Job Market Paper
Count outcomes in economics are often top coded, and the negative binomial model assumes a Gamma distribution for the unobserved heterogeneity. Under censoring, a wrong heterogeneity distribution makes the slope estimates inconsistent. I develop sieve maximum likelihood estimation for the censored mixed Poisson regression model, with the heterogeneity density represented by wavelets or a shallow neural network with an adaptive activation function. I prove identification, consistency, and convergence rates. Under censoring the intercept has zero efficient information, while the slopes have positive information when the censoring threshold is at least two, and the slope estimator is asymptotically normal at the semiparametric efficiency bound under two high level maintained conditions. In simulations, the sieve estimators give valid inference for the slopes and correct the bias of the negative binomial model when the heterogeneity is not Gamma. In an application to doctor visits in the RAND Health Insurance Experiment, the sieve estimators fit the censored counts better than the negative binomial model on held-out data, and the estimated heterogeneity distribution has two groups, which a Gamma distribution cannot capture.
Presented at:
York University, Econometrics Seminar Series, November 2026 (scheduled)
Canadian Economics Association (CEA) Annual Meeting, Vancouver, 2026
York University, Econometrics Seminar Series, 2025
Canadian Economics Association (CEA) Annual Meeting, Montréal, 2025
Canadian Econometric Study Group (CESG), Ottawa, 2025 (poster)
Working Papers
“Semiparametric Estimation of Count Regression Models Revisited” (with Paul Rilstone)
This paper revisits the semiparametric approach of Gurmu, Rilstone and Stern (1999) to count regression models with unobserved heterogeneity, for counts observed without censoring. The regression parameters and the heterogeneity density are estimated jointly by sieve maximum likelihood with Laguerre and shifted Legendre sieves. We give conditions for identification and derive consistency, convergence rates, and asymptotic normality of the regression parameters. Simulations and applications to health care utilization data show how the estimators perform in practice.
Presented at:
Canadian Economics Association (CEA) Annual Meeting, Montréal, 2025
30th International Panel Data Conference, France, 2025 (presented by co-author)
Canadian Econometric Study Group (CESG), Ottawa, 2025 (poster)
Research in Progress
“Poverty Shocks and the Frequency of Intimate Partner Violence in Canada: A Semiparametric Count Data Approach”
Applies the sieve estimators of my job market paper and working paper to right censored counts of violent incidents in the General Social Survey on Victimization.