Statistically compare two proportions on multiply imputed data

[Mengmeng Li]

Dear all,

I calculated two proprtions based on 50 multiply imputed datasets and received the following output:

. mi estimate, post: proportion aces_exp1, over(sex)

Multiple-imputation estimates Imputations = 50
Proportion estimation Number of obs = 607
Average RVI = 0.0174
Largest FMI = 0.0272
Complete DF = 606
DF adjustment: Small sample DF: min = 582.52
avg = 591.04
Within VCE type: Analytic max = 599.55

-------------------------------------------------------------------------
| Normal
| Proportion Std. err. [95% conf. interval]
------------------------+------------------------------------------------
aces_exp1@sex |
Never/rarely Male | .6100332 .0285005 .554057 .6660094
Never/rarely Female | .5622876 .0284572 .5063996 .6181756
Sometimes/often Male | .3899668 .0285005 .3339906 .445943
Sometimes/often Female | .4377124 .0284572 .3818244 .4936004
-------------------------------------------------------------------------

I want to statistically compare two proportions: ".3899668" vs. ".4377124" and did not find the informaiton from "mi estimate" manual.

Your input is greatly appreciated.

Thank you.

Mengmeng

[Clyde Schechter]

-help mi test-

[Mengmeng Li]

Hi Clyde,

Thank you for your suggestion.

I also looked into "mi test" command before posting my initial question. From my reading of "mi test" help file, the command only compares coefficients or a equation of coefficients from a regression model. I did not see any information about two proportions comparison. Did I miss anything?

Best,
Mengmeng

[Clyde Schechter]

If you look at e(b) after you run -proportion- you will see that the proportions are in fact coefficients of a linear probability regression model. So -test- is applicable. All this is also true after -mi estimate-.

[Mengmeng Li]

Thank you so much for your guidance, Clyde! I got it to work this time. It appears that I can also specify the calculation that I wanted to do statistical comparisons right after the "mi estimate" command, like this:

mi estimate, (_b[[email protected]]-_b[[email protected]]): proportion aces_exp1, over(sex)

Best,
Mengmeng

[Clyde Schechter]

mi estimate, (_b[[email protected]]-_b[[email protected]]): proportion aces_exp1, over(sex)

Yes, I had forgotten that you can do that with -mi estimate-. Thanks for reminding me of it! But you have one detail wrong: there should be no comma after -estimate-.

[Mengmeng Li]

Thanks for flagging this, Clyde!