Latent Class Models for Analysis of Response Error and Rotation Group Bias in the Current Population Survey

• Activity Number: 400
• Type: Topic Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Khandaker Mansur*+ and Bac Tran
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: latent class models ; panel design ; unemployment ; rotation group bias ; misclassification errors ; response errors

Abstract

The Current Population Survey (CPS) uses a 4-8-4 rotating panel design. For any given month, the CPS sample can be grouped into eight subsamples corresponding to the eight rotation groups. It is well known that in a rotating panel design survey estimates for the same characteristics from different rotation groups relating to the same time period have different expected values, depending on the length of the time they have been included in the sample. This phenomenon is called rotation group bias. Latent class models are used here for the analysis of response error in the CPS. Latent class analysis (LCA) is used here to estimate the misclassification rates and to assess their impact on labor force estimates. The paper investigates the validity and efficiency of using LCA to estimate rotation group bias, response error and response probabilities. This paper studies response errors in the CPS and assesses their impact on the rotation group bias. The paper also discusses Shockey's results on rotation group bias and attempts to determine why the sizes of his rotation group bias are much larger than those reported by other authors, specifically Bailar (1975) and Mansur (1999) estimates. The paper also compares the response probabilities to those obtained from the Markov Latent Class Analysis presented in Tran (2003).