teffects

[Rob Wood]

Hi all,

When using teffects psmatch, for example, how is the reported p-value calculated? Is it based upon a standard test or derived from the t-statistic? I can't find the answer in the documentation anywhere.

Thanks,

Rob.

[Steve Samuels]

The Examples in the "Remarks and examples" section of Manual entry for teffects psmatch label the test statistic "z", which means the test is a z-statistic, consisting of an approximately Normal numerator statistic divided by a denominator standard error and referred to a standard Normal distribution (mean zero, variance 1) for calculation of p-values.

[Rob Wood]

Hi Steven,

Thank you for clarifying this for me.

Thanks,

Rob.

[John Orme]

Hi All,

Does anyone know if the standard error used to calculate this z-statistic (in Stata 15) take account of the matched nature of the samples?

Thanks for any help you can provide with this...John

[Joerg Luedicke (StataCorp)]

Hi John,

Estimation of standard errors in teffects psmatch take into account that the propensity score is estimated. The implementation follows the derivations from Abadie & Imbens (2012). The aspect of having matched sets is not relevant in this context. For further details about implemented methods, please have a look at Stata's [TE] Treatment Effects manual .

Joerg


Reference:

Abadie, A., and G. W. Imbens. 2012. Matching on the estimated propensity score. Harvard University and National Bureau of Economic
Research. http://www.nber.org/papers/w15301.

[John Orme]

Many thanks, Joerg! John.