predicted probability after teffects ipwra

[ajay pasi]

Hi Statalisters,

I'm using the teffects ipwra command for a research project and I have a question. Since I'm new to this estimation strategy, my question might seem basic.

In a logit analysis, we can easily find the predicted probabilities of covariates using the margins command after running the logit model. I'm wondering how to do this after running a teffects model. Specifically, I'm interested in finding the margins for gender (male and female).

Is my question valid one?

Any guidance would be greatly appreciated.

Warm regards,
Ajay

[Jeff Wooldridge]

I don't know if there's a way to do that using teffects ipw (as you did above) or teffects ipwra. But you can always get this by just estimating the propensity score model by logit and use the margins command -- just as you describe. Both IPW and IPWRA, as defaults, use standard logit models for the PS.

[George Ford]

the covariates are not in the final equation.

use https://www.stata-press.com/data/r18/cattaneo2 , clear

teffects ipw (bweight) (mbsmoke foreign alcohol mage medu fage fedu , logit)

qui logit mbsmoke foreign alcohol mage medu fage fedu
qui predict double ps if e(sample)
qui gen double ipw = 1.mbsmoke/ps + 0.mbsmoke/(1-ps)
reg bweight i.mbsmoke [pw=ipw]

I'm interested in how Jeff proposes to get the effect of female. Easy enough to get the margins on mbsmoke, but not sure where to go from there.

[George Ford]

This?

use https://www.stata-press.com/data/r18/cattaneo2 , clear

teffects ipw (bweight) (mbsmoke foreign alcohol mage medu fage fedu , logit)

logit mbsmoke foreign alcohol mage medu fage fedu
qui predict double ps if e(sample)
qui gen double ipw = 1.mbsmoke/ps + 0.mbsmoke/(1-ps)
margins, over(foreign) post
local f0 = e(b)[1,1]
local f1 = e(b)[1,2]
reg bweight mbsmoke [iw=ipw]
margins, at(mbsmoke = (`f0' `f1'))

[ajay pasi]

Thank you Jeff Wooldridge and George Ford