Weighting data with AIPW and teffects

[Pierre Dupuis]

Hello everyone,

Actually, I try to understand how to use an IPW estimator.
I have 11 treatments variables in a panel dataset, and they can all takes a value of -1, 0 or 1.
Because of all of this, I choose to use the AIPW estimator, with an ordered logit model.
I have learned a bit how to use it, how it works, but I still have questions:
If I use, for example,

teffects aipw (outcome) (treatments covariates, logit)

there is no need to use xtlogit then predict? It is my understanding I have to choose between these two methods, am I right?
How to differenciate treatments and covariates in the code?

Thanks in advance for your assistance.

PS: I didn't start to test some code, so I may have further questions.

[Pierre Dupuis]

OK, it appears that I misread something: teffects doesn't work with treatments which contains negative values.
Is there any way to use IPW methods with negative values for treatments?
As I said, treatments in my dataset takes the value of -1; 0 or 1. These variables refers to economic policies, which can be loosened (value=-1), tightened (+1), or there is no change (0). The dependent variable is the GDP.
Here: https://stats.stackexchange.com/ques...ive-treatments my understanding is that I should modify the negative value. It seems strange to me, because we expect a negative relation between a tightened policy and the GDP.

Any suggestions please?

PS: sorry for the double post.