Propensity Score Matching with/ without caliper

[Cathryn Knight]

Hello,

I am conducting propensity score matching and am currently in the process of choosing the best matching algorithm. The following commands (one with, one without caliper) both turn out the same results:

psmatch2 DyslexicHybrid, outcome(MCS6_FCENGL00) pscore(pc_pscoreAll) neighbor(1)
There are observations with identical propensity score values.
The sort order of the data could affect your results.
Make sure that the sort order is random before calling psmatch2.
----------------------------------------------------------------------------------------
Variable Sample | Treated Controls Difference S.E. T-stat
----------------------------+-----------------------------------------------------------
MCS6_FCENGL00 Unmatched | 2.64098074 3.05025563 -.409274894 .030343766 -13.49
ATT | 2.64098074 2.99649737 -.355516637 .042629207 -8.34
----------------------------+-----------------------------------------------------------
Note: S.E. does not take into account that the propensity score is estimated.

| psmatch2:
psmatch2: | Common
Treatment | support
assignment | On suppor | Total
-----------+-----------+----------
Untreated | 9,193 | 9,193
Treated | 571 | 571
-----------+-----------+----------
Total | 9,764 | 9,764

psmatch2 DyslexicHybrid, outcome(MCS6_FCENGL00) pscore(pc_pscoreAll) caliper(0.12201792) ne
> ighbor(1)
There are observations with identical propensity score values.
The sort order of the data could affect your results.
Make sure that the sort order is random before calling psmatch2.
----------------------------------------------------------------------------------------
Variable Sample | Treated Controls Difference S.E. T-stat
----------------------------+-----------------------------------------------------------
MCS6_FCENGL00 Unmatched | 2.64098074 3.05025563 -.409274894 .030343766 -13.49
ATT | 2.64098074 2.99649737 -.355516637 .042629207 -8.34
----------------------------+-----------------------------------------------------------
Note: S.E. does not take into account that the propensity score is estimated.

| psmatch2:
psmatch2: | Common
Treatment | support
assignment | On suppor | Total
-----------+-----------+----------
Untreated | 9,193 | 9,193
Treated | 571 | 571
-----------+-----------+----------
Total | 9,764 | 9,764


Does this suggest that the nn matching is happening within the caliper distance anyway?

Thank-you.

[Dimitriy V. Masterov]

You should put your output inside code delimiters to make it easier to read (the octothorpe #).

Could you put up kernel densities of your PS by treatment group? The caliper might be so large that it is a constraint that does not bind at all.

Also, did you listen to Stata's advice and shuffle your data before matching?