There’s been a back and forth in these forums for a number of years about how to work with survey data from complex sampling designs (For example Survey weighted regression). There are certainly several different approaches. I encourage those interested to read our recent article. Although we favor the survey weighted pseudo-posterior approach, there are trade-offs between different approaches. Stan is particularly a relevant platform, not only for model fitting, but the algorithmic differentiation provides a tool for non-Bayesians (or traditional survey statisticians) to estimate asymptotic design effects to get consistent point and interval estimates. We highlight the importance of Stan in this article and hope the work encourages non-Bayesians in survey statistics to branch out. We also encourage those comfortable with Bayesian methods to learn about mis-specified likelihoods and their impact on inference.
Here is the ArXiv preprint:
This article is a U.S. Government work and is in the public domain in the USA
3 Likes
@mrwilli thanks for sharing this! I’m looking forward to going through it.
Cool! Looking forward to checking it out. @andrewgelman might be interested too.