Adjusting for survey weighting in Stan


I am curious about whether there is any reference for how to apply the idea of weight adjustment in sample survey data into a Bayesian missing data imputation workflow in Stan? Currently we are trying to apply the multivariate censored regression model (without any hierarchical structures) to impute some missing values in the survey data.


Sorry for letting your question fall through. The expert on surveys would be either @lauren or @andrewgelman. I think (I don’t understand these questions much) that the recommendation might involve using “multilevel regression with post-stratification” (i.e. estimate the response for all subgroups and then use demographic data in the prediction phase to simulate appropriate number of responses from each subgroup; hope I am not misrepresenting that) which should IMHO fit reasonably well with the imputation you already have.

Does that make sense?

1 Like

Yes and thanks so much for your advice!

I would read more about MRP.