Evaluating census data quality using intensive reinterviews: a comparison of U.S. Census Bureau methods and Rasch methods.

"In this paper, we consider Rasch measurement methodology as an alternative to conventional U.S. Census Bureau methodology for evaluating the quality of data obtained using different measurements of the same characteristic. Rather than assuming, like the Census approach, that respondents have true states corresponding to the categories of a questionnaire item, the Rasch approach assumes that measuring instruments and survey respondents vary continuously on one or more common dimensions called latent traits. Unlike the Census approach, the Rasch approach requires that the properties of measuring instruments be invariant across different subclasses of respondents, generating parameter estimates that are sufficient statistics for elementary discrete sampling models. The different conclusions that can be drawn from the Census and Rasch approaches are illustrated by an evaluation of questionnaire items designed to measure the limitation or prevention of work because of physical or mental disability."