Difference between ATE and ATET in the context of teffects IPW

[Kyle Huisman]

I've been searching through the Stata manuals and various forums for a while but I've not been able to find a straight answer as to the exact difference between the calculation of ATE and ATET in the context of teffects IPW, though I understand the theoretical difference between ATE and ATET that you are averaging the difference between the treated and hypothesized untreated outcome for the treatment group only. According to GPT-4, the difference between ATE and ATET in the context of IPW is that in ATET the mean outcome of the treatment group is not weighted by the inverse probability weights. Is this correct, or is there more too it?

[FernandoRios]

Nothing better than try it yourself

webuse cattaneo2, clear
teffects ipw (bweight) (mbsmoke mmarried c.mage##c.mage fbaby medu, logit), atet 
teffects ipw (bweight) (mbsmoke mmarried c.mage##c.mage fbaby medu, logit), ate 

logit mbsmoke mmarried c.mage##c.mage fbaby medu
predict pr
gen wgt_ate = (1/pr)*(mbsmoke==1) + 1/(1-pr)*(mbsmoke==0)
gen wgt_atet = (mbsmoke==1) + pr/(1-pr)*(mbsmoke==0)

reg  bweight mbsmoke [w=wgt_ate]
reg  bweight mbsmoke [w=wgt_atet]

[Kyle Huisman]

Of course there isn't! Can't believe I did not think to do that. Thanks, Fernando.