Hi all,
This paper may be of interest to Stan users who are working with spatial data. It uses Stan to model observational error in small area survey data, such as American Community Survey estimates. The paper encourages health researchers to pay a lot more attention to survey standard errors, and spatial autocorrelation, and proposes a model specification to deal with these things together.
The paper is also engaging with an old Census Bureau aim of leveraging spatial information to improve estimates and reduce uncertainty, since the standard errors (SEs) can be quite large. There’s a little bit of probability theory on this in an appendix.
While the article doesn’t discuss it, the supplementary material provides some new Stan and R code for CAR models. Relatively speaking, the model is pretty fast (I’ve been using it for US county data, and sampling finishes in between three and ten minutes depending on other aspects of the model). The code is all in the github repo (link below) with a couple short demonstrations.
Donegan, Connor, Yongwan Chun and Daniel A. Griffith. 2021. “Modeling Community Health with Areal Data: Bayesian Inference with Survey Standard Errors and Spatial Structure” International Journal of Environmental Research and Public Health 18, no. 13: 6856. https://doi.org/10.3390/ijerph18136856.
@article{donegan_2021,
title={Modeling community health with areal data: {B}ayesian inference with survey standard errors and spatial structure},
author={Donegan, Connor and Chun, Yongwan and Griffith, Daniel A.},
journal={Int. J. Env. Res. Public Health},
year={2021},
volume={18},
number={13},
doi={10.3390/ijerph18136856},
pages={6856}
}