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

  • Use of pstolerance with teffects aipw

    Hi everyone,

    I'd like to use the IPW estimator in my model, with treatment effects.
    The treatment is a dummy variable, but the outcome is a continuous variable which can take negative and positive values (it's a growth rate).
    I have a panel dataset of 28 countries, for the periode 1991m1 - 2019m6 (I have a monthly frequency).
    When I use teffects aiwp (or teffects ipw), I get this:
    "treatment 8 has 636 propensity scores less than 1.00e-05, treatment overlap assumption has been violated; use option osample() to identify the overlap violators"
    So I tried :
    - To use pstolerance: same error when I apply a pstolerance of 1e9. When I choose a pstolerance of 1e15, I have something like this:
    Iteration 0:   EE criterion =  6.548e-13  
    Iteration 1:   EE criterion =  4.803e-20  (not concave)
    Iteration 2:   EE criterion =  1.363e-20  (not concave)
    Iteration 3:   EE criterion =  2.963e-21  (not concave)
    Iteration 4:   EE criterion =  1.019e-21  (not concave)
    Iteration 5:   EE criterion =  5.906e-22  (not concave)
    Iteration 6:   EE criterion =  3.443e-22  (not concave)
    Iteration 7:   EE criterion =  2.236e-22  (not concave)
    Iteration 8:   EE criterion =  1.690e-22  (not concave)
    Iteration 9:   EE criterion =  1.287e-22  (not concave)
    Iteration 10:  EE criterion =  1.019e-22  (not concave)
    Iteration 11:  EE criterion =  2.569e-23  (not concave)
    Iteration 12:  EE criterion =  2.115e-23  (not concave)
    and it never stop.
    To use osample(newvar): I have obtained this: treatment 8 has 636 propensity scores less than 1.00e-05, treatment overlap assumption has been violated by observations identified in variable osample(newvar)
    It's not a very common issue (I didn't see a lot of topics about this subject), and I don't know how to solve it.
    Maybe it is because my treatment doesn't appear often: I have 6888 observations and the dummy takes the value of 1 only 48 times.

    Thanks in advance for your answers