New command (-vbkw-) for propensity scores with multiple treatments

[Melissa Garrido]

-vbkw- is now available on SSC (thank you to Kit Baum!)

This command allows you to implement vector-based kernel weighting (VBKW) in propensity score analyses involving categorical treatment variables.
VBKW is is more robust to propensity score model misspecification than the traditional IPTW approaches.

More info about VBKW here: https://www.hsrd.research.va.gov/for...3970-notes.pdf and https://onlinelibrary.wiley.com/doi/....1002/sim.8836

[John Matrix]

Hello,
Can anyone help me with my code, please? I am trying to use propensity score matching for a complex survey data - Medical Expenditure Panel Survey, where I should use sample weights/svyset.
Any idea how I can accomplish that? Stata is giving me an error message when trying weights:

teffects psmatch ( ttlpx ) (time age female white) [pweight=weight]
pweights not allowed
r(101);

[Stefan Sliwa]

Hi Melissa, thanks for that (and to Kit Baum!)
A quick question on your command. I see that vbkw only allows for setting with categorical treatment.
In my setting, I first have to weight+match on a binary treatment variable (Treatment - Control), before I disentangle further (High intensity treatment - Low intensity treatment - Control). For the latter regression, I was thinking of applying vbkw for pre-processing and creating common support between the three treatment groups. Which weighting+matching approach would you recommend me for the binary treatment? In other words, which would be closest to the underlying mechanics of vbkw, so that I can reasonably use both approaches on the same dataset?

Thank you!

[Melissa Garrido]

For a binary treatment, you can use kernel matching (it implicitly assigns kernel weights to the vector of pscores, which only includes p(no treatment) and p(treatment)). There's an option to do so under the user-written psmatch2 command.