Measuring Mobility

Abstract

I revisit intergenerational mobility in the United States with special attention to biases from imperfectly linked census data. Record linkage errors tend to attenuate ordinary least squares (OLS) estimates such as the association between the status of parents and children, thereby exaggerating levels of mobility. I propose two methods for addressing bias due to imperfectly linked data. In chapter 1, I propose a maximum likelihood estimator for a model of occupational misclassification (MCML), building on results in the non-classical measurement error literature. The MCML estimator reduces bias in estimates by between two-thirds and three-fourths in settings similar to historical census record linkage. In chapter 2, I apply the MCML estimator to re-examine intergenerational mobility in the U.S. between 1850 and 1910. MCML estimates of the intergenerational elasticity of occupation status (IGE) are 20% to 30% higher than OLS estimates and relatively stable over time, suggesting that previous estimates of the IGE overstated historical levels of mobility. In chapter 3, I develop a Bayesian approach to estimation and inference with imperfectly linked data. Modeling the data generating process for imperfect record linkage allows for uncertainty quantification, though I find that estimates remain highly sensitive to assumptions about the linking process. Taken together, the results in this thesis highlight that qualitative conclusions in and beyond economic history may depend on assumptions about linked data.