This dissertation studies how behavioral frictions shape consumer choice and policy evaluation in markets where product attributes, health risks, and regulatory environments are difficult for consumers to process. Across three essays, I develop and apply discrete choice models that separate standard preference heterogeneity from mechanisms such as stigma, internalities, limited consideration, and imperfect attention. The empirical applications focus on food safety and tobacco regulation, two settings in which consumer choices have important welfare and policy implications and in which observed demand may not directly reveal stable underlying preferences.The first essay studies the demand for Fukushima rice after the 2011 Fukushima nuclear power plant accident. Following the accident, consumers and retailers sharply reduced demand for rice labeled as originating from Fukushima prefecture, even though objective food-safety risk was limited by inspection and regulation. Using scanner data from Japanese retail stores, I estimate a structural model of demand and supply that allows the observed demand decline to operate through three channels: a change in perceived utility, limited consideration of Fukushima rice, and retailer assortment decisions. The estimates indicate that the demand penalty is driven primarily by consumer stigma. Approximately 90 percent of the decline is attributable to changes in perceived utility, while retailer assortment accounts for roughly 10 percent. Limited consideration appears to play a minimal role. Counterfactual simulations suggest that price discounts targeted only to Fukushima rice are insufficient, but broader discounts that also include blended rice could substantially recover demand. The second essay develops a stated-choice approach to measuring smoking internalities and applies it to the welfare analysis of graphic warning labels. In a discrete choice experiment, adult smokers choose among cigarettes, e-cigarettes, and quitting both for immediate consumption and for future consumption. Interpreting immediate choices as decision utility and future choices as a proxy for experienced utility, the analysis estimates an average smoking internality of $3.88 per pack, with substantial heterogeneity and alternative estimates ranging up to $19.12. The essay then incorporates these internality estimates into a cost-benefit analysis of graphic warning labels. The implied net welfare effect ranges from negative to substantially positive depending on the maintained internality measure. The results therefore do not support a robust conclusion that graphic warning labels are welfare-improving on average and suggest that the policy may be poorly targeted, generating larger welfare losses for smokers with smaller internalities or stronger self-control. The third essay examines imperfect attention in discrete choice experiments. Standard DCE models typically assume that respondents process all experimentally varied attributes, but this assumption may be inappropriate when tasks involve complex or unfamiliar policy scenarios. I develop a measurement-augmented latent-attention model that combines panel choice data with noisy post-experiment recall measures. The model separates preference heterogeneity from heterogeneity in attribute processing and is estimated using a generalized expectation-maximization algorithm. Applying the framework to a menthol tobacco regulation experiment, I find that attention-adjusted models improve in-sample fit and materially change structural interpretation. In particular, once incomplete processing is modeled explicitly, the estimated disutility of illegal menthol markets becomes substantially larger. Counterfactual simulations predict less quitting and greater persistence of menthol demand under prohibition than a benchmark full-attention mixed logit model. Together, the three essays show that policy-relevant demand estimates can be misleading when behavioral mechanisms are left implicit. Stigma can masquerade as a standard product preference, internalities can alter welfare conclusions even when choices are well measured, and imperfect attention can attenuate estimated responses to experimentally varied policy attributes. Accounting for these mechanisms changes both the interpretation of observed choices and the evaluation of counterfactual policies. The dissertation therefore contributes to behavioral industrial organization, health economics, and applied discrete-choice modeling by showing how richer behavioral structure can improve empirical policy analysis in regulated markets.