teffects psmatch - Identifying patients in the treated and control group post-matching

[Tanner Smida]

I am performing propensity score matching using teffects psmatch. I would like to generate a table containing the baseline characteristics of patients after propensity score matching. To do this, I generated a match variable [teffects psmatch (outcome) (treatment X1 X2 X3 etc.), atet gen(match)]. Is it correct to use all 'treated' patients with a non-missing match value and all 'control' patients with a non-missing match value to generate this table? If not, how can I identify patients belonging to the post-propensity score matching treatment and control groups?

In addition, when using default settings, I am getting up to 10 matches (controls being used as matches up to 10 times) despite specifying 1:1 matching. Could this just be a feature of the sample size of my control/treatment groups (treatment several times smaller than control), or is there an additional term I should add to my code? Stata output attached.

Thank you for your time and effort.

[George Ford]

When a matched control is matched more than once, it will be counted more than once.

This is a kluge, but shows you what's up. Not sure if there's something else in there that could automate this. The t-stat is not the same but the effect size is.

sysuse auto, clear

g id = _n
capture drop match1
teffects psmatch (price) (foreign mpg weight turn displacement), atet gen(match)
egen matched = anymatch(id), v( 24 3 3 44 3 24 16 24 44 24 3 20 3 43 20 44 24 3 16 3 44 1)
replace matched = 1 if foreign
replace matched = 7 if id==3
replace matched = 2 if id==16
replace matched = 2 if id==20
replace matched = 5 if id==24
replace matched = 4 if id==44
reg price foreign  [fw=matched] if matched>0
tabstat mpg weight turn displacement [fw=matched] if matched>0, by(foreign)

[George Ford]

psmatch2 will give you the weights (_weight), making it easier.