Teffects aipw

[Ilse van Vliet]

Hi there,
I am using teffects aipw but keep receiving errors and I would be very very grateful if someone can help me out. For convenience I wrote my questions in bold.

This is the command I used:

teffects aipw (anxiety cancer geslacht herkomstgroep c.leeftijd sted dutch, poisson) (treatment c.leeftijd cancer bmi geslacht herkomstgroep sted, hetprobit(leeftijd)), wnls vce(cluster nomem_encr) osample(test1)

- I chose poisson because anxiety is measured on a scale of 1-6 (Stata Guide says poisson should be used for counts or nonnegative outcomes).
- I chose hetprobit because I don't think the error terms are homoskedastic (is there a way I can check this?)
- I chose WNLS because this estimator may be more robust to outcome model misspecification than maximum likelihood.
- I chose vce(cluster nomem_encr) because I have longitudinal data (nomem_encr is personal identifier).

Then I receive this error:

treatment 1 has 174 propensity scores less than 1.00e-05
treatment overlap assumption has been violated by observations identified in variable osample(test1)

Can I just delete observations that violate the overlap assumption?

If yes, this would be my next code:

drop if test1==1
teffects aipw (anxiety cancer geslacht herkomstgroep c.leeftijd sted dutch, poisson) (treatment c.leeftijd cancer bmi geslacht herkomstgroep sted, hetprobit(leeftijd)), wnls vce(cluster nomem_encr)

But then I receive this error:

Iteration 0:   EE criterion =  .05195572  (not concave)
Iteration 1:   EE criterion =    -1.#IND  (backed up)
Hessian is not positive semidefinite
gmm estimation failed

What does "Hessian is not positive semidefinite" mean?

Is there a way I can make this gmm estimation work?