Treated VS control observations

[Angeliki Skoura]

Dear Statalisters,

I have a concern about the number of matced observations from my sample. In other words in my research I am looking at listed and unlisted companies on the stock exchange. So in my attempt to make the sample comparable I applied several matching methods such as: propensity score matching- ( kernel matching, nearest neighbor matching , stratification Matching) and local linear regression matching. But I am confused because each of these methods gives me different result for the treated and control observations and spesifically the local linear regression matching has a lot of control observations . To be clear I am showing you the detailed results of each method.
Your advice will be very useful to clarify how many observations I have in each group.
Please let me know if I need to clarify anything further.
Thank you in advance,

Best wishes
Angeliki

(Stata 16.0 MP)

Kernel matching method:

Nearest Neighbor matching method:

Stratification Matching:

Local linear regression matching:

[Jared Greathouse]

I don't see what the problem is. The point estimates and even the standard errors are within the same neighborhood of one another, even if they're not all EXACTLY the same. Anyways, that's not the point.

The reason you're getting these differences is because the underlying guts of the mathematical computations are different. They make different assumptions and were built for different problems, all of which I don't know the precise answer to. I haven't used matching since I was like 19 or 20 4 or 5 years ago, and I certainly didn't interrogate how each of these were different. So as boring as it may sound, this boils down to math. If I were you however, I'd worry more about which sort of matching is appropriate, and if you should even want to match on the propensity score at all (see King's Why Propensity Scores Should not Be Used For Matching). Maybe coarsened exact matching would be ideal for you.

Point is, why the numbers differ isn't nearly as important as which one from a very practical perspective suits your question the best.

[Angeliki Skoura]

Thank you Jared Greathouse for your feedback! I appreciate your help!

Best wishes,
Angeliki