Using Latent Class Models to Better Understand Reliability in Measures of Labor Force Status

• Activity Number: 614
• Type: Topic Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Bac Tran*+
• Companies: U.S. Census Bureau
• Keywords: Reinterview ; Latent Class Analysis ; First-order Markov ; Second-order Markov ; Mover-Stayer

Abstract

The Current Population Survey (CPS) conducts reinterviews (RI) to measure the reliability of labor force (LBFR) status. Whether or not latent class analysis (LCA) can be used instead of RI, these models are able to identify sources and causes of unreliability as well as errors in its estimation. LCA uses substantially increasing the sample size without conducting RI. Agreed with RI results, we have been using LCA in parallel with RI to estimate the LBFR response error in CPS. We began with a first-order latent Markov model, but the challenge is it assumes unobserved homogeneity that is not always true. Mover-Stayer (MS) model solves the homogeneity problem, and outperformed the first-order model in estimating CPS response error, but if the dependency (second-order) goes beyond the previous state, we may be missing information. This paper presents different error models with second-order.