Robust Estimation for the Annual Survey of Public Employment & Payroll Using Mixture of Linear Mixed-effects Models with the MCMC Procedure

• Activity Number: 662 - State, County, and Local Government Statistics
• Type: Contributed
• Date/Time: Thursday, August 3, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Giang Trinh* and Bac Tran
• Companies: Census Bureau and US Census Bureau
• Keywords: Linear mixed-effects models ; Mixture models ; Bayesian Method ; MCMC procedure

Abstract

The Public Sector Sample Design and Estimation Branch uses Horvitz-Thompson, Empirical Best Linear Unbiased Prediction (EBLUP) and Bayesian approach to small area estimation (SAE) for the Annual Survey of Public Employment & Payroll (ASPEP). The EBLUP estimator is based on a linear mixed-effect model (LMM) with errors that are assumed to be normally distributed. In this study we provide a robust estimate for the total number of full-time employees in the ASPEP using Bayesian method for a LMM assuming errors governed by a mixture of normal distributions. We specify the Markov Chain Monte Carlo (MCMC) procedure in order to produce samples for the LMM's parameter space. We then compare our research method to the existing methods being used at the U.S. Census Bureau. The Census of Governments (CoG), Survey of Public Employment & Payroll data of 2007 and 2012 were used for the evaluation of this research.