Support boundaries for propensity score

[Bill Smith]

I'm using Sata 14.2 under Win 7E. I have run the following code to check the boundaries of common support in a propensity score model:

. psmatch2 anesthesia encounter_age bsa2 qrs2 rightventsize, /*
*/kernel k(biweight) common logit

tab _pscore _support

The results are interesting:

 psmatch2: |   psmatch2: Common
Propensity |        support
     Score | Off suppo  On suppor |     Total
-----------+----------------------+----------
  ...
       .61 |         0          1 |         1
       .66 |         0          1 |         1
       .68 |         1          0 |         1
       .71 |         0          1 |         1
       .74 |         0          1 |         1
       .76 |         0          1 |         1
       .77 |         0          1 |         1
       .83 |         1          0 |         1
       .85 |         1          0 |         1
       .92 |         1          0 |         1
-----------+----------------------+----------
     Total |         4        225 |       229

How can .68 be off support but everything above it is on support as well four observations below it? I want t specify the boundaries of support in kmatch, but it appears that it requires a continuous interval of support to my understanding. What should I do?

[Bill Smith]

I'm still struggling with how this could happen. Does the fact that .68 is off support suggest there is no match in the controls?

As far as the solution is concerned, in kmatch I could specify the interval of common support (.01 .77) but limit the analysis to only cases where common support=1. Thoughts?

[FernandoRios]

Hi Bill
Perhaps if you do a tabulation by common support and treatment, may help you understand why those observations are not in common support.
Perhaps there is simply no one in the treatment group with a pscore close to 0.68.
While i do not think there is reason to worry about it, in terms of the solution, it is always smart doing what you suggest, so you have a bit more control in case there is something off with the -psmatch2- program.
Best
Fernando