Shapley decomposition: iop command

[Bham Paolo Candio]

Hello,

I am working on a distributional analysis of excessive alcohol consumption (dependent variable) and trying to obtain the Shapley values (r2) for the covariates.

I have first used the rifhdreg command to estimate the coefficients (betas) at the 50th - rifhdreg excess sex origin limitill , rif(q(50))- , 75th and 95th quintiles and then wanted to used the iop command to obtain that contribution to the explained variance by the specified covariates across those three quintiles. The issue is that iop does not seem to be designed for unconditional quantile analyses. Am I right in thinking so? Is there any solution or experience (STATA code) you can share? Any help with this will be very much appreciated.

Thank you
Paolo

[Alexia Muller]

I have the same question. Have you solved the problem?

[Joseph Luchman]

Shapley value decomposition of an R² for independent variables for most models is possible using the -domin- or -domme- software on SSC (see this article for a discussion of both).

I am not familiar with the -rifhdreg- command myself nor am I familiar with the fit statistics it produces but, as is noted in the linked article, so long as the command follows standard depvar indepvars formulation (for -domin-) or accepts constraints (for -domme-), it should be possible to decompose an R² using the Shapley method (note that in the command the Shapley values are called "General Dominance Statistics").

- joe