Hi Statalist,

I am trying to create a matched cohort of patients based on a single lab value (CEA). I am looking to see if Black patients have better outcomes than white patients with colon cancer. The Black patients have higher baseline CEA in my cohort, so I want to identify a 1:1 (or 2:1 if the match can still be pretty good) pairing to identify a sub-group of white patients with similar CEA levels as the Black patients.

Based on other statalist threads, I used the following code to get a 1:1 match:

Code:
psmatch2 white_vs_black CEA, noreplacement
To better look at the matching, I then use the following code:
Code:
gen pair = _id if _treated==0
 replace pair = _n1 if _treated==1
bysort pair: egen paircount = count(pair)
When assessing the pairs this way, the pairing is not that close. The mean CEA among the 66 white patients is 1.8 while for the 66 Black patients it is 28. I tried changing my code to

Code:
psmatch2 white_vs_black CEA, noreplacement common caliper (0.01)
and the matches get a little better but not by much.

After several days of trying to figure this out, I went ahead and matched the patients by hand and was able to get the mean CEA for my hand-matched white patient cohort to be 28.1 (very close to the mean of 28 in the Black patient cohort).

What am I doing incorrectly that is preventing Stata from achieving the same matching that I could get by hand?