Announcement

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

  • Graph treatment effects by propensity score

    Hi Everyone,

    I am trying to understand some of the heterogeneity in my study in which I am working by matching data. I am estimating the effect of treatment on Y matching on two controls. To estimate the effect I am using the following teffects command:


    Code:
     teffects psmatch (Y) (treatment dist_control dir_control, logit), control(0) //can't use rdist - perfectly predicts in some cases
    However now I would like to see how the treatment effect varies by the propensity score and graph the results. Could someone help me in figuring out how to do this? I am struggling to extract the propensity scores following teffects and am not sure how I'd graph it against the treatment effect.

    Your help would be highly appreciated.

    Thanks,
    John




    Code:
    * Example generated by -dataex-. To install: ssc install dataex
    clear
    input float Y long treatment float dir_control double dist_control
     11501085 1  9   7
     18502764 1  0  21
     45422704 0  3 797
     57362584 1  1 518
      3086949 1  0 203
     33602376 1 18 704
     63366988 0  0 319
     12594366 0  0  82
    103255616 0  4 704
     34128948 0  1 673
     49481792 0  2 508
      6682098 1  9 179
    255873248 0  0 439
     10327239 1 12 128
     50632036 0  4 369
    144097792 0  3 179
     14438661 0  0 497
    154696080 0  2 641
     15464507 0  0 162
      6838070 1  0  15
     44862188 1  1 228
     80270224 0  2 797
     39363636 1  1  17
     37861964 0  5 102
       388532 0  0 155
     21386012 1 11 476
     28154936 0  2 334
    100138848 0  1 317
     48463480 0  6 644
     15977832 1  9 153
     11841544 1  0 276
     30547218 1  1 134
       270700 0  0 509
      8363518 0  2 431
     66027236 0  0 651
      6971777 1  1 518
     66641204 0  7 565
     23864886 0  1 213
     31558040 1  0  31
     11540607 1  0  17
      5095038 1  1 287
      3793614 1  0  28
     19889300 0  6 369
      1844610 1  1 588
      6283742 1  0  23
      8064972 1  1 213
     18655776 0  5 545
     99880736 1  4 446
     13637433 1  5 190
      6995302 1  5 486
    end

  • #2
    Is this what you want?

    Code:
    teffects psmatch (Y) (treatment dist_control dir_control, logit), control(0) gen(nn)
    predict ps*, ps
    predict te ,te
    scatter te ps2

    Comment


    • #3
      Thanks so much , that is what I was looking for.

      Comment


      • #4
        Scott Merryman One follow up here: does the ,te option provide treatment effects? I saw somewhere else that it means technical efficiency and I was not quite sure what that meant.

        Comment


        • #5
          Yes, see -predict- under -help teffects postestimation-. With the stochastic frontier models, -frontier- or -xtfrontier- the te option is the technical efficiency.

          Comment


          • #6
            Thanks for the clarification Scott. Again, your help is much appreciated.

            Comment

            Working...
            X