| The dissertation consists of two essays, related by their use of longitudinal linked data. The first essay is concerned with the role of heterogeneity in determining labor market outcomes. I develop an equilibrium matching model with heterogeneous workers, firms, and worker-firm matches, and apply it to longitudinal linked data on employers and employees. Agents have complete information about worker and firm heterogeneity, and symmetric but incomplete information about match quality. They learn its value slowly by observing production outcomes. I show that under a simple CRS technology, the Nash-bargained equilibrium wage is linear in a person-specific component, a firm-specific component, and the posterior mean of beliefs about match quality. The optimal separation policy is characterized by a reservation level of beliefs about match quality. The reservation value varies across workers and firms, and is monotone in tenure. I apply the theoretical model to data from the Longitudinal Employer-Household Dynamics (LEHD) Program at the US Census Bureau. I estimate both fixed and mixed model specifications of the equilibrium wage function. I recover structural parameters of the matching model, and test a number of its predictions. I find considerable support for the matching model in these data.;The second essay (joint with John M. Abowd) digresses from issues specific to labor markets, and considers the problem of protecting confidentiality in longitudinal linked data. Data combined from several sampling frames for statistical analysis can pose a significant disclosure risk. We draw upon Bayesian multiple-imputation methods for missing data to develop disclosure limitation methods that are sufficiently flexible to accommodate a variety of data structures, that preserve complex dynamic relationships between variables, and most importantly, prevent disclosure of confidential information. The basic idea is to replace confidential data items with multiple values drawn from the posterior predictive density of an appropriate generalized linear model. We consider the case where confidential data can be used to condition the imputation model, and where it cannot. In a series of detailed applications and examples, we show that both masked and simulated data preserve important statistical properties of the confidential data, but contain no confidential data items. |
