Why does the number of observation differ after the PSM estimation?

[Go Shimada]

I used attnd command to estimate nearest-neighbor matching. Then, I got the result as follows.


*ATT estimation with Nearest Neighbor Matching method
*(random draw version)
*Analytical standard errors

*---------------------------------------------------------
*n. treat. n. contr. ATT Std. Err. t
*---------------------------------------------------------

* 69 42 0.667 0.160 4.175

*---------------------------------------------------------
*Note: the numbers of treated and controls refer to actual
*nearest neighbour matches


I am curious why the number of observations differs between treatment (69) and control (42). I thought the number of observations balance as it performs matching.


I also checked the STATA seminar document and then found out the document's result also shows the different observation numbers.
https://www.stata.com/meeting/italy1...t14_grotta.pdf

Does anybody know why the observations differ?

[David Radwin]

The Statalist FAQ advises posters, "If you are using community-contributed (also known as user-written) commands, explain that and say where they came from: the Stata Journal, SSC, or other archives." attnd was written by Sascha O. Becker and Andrea Ichino and is available from the Stata Journal (net sj 5-3 st0026_2).

I'm not familiar with this program, but it looks like it uses a many-to-one matching process. If you want to restrict matches to common support, which does not guarantee that the number of treatment and control cases are the same, use the comsup option.