Propensity Score Matching with psmatch2 / teffects

[Franziska Hittmair]

Hello,
I am trying to perform propensity score matching in order to calculate weights and perform a regression with the matched sample.

I tried using both teffects and psmatch2.

I want to match on one continuous variable (x1) and 2 categorical variables (x2, x3).

. psmatch2 treatment x1 i.x2 i.x3, out(y) logit

this code works and provides me with the _weight output variable i need for regressions using

 [fweight=_weight]

However, if i want to perform a 1:k match like this

. psmatch2 treatment x1 i.x2 i.x3, out(y) logit neighbor(3)

my _weight variable becomes an integer and therefore i cannot use frequency weights in my regression anymore

(error: may not use noninteger frequency weights
r(401)

How do I change that and calculate frequency weights from my non-integer weights that the 1:k matching with psmatch2 provides me?

Also, i tried using the teffects command.

However, teffects does not allow the categorical specification with i.x2 or i.x3.
I tried using dummy variables instead for all my categories but then i have too many perfect predictors.
Is there any way i can still use teffects psmatch with categorical variables ?
Is it ok to just use x2 and x3 without specifiying them as categorical variables, e.g:

teffects psmatch (Y) (treatment x1 x2 x3), gen(match)

thank you

[David Radwin]

I'm not sure that you need to use the _weight variable at all for the "match first, then regress on the matched sample" approach. See the Stata code in appendix SA2 of the following article and the article itself.

DuGoff, E.H., Schuler, M., and Stuart, E.A. (2014). Generalizing Observational Study Results: Applying Propensity Score Methods to Complex Surveys. Health Services Research, 49(1): 284–303. https://onlinelibrary.wiley.com/doi/...475-6773.12090