‘nnmatch’ vs ‘teffects nnmatch’

[sang baum kang]

I am trying to use average treatment effect (ATE) and average treatment effect on the treatment (ATET) using multivariate matching technique as explained by the following article published in The Stata Journal.

Abadie, Alberto, David Drukker, Jane Leber Herr, and Guido W. Imbens. "Implementing matching estimators for average treatment effects in Stata." The Stata Journal 4, no. 3 (2004): 290-311.

(I also attach a pdf file of the paper to this posting.)

According to the article, I should use nnmatch command. However, I do not have nnmatch command in my version of Stata (version 14.2). Instead, teffects command has nnmatch subcommand.

With that said, my question is as follow. Does ‘teffects nnmatch’ in version 14.2 implement the calculation of ATE and ATET in a precisely same way as described in the aforementioned paper (i.e., Abadie, Drukker, Herr, and Imbens, 2004)?

I ask this question because Stata Treatment-Effects Reference Manual does not have details about how nnmatch works but Abadie, Drukker, Herr, and Imbens (2004) does. That is why I want to make sure that ‘teffects nnmatch’ in my current version of Stata (version 14.2) is the same as ‘nnmatch’ as described in Abadie, Drukker, Herr, and Imbens (2004).

Thank you for considering my question.

[David Radwin]

According to the Stata 17 PDF documentation, "teffects nnmatch . . . was previously implemented in Stata as discussed in Abadie et al. (2004)."

When I compared results, the point estimates from teffects nnmatch in Stata 17 match the first two examples in the Stata Journal article, but the standard errors are different. I am not sure why there is a discrepancy. Maybe someone else can explain.

. version
version 17.0

. use artificial

. teffects nnmatch (y x) (w)

Treatment-effects estimation                   Number of obs      =          7
Estimator      : nearest-neighbor matching     Matches: requested =          1
Outcome model  : matching                                     min =          1
Distance metric: Mahalanobis                                  max =          2
------------------------------------------------------------------------------
             |              AI robust
           y | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
ATE          |
           w |
   (1 vs 0)  |   .1428571   .9709082     0.15   0.883    -1.760088    2.045802
------------------------------------------------------------------------------

. use ldw_exper

. teffects nnmatch (re78 age educ black hisp married re74 re75 u74 u75) (t), nneighbor(4)

Treatment-effects estimation                   Number of obs      =        445
Estimator      : nearest-neighbor matching     Matches: requested =          4
Outcome model  : matching                                     min =          4
Distance metric: Mahalanobis                                  max =         11
------------------------------------------------------------------------------
             |              AI robust
        re78 | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
ATE          |
           t |
   (1 vs 0)  |   1.903716   .7477776     2.55   0.011     .4380988    3.369333
------------------------------------------------------------------------------