Using Double Sampling to Correct Gross Flows for Misclassification Error: Moment-based Inference vs. Likelihood-based Inference

• Activity Number: 216
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Nikolaos Tzavidis*+
• Companies: University of Southampton
• Keywords: measurement error ; missing data ; panel surveys ; EM algorithm ; re-interview surveys ; missing information principle

Abstract

Longitudinal surveys provide a key source of information for analyzing dynamic phenomena. Typical examples of longitudinal data are gross flows between a finite number of states. Sample surveys are, however, affected by nonsampling errors. We investigate the use of double sampling for correcting discrete longitudinal data for misclassification error. In a double sampling context, we assume that along with the main measurement device, which is affected by misclassification error, we can use a secondary measurement device, which is free of error but more expensive to apply. Inference is based on combining information from both measurement devices. Traditional moment-based inference is reviewed and contrasted, under alternative double sampling schemes, with a proposed likelihood-based method that works by simultaneously modeling the true transition process and the measurement error process within the context of a missing data problem. Variance estimation, under both approaches, is discussed. Monte Carlo simulation experiments indicate that the proposed likelihood-based method offers significant gains in efficiency over the traditional moment-based method.