Estimation of government employment using multivariate hierarchical Bayes modeling
Scott, Andrew John
MetadataShow full item record
Small area estimation is a topic of considerable interest for many agencies wishing to use data collected at the national level to accurately estimate statistics speci c to subsets of the general population. Our approach used a lognormal multivariate hierarchical Bayes (MVHB) mixed model to generate improved estimates of small area totals for Census of Governments data drawn through strati ed random sampling (SRS) and skewed data drawn through probability proportional to size (PPS) sampling. Additionally, we explored the inclusion of design weights as model covariates for PPS-sampled data. Gibbs samplers were developed to provide estimates of posterior mean and variance. We show that the MVHB models produced improved estimates with decreased posterior variance, when compared to their univariate counterparts for both the PPS and SRS samples.