Intra - and Intergenerational Transmission of Wage Underpayment

Abstract

This thesis examines wage underpayment using over 54 years of longitudinal data from the Panel Study of Income Dynamics (PSID). Drawing on a primary analytical sample of 3,262 workers observed for at least 15 career wage years between 1985 and 2023, and a matched intergenerational sample of 1,532 parent-child pairs, the study investigates whether wage underpayment is structurally determined, persists within careers, and transmits across generations. To measure underpayment, the thesis develops a multi-method framework combining a classical OLS benchmark, out-of-fold machine learning prediction using Elastic Net, Random Forest, and XGBoost, and Stochastic Frontier Analysis, applied across four classification thresholds. The two-way fixed effects dynamic panel model finds that underpayment exhibits genuine withincareer persistence, with an SFA persistence parameter of 0.286 in the primary cohort that rises to 0.306 among workers with at least 20 career wage years, indicating that underpayment becomes more entrenched rather than self-correcting as careers progress. Transition matrices confirm that all three wage states, underpaid, fair paid, and overpaid, are sticky, with underpaid workers facing a 45.7% probability of remaining underpaid in the following period. The intergenerational analysis finds a transmission coefficient of β=0.065 under SFA, stable across five measurement models and three cohort definitions, indicating that parental career underpayment significantly predicts child career underpayment after controlling for the child's own productive characteristics. Union membership among second-generation workers reduces the intergenerational transmission to zero, while Black children show near-zero transmission differentials at stricter thresholds, consistent with structural discrimination crowding out the parental channel. By embedding wage underpayment within a dynamic and intergenerational framework, this thesis advances our understanding of how labor market failures compound within careers and across generations in ways that standard income-level approaches cannot detect.