In marketing and other social sciences, researchers often use field data to empirically study how people make decisions in naturalistic environments. There are numerous theoretical and practical challenges to doing so, and in this dissertation, I propose methodological approaches to address two such challenges. First, people often make complex decisions that are described in terms of high-dimensional or unstructured variables (e.g., writing text or choosing an assortment from a large set of options) which are difficult to analyze relative to simpler decisions (e.g., binary choices). Second, when analyzing how people's decisions are affected by a major event (e.g., regulatory changes or a global pandemic), events often affect a large population of interest simultaneously, making it difficult to assess the impact of the event relative to a counterfactual where the event did not occur. In Chapter 1, I address the first challenge in the context of non-cooperative games. I develop a novel neural network architecture that enables behavioral analysis of complex games by estimating a game's payoff structure (e.g., win probabilities between pairs of actions) while simultaneously mapping agent actions to a lower-dimensional latent space that encodes strategic similarities between actions in a smooth, linear manner. I apply my method to analyze a unique dataset of over 11 million matches played in a competitive video game with a large array of actions and complex strategic interactions. I find that players select actions that counterfactually would have performed better against recent opponents, demonstrating model-based reasoning. Still, players overrely on simple heuristics relative to model-based reasoning to an extent that is similar to findings reported in lab settings. I find that noisy and biased decision-making leads to frequent selection of suboptimal actions, which corresponds to lower player engagement. This demonstrates the limits of player sophistication when making complex competitive decisions and suggests that platforms hosting competitions may benefit from interventions that enable players to improve their decision-making. In Chapter 2, I address the second challenge, proposing a general and flexible methodology for inferring the time-varying effects of a discrete event on consumer behavior when the event spans the target population being analyzed, such that there is no contemporaneous "control group" and/or it is not possible to measure treatment status. I achieve identification by exploiting the empirical regularity of customer spending patterns across cohorts (i.e., groups of customers who adopted the same product or service at different times), comparing purchasing behavior across cohorts who were affected by the event at different points in their tenure. My method applies nonparametric age-period-cohort (APC) models, commonly used in sociology but with limited adoption in marketing, in conjunction with a predictive model of the counterfactual no-event baseline (i.e., an event study model). I use this method to infer how the COVID-19 pandemic has affected 12 online and offline consumption categories. My results suggest that the pandemic initially drove significant spending lifts at e-commerce businesses at the expense of brick-and-mortar alternatives. After two years, however, these changes have largely reverted. I observe significant heterogeneity across categories, with more persistent changes in subscription-based categories and more transient changes in categories based on discretionary purchases, especially those of durable goods.

Not available / text