teffects psmatch - how to check if the matching is acceptable?

[Luna Viana]

Dear all,

I am struggling with the function teffects psmatch.

I need help with how to evaluate if a reasonable matching was achieved.

After running the teffects pmatch i run tebalance summarize, baseline. But what threshold should I use to evaluate the matching?

Here is an example:

the variables are standardized.


I believe one way would be to perform a ttest of the difference in means. But I don´t know how to consider only the matched observations.

Help.
Thank you
Luna

[Maxence Morlet]

With propensity score matching, you generally plot propensity score histograms for treated and control and check the overlap (a.k.a common support).

[George Ford]

You don't want a t-test but a standardized difference (like a t-test but without sample size). t-test are too sensitive to sample size. See Imbens & Wooldridge JEL.

covbal is a useful approach and it permits weights.