A Microlevel Latent Class Model for Measurement Error in the Consumer Expenditure Interview Survey

Previous research by Tucker et al. (2005) and Tucker et al. (2006) attempts to identify a latent construct that predicts the amount of measurement error in expenditure reports on the Consumer Expenditure Interview Survey (CEIS). While this work was successful in identifying a construct that predicts measurement error in expenditure reports, it is more sensitive to falsely negative reports of the entire purchase than it is to the underreporting of the amount of expenditure for that purchase. Current research focuses more deeply on the underreporting of expenditure amounts for a number of different commodities. Together, with previously explored indicators such as, the number of contacts, missing on the income question, the length of the interview, and the use of records, we examine new indicators describing the experience of the CU throughout the panel, resulting in a new latent construct. Although we find that our newly developed latent constructs have strong validity, appearing to measure report quality, it does little to explain either the overall level of expenditure or the difference in expenditure reports between interviews.