Methods for Fitting a Markov Latent Class Analysis for the National Crime Victimization Survey

• Activity Number: 628
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Marcus Berzofsky*+ and Paul P. Biemer and William Kalsbeek
• Companies: RTI International and RTI International and The University of North Carolina at Chapel Hill
• Keywords: Markov latent class analysis ; model assumptions ; NCVS ; Data Sparseness ; measurement error ; classification error

Abstract

This paper presents the methods for the first assessment of classification error (measurement error for categorical data) in the National Crime Victimization Survey (NCVS). The NCVS is the only national survey that captures information on victimizations that are both reported and unreported to the authorities. To estimate reporting errors we use Markov latent class analysis (MLCA), a modeling technique that can be used to estimate classification error in panel data that does not require a gold-standard (error-free) measurement. This paper proposes a process by which an MLCA can be conducted on complex survey data, ensuring that all key assumptions are met or corrected for such that parameter estimates are valid. To conduct this analysis, we used a special longitudinal file containing all respondent waves from a sample of NCVS households. We determined that a model with fully constrained transition probabilities and partially constrained classification error probabilities fit the NCVS data the best.