Iterations convergence problem with teffects ipwra for cross-sectional data

[Veda Ph Sharma]

Hi everyone,

am trying to run a multi-value treatment effects regression with access to credit sources as the dependent variable named as 'Access_C' (0-no access, 1-access to formal sources, 2-access to informal sources). I am using a set of socio-economic variables as the explanatory variables. However, when I run the teffects ipwra command in Stata 16.1, the iterations are not converging. The outcome variable is Total expenditure on farm inputs (Tot_Expen)

Code:

[ teffects ipwra ( Tot_Expen age gender educ ST SC OBC land_hectare consumer_exp_month if_hh_memb_registorg, poisson) ( Access_C age gender educ ST SC OBC land_hectare consumer_exp_month if_hh_memb_registorg), atet control(0) tlevel(2)]



Tot_Expen, age, land_hectare, consumer_exp_month are continuous variables

gender, educ, ST, SC, OBC, if_hh_memb_registorg are dummy variables


The iterations are not stopping even after 50 iterations and says 'not concave' after each iteration.

[CODE]
[Iteration 0: EE criterion = 6.040e-16 (not concave)
Iteration 1: EE criterion = 2.871e-17 (not concave)
Iteration 2: EE criterion = 5.942e-18 (not concave)
Iteration 3: EE criterion = 2.361e-18 (not concave)
Iteration 4: EE criterion = 2.343e-18 (not concave)
Iteration 5: EE criterion = 2.343e-18 (not concave)
Iteration 6: EE criterion = 2.342e-18 (not concave)
Iteration 7: EE criterion = 2.342e-18 (not concave)
Iteration 8: EE criterion = 2.342e-18 (not concave)
Iteration 9: EE criterion = 2.342e-18 (not concave)
Iteration 10: EE criterion = 2.342e-18 (not concave)]

Could anyone help me figure out the problem?

Thanks in advance.

[FernandoRios]

Change the convergence criteria and or change scale of your dep variable.
alternatively do the analysis by hand. That’s faster but harder to setup

[Jacopo Tramontano]

Hi FernandoRios , I have the same problem as OP. Could you help us develop a "by hand" approach? I am ignorant in GMM.
I also encountered your other reply in a similar thread that suggested limiting the number of iterations, but how? In teffects ipwra I don't see an option. Thank you for any help!

[FernandoRios]

You may want to look into Ben Jann on using influence functions. He has code to replicate most of the teffects estimations
for teffect look into maxiter() option