Measuring and Modeling Human Capital : Confirmatory IRT, Poor-Proxy Bias, and Latent Convection

Abstract

This dissertation develops and applies sophisticated Item Response Theory (IRT) methods to address fundamental measurement challenges in cognitive testing, focusing on the Armed Services Vocational Aptitude Battery (ASVAB) data from the National Longitudinal Survey of Youth (NLSY). The first chapter implements a confirmatory multidimensional IRT model that accounts for both general cognitive ability and domain-specific skills while incorporating survey weights and handling missing responses. This hierarchical framework aligns with modern theories of cognitive ability and provides practically useful estimates for labor market analyses, offering substantially more information than traditional approaches based on raw scores or principal components. The second chapter applies these IRT methods to investigate temporal changes in returns to education and cognitive skills between the 1980s and 2000s, addressing methodological challenges in comparing paper-and-pencil tests (NLSY79) with computerized adaptive testing (NLSY97). By constructing IRT-based scale scores for the paper-and-pencil test rather than converting adaptive test scores to paper-and-pencil equivalents, this approach reduces measurement error and explains between 5% and 32% of previously observed temporal gaps in returns to cognitive skills and education. While poor-proxy bias accounts for some of the temporal changes in wage returns, the fundamental pattern of declining returns to cognitive skills alongside rising returns to education remains robust, thus suggesting that measurement issues explain some but not all of these labor market changes.