exact matching on teffects

[Aloysius Siow]

Hi:

I am using exact matching and nnmatch in teffects.

In my test data, I have one observation in the treatment group which does not have an exact match. Why does the program terminates rather than estimate the treatment effect dropping that one observation?

Also, I am asking for ATET estimate. Why does the number of observations include all treatment and control observations rather than just the number of treatment observations?

Thanks,

Aloysius

[Mike Lacy]

I posted several years ago on the inconvenience of this feature of -teffects-, and had a quick response from StataCorp. I have not followed up since then to find out if there is a better solution. What I was doing was something like: -teffects .... if (ID != 1234) - In my case, as you will see, I had to repeatedly do this in order to discover all the problem IDs, so I ended up with something
like ... if !inlist(ID, 1234, 5678, 9876, ......) I'd like to think that there is a nicer solution to this now, so perhaps someone else will jump in.

[Aloysius Siow]

thanks. the STATA answer makes sense but as you said, dumb for a large dataset.