Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Individual Caliper for Variables Nearest-Neighbour Matching (psmatch2)

    Dear Community,

    I aim to apply nearest neighbour matching using the mahalanobis option in the the psmatch2 package and Stata15. Given the syntax of this package, I can use the option 'caliper' to define the maximum distance between controls (I assume by setting it to a certain standard deviation?). My question relates to the situation where the importance of deviations varies across the different covariate-matching variables. In other words, I'd like to assign a smaller tolerance to certain covariates. Is this anyhow possible in the psmatch2 or any comparable environment? From my point of view, it would have to be anyhow combined with the possibility of including a list of variables. I googled it and checked the forum search, but could not find any corresponding information.

    To provide a minimum working example, here the stata code I am currently working with:
    Code:
    psmatch2 var_treatment, mahalanobis (covariat1 covariat2 covariat3) outcome(dep_var) neighbor(5) caliper(0.3) kernel kerneltype(epan)
    Any help is highly appreciated!

    Best,
    Rob

  • #2
    When you use mahalanobis matching, the distances for all covariates are combined into a single number, the mahalonobis distance. The caliper you specify sets a limit on this distance. Therefore, it is not possible to specify that you only take matches where the distance for a particular variable is limited. I suppose you could get somewhere using cem (coarsened exact matching, ssc) but I wouldn't know a straightforward way to do what you ask in Stata without programming the matching command yourself.

    Comment


    • #3
      Originally posted by Jesse Wursten View Post
      When you use mahalanobis matching, the distances for all covariates are combined into a single number, the mahalonobis distance. The caliper you specify sets a limit on this distance. Therefore, it is not possible to specify that you only take matches where the distance for a particular variable is limited. I suppose you could get somewhere using cem (coarsened exact matching, ssc) but I wouldn't know a straightforward way to do what you ask in Stata without programming the matching command yourself.
      Dear Jesse,

      Thank you very much for this answer. Your explanation regarding the mahalonobis distance is very plausible to me. I will take a look at the cem command and see whether I can find a way to do it in this environment.

      Thanks again and best,
      Rob

      Comment


      • #4
        Let us know if you find something (someone else might have the same problem)!

        Comment

        Working...
        X