Estimating models of learning in individual decision making with an application to youth smoking.

摘要

In the first chapter of my dissertation, I examine the dynamics of youth smoking behavior using a model of rational addiction with learning. Individuals in the model face uncertainty regarding the parameters that determine their utility from smoking. Through experimentation, individuals learn about how much they enjoy smoking cigarettes as well as the effects of reinforcement, tolerance, and withdrawal. The addition of learning to the dynamic optimization problem of adolescents provides an explanation for the experimentation of the non-smoker. I estimate the parameters of the model using data from the National Longitudinal Survey of Youth 1997 and compare the overall fit of the model to the model without learning. The estimated model is also used to analyze the effect of cigarette taxes and anti-smoking policies. I find that the model with learning is better able to fit the observed data and that an increase in cigarette taxes are not only effective in reducing the level of youth smoking, but can even increase welfare for some individuals.;In the second chapter (with Jonathan James), we show how the conditional choice probability (CCP) estimation procedure of Arcidiacono and Miller (2011) can be extended to feasibly estimate structural learning models. Although the focus of the paper is the specific application to learning models, the procedure could be used to estimate any model with continuous unobserved heterogeneity. Monte-Carlo simulations show that the CCP method can provide significant computational savings relative to Simulated Maximum Likelihood.;In the third chapter (with Forrest Spence), we investigate whether an individual's subjective price beliefs reflect the empirical distribution of prices and whether an individual learns about features of the price distribution through experience in the market. We use data on subjective price beliefs from a survey of 1,224 college students, and find that inexperienced individuals tend to expect online prices to be higher than what is observed empirically. However, consumers with more experience in the marketplace generally have more accurate beliefs about the price distribution, which is consistent with learning.