I am new to matching test/control. I saw that in R, we could use either Matching or MatchIt to select matches. What are differences between the two ways to create matches? Could you provide examples for each of them?
Thanks!
There are a number of differences between MatchIt and Matching for performing matching to reduce confounding in observational studies. The most important is the philosophy underlying each package: MatchIt considers matching to be a form of nonparametric preprocessing as described in Ho et al. (2007), while Matching consider matching to be a specific estimator of a difference in means, as described in Abadie and Imbens (2006). This distinction manifests in that Matching provides effect estimates after matching, whereas MatchIt only provides a matched dataset that can be supplied to a model to estimate the treatment effect.
There are many other differences between the two packages. MatchIt provides an interface to a large variety of matching methods, including nearest neighbor matching, optimal pair matching, optimal full matching, (coarsened) exact matching, genetic matching, propensity score subclassification, and cardinality matching, some of which rely on other packages (including Matching!). Matching only performs nearest neighbor matching and genetic matching.
The implementation of genetic matching in both packages is about the same because MatchIt calls functions in Matching to perform the genetic matching. By default, MatchIt performs genetic matching on the covariates and propensity score, whereas, in Matching, the propensity score must be manually estimated and added along with the covariates. There are some other slight differences that are generally not too relevant.
The implementation of nearest neighbor matching is where the two packages differ the most, though in most cases it is possible to replicate the results of one with the other. The key differences are in the defaults; most capabilities are the same between the two packages. One difference is that Matching can perform nearest neighbor matching with ties, whereas MatchIt does not allow ties. With numerous continuous covariates, ties rarely occur anyway. Below are some of the differences between the packages:
MatchIt is nearest neighbor matching without replacement on the propensity score; the default in Matching is nearest neighbor matching with replacement and ties on the scaled Euclidean distanceMatching, the matches occur in the order of the data; in MatchIt, the user can select the matching order (the default is in descending order of the propensity score)Matching, the ATE can be chosen as the estimand when matching with replacement; in MatchIt, only the ATT and ATC are availableMatching uses very fast C code to perform the matching; MatchIt uses less optimized Rcpp codeMatching allows you to customize the matrix used to weight the importance of each covariate in the matching; MatchIt allows you to supply a custom distance matrix itselfThere are a few other minor differences in options, but the other biggest differences come in the interface. I find MatchIt significantly easier to use, especially for beginners, because the interface is similar to that of other R functions like lm(), whereas Matching requires a bit more manual programming and knowledge of R to use all options effectively. The documentation of MatchIt is also significantly more detailed and extensive (note that I wrote it), which can make it easier for beginners. MatchIt also provides more tools for visualizing balance.
Which should you use? I say MatchIt if you are a beginner or want many options, and Matching if you are an econometrician who wants to use the specific estimator Matching implements. The interface of MatchIt and the ability to use a whole suite of matching methods, along with its extensive documentation aimed at beginners, make MatchIt a better choice for those just getting started in my opinion.