me to give an example of data for this kind of dataset?

[Weiwen Ng]

I haven't personally done mi + svy before, so I hope this syntax is right. However, instead of svyset beforehand, you need to type mi svyset (svy parameters) to declare the data to be MI and svy data Note the lack of colons.

Also, I just realize that you're tabulating the data. I don't know if svy's tabulate is considered to be an estimation command, and mi will only support estimation commands. If tabulate works, you'd type

mi est: svy: tabulate povcat_i

That said, I am not sure it works. Now, you could get the mean of each category, I think, by

mi est: svy: regress i.povcat_i, nobase