K-matching

[Angeliki Skoura]

Hello everyone,

How can I perceive the statistical significance of variable in the k-matching?

(Stata 16.0)

Thanks for your help.
Angeliki

[Felix Bittmann]

Are you referring to kernel matching? Most commands report analytical SEs ans p-values but you can also use the bootstrap. kmatch supports both. Or you mean about the interpretation of the p-value? It is not different than from a regular regression model. Any statistics textbook will do.

[Øyvind Snilsberg]

-kmatch- reports a coefficients table of estimated treatment effects. These coefficients being statistically significant suggest that the treatment has an effect on the outcome based on the matched observations.
In the output you shared in a previous post (https://www.statalist.org/forums/for...1649060-kmatch), the estimated ATT was not statistically significant from 0.
The result from -lincom ATT - NATE- (previous post) simply suggests that the matched treatment effect is different from the naive treatment effect.

[Angeliki Skoura]

Yes i am referring to kernel matching and p-value's statistical significance. I will try it.

Thanks for your helpful advice!

Angeliki

[Angeliki Skoura]

Thank you very much @Øyvind Snilsberg & @Felix Bittmann for your time. I understood it very well.

Your help was invaluable.

Best wishes,
Angeliki

[Angeliki Skoura]

Another question that came up after running the kmatch command:

Code:
"kmatch md publiccomp ta poso_015 (apopbtdtziros2), nate att nn"

Finally, which statistical significance should I focus on ATT or NATE or both together?

Output:

Multivariate-distance nearest-neighbor matching

Number of obs = 660,443
Neighbors: min = 1
Treatment : publiccomp = 1 max = 47
Metric : mahalanobis
Covariates : ta poso_015

Matching statistics
------------------------------------------------------------------------------
| Matched | Controls
| Yes No Total | Used Unused Total
-----------+---------------------------------+--------------------------------
Treated | 218435 0 218435 | 141011 300997 442008
------------------------------------------------------------------------------

Treatment-effects estimation
------------------------------------------------------------------------------
apopbtdtzi~2 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
ATT | -322.0717 661.8257 -0.49 0.627 -1619.229 975.0853
NATE |-1621.53 659.3175 -2.46 0.014 -2913.771 -329.2886
------------------------------------------------------------------------------

Thanks,
Angeliki

[Felix Bittmann]

Be careful, the NATE is just the group difference without adjusting for anything. Its what a t-test reports. You want the ATT.

[Angeliki Skoura]

Thank you so much!
Angeliki