Job Market Paper

“Determinants of Electric Vehicle Adoption: Evidence From the California Vehicle Survey”

Abstract: This study investigates the determinants of electric vehicle (EV) adoption using a Bayesian sample selection model applied to stated preference data from the 2017 California Vehicle Survey. We address potential self-selection bias by distinguishing between current EV owners and non-owners in their future vehicle choices. Employing Markov chain Monte Carlo methods, we estimate the differential impacts of vehicle attributes and consumer characteristics on EV purchase decisions for these two groups. Our findings reveal that while both EV owners and non-owners are sensitive to vehicle prices and operating costs, EV owners exhibit greater price elasticity. Socioeconomic factors, including gender, education, and housing type, are found to significantly influence EV adoption. Counterfactual policy simulations indicate that reducing EV prices would be more effective in promoting adoption compared to increasing gasoline taxes or gasoline vehicle prices. This paper contributes to the literature by introducing Bayesian techniques for policy effect estimation and model selection in discrete choice settings. The results provide valuable insights for policymakers and manufacturers in designing targeted strategies to accelerate EV adoption across different consumer segments.

Working Papers

"What Determines Demand for Organic Food: A Multivariate Ordered Probit model"

Abstract: This study explores the determinants of organic grocery consumption in three general categories, including deli and dairy products, dry grocery and frozen food products, and fresh produce. Evidence from the Multivariate Ordered Probit model provide insights into the varying impacts of the same demographic factors on organic purchase decisions across categories in terms of direction and magnitude.

"Simulated Likelihood Estimation and Comparison of Methods"

Abstract: Analysis of limited dependent variable (LDV) models often encounters difficulty evaluating an agent’s choice probability. Despite special cases such as multinomial logit, nested logit, and ordered logit specifications, the integrals of agent’s choice probabilities in most discrete choice models have no closed-form expression. This paper introduces Markov Chain Monte Carlo(MCMC) and hybrid methods that combine the MCMC with classical methods into likelihood estimation for discrete choice models. This study aims to show that in addition to their continuity and differentiability in parameters, the MCMC methods also exhibit computational efficiency compared to classical estimation methods.