Sensitivity analysis after -teffects ipwra-?

[David Torres]

Hi,

I'm using Stata 14 and have modeled the effect of a treatment on an outcome using the -teffects- command. In particular I'm using the inverse probability-weighted regression adjustment (ipwra) option. What I'd like to know is if there's a way to conduct a sensitivity analysis after running my model?

For more context, here is my code:

teffects ipwra (eng1score1516 c_readscore1415 black hispanic asian othrace male agecentered disadvantaged yesatrisk yeslep yesspecial yesmagnet yesgifted yeshomeless black29pct econ73pct atrisk62pct cad_gap cad_progress cad_post cad_read cpemallp ir_school) (treat black hispanic asian othrace male disadvantaged black29pct ir_school agecentered), vce(robust) ate

Iteration 0: EE criterion = 1.056e-18
Iteration 1: EE criterion = 2.299e-25

Treatment-effects estimation Number of obs = 10,347
Estimator : IPW regression adjustment
Outcome model : linear
Treatment model: logit

Robust
eng1score1516 Coef. Std. Err. z P>z [95% Conf. Interval]

ATE
treat
(1 vs 0) 58.09149 11.23186 5.17 0.000 36.07744 80.10554

POmean
treat
0 3931.306 6.129642 641.36 0.000 3919.292 3943.32

[Felix Bittmann]

I would like to push this unresolved question up. If there is a multivalued treatment, how can we conduct sensitivity analysis? Looking for ados like regsensitivity, psacalc, sensatt, or sensemakr. These all work only with binary treatments.

[Jeff Wooldridge]

Felix: I would think the main effects of interest are ATE(g) or ATET(g), where g is the treatment level. These are with respect to the control state, so Y(g) - Y(0). IPWRA is the same as a bunch of binary treatment comparisons, comparing g with 0. So I think you can apply the tools for binary treatments. It’s a bit clumsy, I suppose.