Announcement

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

  • Propensity scores for three treatment groups

    I am interesting in building a propensity-score model in which there are three treatment groups. The three treatments are "equivalent," in that none is patently an "intervention" or "control" group - they're just three interventions. I scanned earlier Statalist posts, and while I was able to find small pieces of information on this topic, those pieces were not sufficient for me to visualize the entire process.

    I am able, for example, to use -mlogit- and produce, for each subject, the probability of being in each of the three treatment groups -- but where do I go from there? It would be ideal to have a well-written, easily-accessed on-line tutorial that illustrates the running of a three-group propensity model from beginning to end.

    Also, the three groups in my data set are of badly unbalanced size -- the respective numbers of cases in the three groups are about 250, 150, and 46. Would the small size of that last group make a propensity-score model impractical from the start?

    Thank you.

  • #2

    Hi Larry,
    In the Stata 13 manual, you may find the section -teffects multivalued- useful.
    It allows you to use inverse probability weighting to create weights for each treatment group.
    The number of cases in each group is not necessarily a limiting factor, but a small n may make it more difficult to achieve balance in observed covariates.

    Hope this helps,
    Melissa

    Comment


    • #3
      Thank you, Melissa. Actually, I'm using Stata 12, but I believe I can get access to the Stata 13 on-line manual. At the moment it is hard to visualize how I could run a model with three different "weights" for each case -- a probability for group A, a probability for group B, and a probability for group C -- but I presume the Stata 13 entry for "teffects multivalued" will explain it!

      Comment

      Working...
      X