I investigate the impact of assessment design on student decision making in the context of a second year economics course. I develop a dynamic structural model of student test taking where students are uncertain about their ability, and have biased beliefs about the returns to studying. In the model, students exert costly effort to earn ordinal test grades centered around a target average. Between tests, students update their beliefs about their ability between tests. Using novel high frequency data collected from a competitive large course setting, I estimate the true returns to studying, students' beliefs about returns to studying, and students' capacity to update under a Biased Bayesian framework. Then I combine the three components to simulate counterfactuals using the dynamic model. I find that returns to study effort are relatively concave, that students overestimate their returns to studying, and that students vastly under-react to test grades when updating. Counterfactuals suggest that biased model perceptions contribute to large increases in study effort, implying that the phenomenon exerts positive influence regarding knowledge acquisition. In addition, assessment design counterfactuals suggest that more frequent, and more objective tests are of benefit to students, as they contribute to a swifter reduction in students' biased beliefs about their own skills.

We study purchase deferral and pricing in digital goods markets using data from PC video game sales on Valve's Steam platform. We document that demand systematically declines prior to predictable sales, consistent with forward-looking consumer behavior. To rationalize this phenomenon, we develop and estimate a structural model of demand with forward-looking consumers and differentiated products, allowing substitution across both time and product. The model shows that firms can use temporary discounts to implement intertemporal price discrimination, segmenting consumers by patience. Quantitatively, we find that approximately 40–50% of consumers are forward-looking, and that discounting increases both consumer surplus and profits for most games. Counterfactuals demonstrate that discounting can outperform optimal uniform pricing. Market structure also plays a key role: competition reduces profits, while centralized pricing increases surplus extraction by internalizing substitution. These results provide an  explanation for high–low pricing in digital goods markets without stockpiling. 

Mainstream economic models assume agents are correctly specified regarding their environment and process information without bias. In many settings, the failure of these assumptions can have a significant impact on learning and choice. To understand these issues better, one must obtain data not only on the actual environment but also on agents' dynamic and potentially noisy perception of the environment. We investigate these issues via first-year students in large mathematics classes who receive test scores throughout the term as signals of their ability and return to effort. These test scores are highly informative, and past test scores correlate highly with future scores. We collect a high-frequency dataset by surveying students in an incentive-compatible manner before and after each test. The dataset is also novel in that we are the first to collect belief data regarding not only students' expected grade but also their belief in the noisiness of the test scores. We find that students overestimate the noisiness of tests by a factor of 2 and update in a biased manner. We conduct a randomized control trial where students are not given additional information but are told about the accuracy of tests. We find students can update their beliefs about the noisiness of the testing correctly, and this leads to improvement in their updated beliefs. This suggests a substantial amount of updating failure due to misperception can be corrected.