Activity Number:

89
 SPEED: Survey Methods, Transportation Studies, SocioEconomics, and General Statistical Methods Part 2

Type:

Contributed

Date/Time:

Sunday, July 28, 2019 : 5:05 PM to 5:50 PM

Sponsor:

Survey Research Methods Section

Abstract #307932


Title:

Bayesian Uncertainty Estimation Under Complex Sampling

Author(s):

Matthew Williams* and Terrance Savitsky

Companies:

National Science Foundation and Bureau of Labor Statistics

Keywords:

pseudoposterior;
credible set;
Stan;
survey sampling;
variance estimation

Abstract:

Complex survey sampling designs typically produce a correlation between the response variables of interest and the survey sampling inclusion probabilities such that the balance of information in the observed sample is different from the underlying population targeted for inference. A data analyst may use a samplingweighted pseudoposterior distribution to estimate the population model on the observed sample. While the pseudoposterior distribution contracts on the true population model parameters, we demonstrate that the scale and shape of the asymptotic distributions are different between each of the MLE, the pseudoposterior and the MLE under simple random sampling. Motivated by the different forms of the asymptotic covariance matrices and the within cluster dependence, we devise a correction applied as a simple and fast postprocessing step to our MCMC draws from the pseudoposterior distribution such that the nominal coverage is approximately achieved for posterior intervals. We implement in R by weaving together functionality from the rstan and survey packages.
