Responsibility Konstantin Korotkiy. Publication [Stanford, California] : [Stanford University], 2023. Copyright notice ©2023 Physical description 1 online resource.
Items from this collection are featured in:
Author/Creator Korotkiy, Konstantin, author. Contributor Sahni, Navdeep, degree supervisor. Thesis advisor Kim, Yewon (Assistant professor of marketing), degree committee member. Thesis advisor Narayanan, Sridhar, 1970- degree committee member. Thesis advisor Stanford University. Graduate School of Business.
Summary This dissertation focuses on understating and modelling seemingly irrational consumer choices. Through empirical analysis, it demonstrates significant deviations from the economically rational behavior and offers structural models to rationalize and predict such behaviors along with the statistical tools to estimate key effects of interest. This dissertation consists of three chapters: Chapter 1, Perception of Monetary Gains in Consumer Search, provides an empirical evidence of satisficing behavior and presents a novel structural consumer search model encompassing satisficing behavior. Chapter 2, Inferring Brand Quality From Prices: Empirical Analysis, analyses the results of the randomized field pricing experiment and shows how higher prices might signal lower brand quality, despite a conventional belief of a positive price-quality relationship. Finally, Chapter 3, Improved Uplift Modeling with Covariate Offsetting, offers a novel statistical variance-reduction tool called covariate offsetting designed to yield better estimates of conditional average treatment effect function in randomized field experiments. The first chapter is a joint work with Dr. Navdeep S. Sahni. Consumer choice and the underlying decision-making mechanisms are the core objects of interest in marketing. Yet, it is extremely challenging to study such decision-making mechanisms without making numerous simplifying assumptions, as observed choices are not only a function of such unobserved mechanisms, but also a function of unobserved preferences, beliefs and private information. In this paper, we analyze a unique dataset on consumers facing very simple choice situations in a natural environment. Specifically, these consumers are presented with the chance to receive a 20% discount on any item within their grocery store basket, with no additional implications tied to their selection. Within these choice situations, we have access to all relevant factors guiding consumers' decision-making processes. Consequently, we are able to gain deeper insights into the decision mechanisms utilized by consumers. Notably, our findings provide empirical evidence of satisficing behavior — a phenomenon wherein consumers place less emphasis on potential gains when their current state is deemed satisfactory. Through our modeling, we capture and illustrate this behavior, demonstrating that customers prioritize absolute monetary gains when confronted with simple choice situations, while attaching greater importance to percentage gains when faced with more complex choices. The second chapter challenges a conventional belief that higher prices invariably signal higher quality. Through a series of a randomized field experiments with a boutique sportswear startup Reprise Activewear, we discovered that increasing prices by 30% for products that traditionally have low sales can lead to a decrease in the brand's total sales by nearly 50%. We develop a theoretical choice model to rationalize these findings: in this model, customers are capable of recognizing a case where the disparity in prices clearly is not justified by a corresponding disparity in production costs and product quality, deducing that the brand's profit margins are excessively high at least for some products. As a result, customers update their beliefs about the quality across the brand's entire product line, and if the prices seem to be particularly unjustified, overall demand is likely to decrease. The third chapter presents my joint work with Dr. Harikesh Nair. In this paper we propose a covariate offsetting method that could be paired with any nonparametric conditional average treatment effect (CATE) estimation technique. These techniques usually estimate CATE by regressing (possibly transformed) outcome on observables via supervised learning algorithms such as random forest or neural network. We see covariate offsetting as a tool of reducing the unnecessary variance in the target variable used by these supervised learning algorithms, which in turn results in more stable and reliable CATE estimates. We explicitly show that minimizing the variance in the inverse probability weighted (IPW) outcome using the proposed offsetting method yields the augmented IPW (AIPW) outcome of Robins et al. [1994] and Robins et al. [1995]; hence we provide an alternative interpretation for the AIPW outcomes. Finally, using both simulated and the real world experimental data, we show that offsetting could significantly improve the performance of some of the most popular CATE estimation methods.
Publication date 2023 Copyright date 2023 Note Submitted to the Graduate School of Business. Note Thesis Ph.D. Stanford University 2023.