Research
My research is in applied econometrics, with a strong computational focus. I study numerical reliability, likelihood-based estimation, duration and survival models, and model specification in longitudinal and repeated-spell data, with current applications to women’s employment duration.
Research Interests and Methods
Applied Econometrics; Computational Econometrics; Generalized Linear Models (GLMs); Duration and Survival Analysis; Longitudinal and Repeated-Spell Data; Likelihood-Based Estimation; Model Specification and Diagnostics.
Job Market Paper
Catastrophic Rounding in the Mixed Multiple-Spell Proportional Hazard Model: The Case of Employment for Younger Women
with Jan Ondrich
This paper studies numerical reliability in the estimation of a multiple-spell Prentice-Gloeckler-Meyer model and applies the framework to employment durations of younger women in the National Longitudinal Survey of Youth 1997. The model accommodates grouped, interval-censored durations and woman-specific unobserved heterogeneity, represented parametrically by gamma frailty or nonparametrically by a Heckman-Singer finite mixture.
The multiple-spell gamma-frailty likelihood contains alternating-sign terms that can suffer severe cancellation in finite-precision computation. The paper develops numerically stable evaluation strategies and shows why optimizer convergence alone need not establish that the accurately evaluated likelihood has been maximized. The empirical analysis also compares a flexible duration-indicator baseline with a parsimonious spell-stratified restricted cubic spline baseline and examines finite-mixture specifications across different numbers of mass points.
Methods: computational econometrics; interval-censored duration models; multiple spells; gamma frailty; finite mixtures; restricted cubic splines; numerical likelihood evaluation.
Draft available soon
Papers
Specification Tests for Women’s Employment Duration: Can One Model Fit All Family Structures?
with Jan Ondrich
Using NLSY97 employment histories, this paper asks whether one duration model fits the same woman equally well as her family circumstances change. It evaluates model adequacy across four family structures—no partner or child, partner only, child only, and both partner and child—using martingale and deviance residuals with the Skillings-Mack test, and examines remaining residual dependence with the Pesaran CD test. The analysis compares specifications without unobserved heterogeneity, gamma frailty, and Heckman-Singer finite mixtures while allowing duration dependence to vary by spell order through flexible spline baselines. Current work also examines alternative measures of children and interactions with partner status to connect specification results more directly to economic interpretation.
Focus: women’s employment duration; changing family structure; repeated spells; unobserved heterogeneity; flexible duration dependence; residual specification diagnostics.
Working Paper
Work in Progress
Spell-Varying Unobserved Heterogeneity
A planned extension examines whether unobserved heterogeneity itself varies with spell order rather than remaining fixed across all employment spells for a woman. This provides a way to study the interaction between flexible duration dependence and flexible heterogeneity in multiple-spell duration models.
