Ultimatch - Nearest Neighbor, Radius, Coarsened Exact, Percentile Rank and Mahalanobis Distance Matching in one Package

[Thorsten Doherr]

Hi everyone,

for all that are interessted in matching (especially by Mahalanobis Distance)...

ultimatch - Nearest Neighbor, Radius, Coarsened Exact, Percentile Rank and Mahalanobis Distance Matching in one Package

Abstract: ultimatch implements various score and distance based matching methods, i.e. Nearest Neighbor, Radius, Coarsened Exact, Percentile Rank and Mahalanobis Distance Matching. It implements an efficient method for distance based matching like Mahalanobis matching preventing the quadratic increment of the runtime. Matched observations are marked individually allowing interactions between treated and counterfactuals. Different methods can be combined to improve the results and/or to impose external requirements on the matched. Among other control variables, it creates mandatory weights to provide balanced matching results, preventing distortions caused by skewed counterfactual candidate distributions, e.g. overabundance of candidates with the same score or within the same coarsened group.

Installation:

ssc install ultimatch

Cheers,

Thorsten

[Thorsten Doherr]

I have updated the ultimatch.ado file. Here is the update history:

2019.08.31

• Prevent the usage of the options limit and exp for Coarsened Exact matching.

2019.08.26

• Fixed a bug in Mahalanobis matching that caused partially greedy behavior in non-greedy mode.
• Included options euclid and mahalamobis to improve flexibility for distance matching.
• Included option full as sub-option to copy to enforce treated/counterfactual tuples.
• Included option radius to explicitly activate radius matching instead of the implicit activation before.
• Percentile Rank is now considered a general transformation instead of a separate matching method.
• ultimatch now supports radius matching for the distance-based matching method.
• Added an additional example.

2019.04.25 (ssc repository version)

• Initial version.

As you can see, those are some hefty updates.
Because of the cumbersome update process of the ssc repository you can download the files from my Github page:

https://github.com/ThorstenDoherr/ultimatch

Delete all traces of the former ultimatch.ado and ultimatch.sthlp from your local ado directory before copying the new version. Look also into the "u" sub-directory.
I will not forfeit the ssc repositoy, but the Github updates will be more regular and better documented.

[Alexander Rodriguez]

Dear Thorsten,
Lovely package, especially for all us 1990's Premier League tragics who long for the return of the Mitre Ultimatch 18-panel (seriously one of the most beautiful footballs, felt so lovely to put the laces through!).
Cheers

[Thorsten Doherr]

Another update to ultimatch thanks to the valuable comments I got during the STATA conference in London:

• Implemented Standardized Differences in Means (SDM) as additional similarity measurement.
• Non-Clustered standard errors are now robust.

The SDM is independent from the sample size and does not falter as opposed to the T-tests, which may show statistically significant but practically inconsequential differences for large groups.
As a rule of thump, a SDM of 0.2 is considered a "small effect" in the literature (what ever that means). But I suggest to rely on matchings with lower SDMs. SDMs have a sign but that only shows the direction on the discrepancy. The absolute value is of interest.

Of course, the accustomed test statistics are also shown.

You can download it as usual from that GitHub page:

https://github.com/ThorstenDoherr/ultimatch

Happy matching
(and kicking the Mitre Ultimatch 18-panel)

[Thorsten Doherr]

The latest ultimatch version is now on SSC for your convenience:

ssc install ultimatch, all replace

It is identical to the current GitHub version.

[Thorsten Doherr]

An new version of ultimatch is available on SSC. Main features are:

• Distance-based matching (eucledian, mahalanobis) now supports the draw option (drawing multiple neighbors).
• Matching results can be reproduced by setting a specific seed (set seed).
• Fixed a bug in score-based matching regarding the combination of "copy" and "single".

ssc install ultimatch, all replace

[Thorsten Doherr]

The latest version of ultimatch (2024.12.02) is avilable on SSC.
It additionaly supports cosine and haversine distance matching. The latter is useful to identity geographic neighborhoods in geocoded data identitfied by lattitude and longitude. In combination with the radius option, it is possible to match a spatial neighborhood to a treated observation.
The hypersphere leeway algorithm for distance based matching is further improved to keep the runtime at tolerable levels.