Propensity Score Matching (PSM) for continuous treatment - levels limited to 20

[Claire Maggi]

Hello everyone,

I want to conduct a Propensity Score Matching for estimating the average treatment effect of a treatment on an outcome variable.
The peculiarity of my case is that I don't have a binary, but continuous treatment, i.e., my treatment variable can have 25 different manifestations (think 25 different intensities of my treatment).

Therefore, I use the teffects regression analysis command in stata:
teffects ra (OutcomeVariable Covariate1 Covariate2 Covariate3) (ContinuousTreatment), atet

However, apparently a continuous treatment variable is only allowed to have 20 different manifestations as I get the following error in stata (stata version 16.0):
treatment variable ContinuousTreatment has 25 levels exceeding the maximum of 20

I would like to avoid somehow adapting my variable to reduce the levels to 20.
Any help or ideas are appreciated - thank you!

[Jeff Wooldridge]

Claire: Your command implies that you're using regression adjustment, not propensity score matching. As far as I know, PS matching is only supported for binary treatments -- not even multiple treatments.

You would have very many treatment effect estimates, anyway. Your teffects ra command, if more than 20 levels were allowed, would estimate an effect going from the zero treatment level to each of the other 24 levels. Do you really want that much information? If you do, you can use

reg y i.treat#(c.x1 c.x2 ... c.xK), vce (robust)

But I'm not sure you want this. I bet it would be sufficient to use

reg y c.treat##(c.x1 c.x2 ... c.xK), vce(robust)