Propensity score match under conditions

[Rachel Sleeps]

Hi,

I am using Stata 13 and use teffects psmatch to undergoe propensity score matching. My question is the following: Can I restrict a matching on certain conditions?
Here is an example:

clear

use http://ssc.wisc.edu/sscc/pubs/files/psm, replace

set seed  42

gen ob=_n

gen ind_1 = round(runiform()*2)+1
gen ind_2 = round(runiform()*2)+1

teffects psmatch (y) (t x1 x2), nn(3) gen(match)

list ind_1 ind_2 if ob==1 | ob==467 | ob==356 | ob==218

teffects psmatch (y) (t x1 x2 ind_1 ind_2), nn(3) gen(_match)

list ind_1 ind_2 if ob==1 | ob==467 | ob==247 | ob==218

I use a given dataset and generate two fictive indicators ind_1 and ind_2. I than calculate the propensity score based on x1 and x2 and the ATE based on a nearest neighbour matching and require at least 3 matches. I did not include ind_1 and ind_2 in the matching - I do that in a second step. the gen() obition allows me to identify which observations get matched. For example in the first case, observation 1 with observations 218, 356, and 467, and in the second case observation 1 with observations 218, 247 , and 467.

However (and rightfully so, since ind_1 and ind_2 are entirely random), observations are matched with differing indicators (see below).

+---------------+
      | ind_1   ind_2 |
      |---------------|
   1. |     1       1 |
 218. |     1       2 |
 247. |     3       2 |
 467. |     1       2 |
      +---------------+

Is there a way to ensure that the treatment and control exactly match on certain criteria?

Best and thanks,
/R

[Rachel Sleeps]

What I am looking for is what the option ematch(varlist) does in teffects nnmatch - namely match exactly on specified variables. However, teffects psmatch does not provide this option.

[Johannes Scharzer]

Sorry, I have no answer but the same issue, so replying to stay stay in the loop

[Rachel Sleeps]

nobody?