Marketing Research Seminar Series

This article provides evidence on the impact of workforce gender diversity on individual-level performance in a microfinance firm in Mexico. Despite increased confidence from recent research in diversity's effects on group-level performance, direct evidence is absent on whose performance within groups increases (more) due to demographic diversity. Leveraging the firm's random transfer policy, we initially do not find evidence of the impact of gender diversity on individual officers' loan collection performance however, further investigation reveals substantial heterogeneity especially by loan officer gender. Male loan officers perform best in gender-balanced branches while underperforming in predominantly male or female branches. Female loan officers, by contrast, show stable performance regardless of the workforce composition. We further explore potential mechanisms behind this impact whether it comes from learning skills for better performance or from getting temporary access to more diverse information in a gender-diverse branch.

Doctoral Student Research Presentations

Welcome and Keynote Address

Introducing a new currency in an existing market offers a potentially influential, alternative currency option—a prediction that prompts the current research into how consumers evaluate cryptocurrencies and the effects on their spending. Nine studies consistently demonstrate that paying with cryptocurrencies reduces purchase intentions, mediated by the pain of payment. However, if a cryptocurrency is positioned as a payment mechanism, consumers' purchase intentions remain similar to those they express for fiat currencies. This mental accounting explanation is confirmed with field data. The authors also demonstrate that the negative impact can be mitigated if the cryptocurrency is introduced as a virtual currency. Finally, the negative impact of cryptocurrencies is mitigated for higher priced products. These results offer important theoretical insights into how payment mechanisms affect purchases, as well as managerial insights about how cryptocurrency providers and retailers can increase uses of cryptocurrency as a payment mechanism.

Self-disclosures, or the voluntary sharing of information about one's vulnerabilities, are common among persuasion agents such as online influencers, entrepreneurs, and marketers. While prior research highlights the 'disclosure-liking effect,' where self-disclosures increase trust and positive evaluations, we investigate the conditions under which self-disclosures may backfire. Drawing on persuasion knowledge theory, we show that self-disclosures by persuasion agents can elicit inferences of manipulative intent (IMI), leading to reduced liking, crowdfunding interest, and purchase intentions. Further, drawing on attribution theory, we demonstrate that perceived controllability moderates the effect of self-disclosures on consumer responses. When the disclosed issue is attributed to causes controllable by the persuasion agent (e.g., overspending), IMI is heightened, reducing liking and consumer support conversely, when the issue is attributed to uncontrollable causes (e.g., genetic diseases), self-disclosures elicit sympathy, enhancing liking. These findings introduce IMI as a novel consequence to self-disclosures in the marketplace and controllability as a key moderator. Substantively, this research speaks to the double-edged nature of self-disclosures, providing insights for persuasion agents navigating the fine line between authentic engagement and perceived manipulation.

Online content-sharing platforms such as TikTok and Facebook have become integral to daily life, leveraging complex algorithms to recommend user-generated content (UGC) to other users. While prior research and industry efforts have primarily focused on designing recommender systems to enhance users' content consumption, the impact of recommender systems on content production remains understudied. To address this gap, we conducted a randomized field experiment on one of the world's largest video-sharing platforms. We manipulated the algorithm's recommendation of creators based on their popularity, excluding a subset of highly popular creators' content from being recommended to the treatment group. Our experimental results indicate that recommending content from less popular creators led to a significant 1.34% decrease in video-watching time but a significant 2.71% increase in the number of videos uploaded by treated users. This highlights a critical trade-off in designing recommender systems: popular creator recommendations boost consumption but reduce production. To optimize recommendations, we developed a structural model wherein users' choices between content consumption and production are inversely affected by recommended creators' popularity. Counterfactual analyses based on our structural estimation reveal that the optimal strategy often involves recommending less popular content to enhance production, challenging current industry practices. Thus, a balanced approach in designing recommender systems is essential to simultaneously foster content consumption and production.

Digital advertising, which uses consumer data to target ads to users, now accounts for most of global ad expenditures. Privacy concerns have prompted regulations that restrict the use of personal data. To inform these policy debates, we develop an equilibrium model of advertising and market structure to analyze the impact of privacy regulation on market outcomes. We test the model's predictions using the launch of Apple's App Tracking Transparency feature, which created a natural experiment that limited the use of consumer data. Leveraging data from all U.S. advertisers on Meta combined with offline administrative data, we find that reductions in digital ad effectiveness led to decreases in investments in advertising, increases in market concentration, and increases in product prices. These effects are economically meaningful in magnitude and suggest potential harms to both firms and consumers from privacy regulation.

Firms often need to develop a timing policy for interventions in response to a customer action. Examples include how long to wait before sending an online review request after a purchase, and time until sending a promotional offer to shopping cart abandoners. Causal inference in such settings, especially when the timeline for intervention can be over an extended period (e.g., a few weeks), is not straightforward due to the ''triple hurdle'' problem. That is, customers (1) need to enter the sample by taking an action, (2) comply with the intervention timing intended by the firm, and (3) choose their response to the intervention. We present a quasi-experiment design that can recover intervention treatment effects by timing when randomization into treatment and control groups is possible but random assignment to intervention timing is not. Specifically, we use (1) propensity score matching using a rich pre-treatment customer history to match the distribution of customer types across customer cohorts, (2) random assignment to treatment and control groups generating an exclusion restriction for the customer outcome equation, and (3) estimation of a conditional average treatment effects as a function of intervention timing. We discuss how the proposed design can allow a firm to test the effectiveness of interventions by timing while minimizing campaign duration and exposure to the intervention amongst the customer base. We demonstrate our approach in the context of shopping cart interventions that target customers leaving an item unpurchased in their online cart and find an inverted U-shaped lift pattern for intervention timing using a discount coupon.

People who persuade can use language that differs in average and variation of certainty. While higher average certainty has been shown to enhance persuasion, excessive certainty may harm it, and the impact of certainty variation has been overlooked. This research uses multiple methods to investigate. We analyze the transcripts of 7,811 sales calls, finding that higher average certainty generally boosts purchase likelihood, though too much certainty can backfire. Notably, at high levels of average certainty, greater variation in certainty reduces its positive effect. Pre-registered scenario-based experiments confirm key findings, showing that perceptions of persuader confidence and credibility mediate these effects, with uncertainty about controllable aspects being particularly harmful to evaluations and purchase intention. This study offers key insights for persuasion and source certainty.

Large Language Models (LLMs) have revolutionized natural language processing but can exhibit biases and may generate toxic content. While alignment techniques like Reinforcement Learning from Human Feedback (RLHF) reduce these issues, their impact on creativity, defined as syntactic and semantic diversity, remains unexplored. We investigate the unintended consequences of RLHF on the creativity of LLMs through three experiments focusing on the Llama-2 series. Our findings reveal that aligned models exhibit lower entropy in token predictions, form distinct clusters in the embedding space, and gravitate towards ''attractor states'', indicating limited output diversity. Our findings have significant implications for marketers who rely on LLMs for creative tasks such as copywriting, ad creation, and customer persona generation. The trade-off between consistency and creativity in aligned models should be carefully considered when selecting the appropriate model for a given application. We also discuss the importance of prompt engineering in harnessing the creative potential of base models. - The emergence of large language models (LLMs) has opened up exciting possibilities for simulating human behavior and cognitive processes, with potential applications in various domains, including marketing research and consumer behavior analysis. However, the validity of utilizing LLMs as stand-ins for human subjects remains uncertain due to glaring divergences that suggest fundamentally different underlying processes at play and the sensitivity of LLM responses to prompt variations. This paper presents a novel approach based on Shapley values from cooperative game theory to interpret LLM behavior and quantify the relative contribution of each prompt component to the model’s output. Through two applications—a discrete choice experiment and an investigation of cognitive biases—we demonstrate how the Shapley value method can uncover what we term ''token noise'' effects, a phenomenon where LLM decisions are disproportionately influenced by tokens providing minimal informative content. This phenomenon raises concerns about the robustness and generalizability of insights obtained from LLMs in the context of human behavior simulation. Our model-agnostic approach extends its utility to proprietary LLMs, providing a valuable tool for marketers and researchers to strategically optimize prompts and mitigate apparent cognitive biases. Our findings underscore the need for a more nuanced understanding of the factors driving LLM responses before relying on them as substitutes for human subjects in research settings. We emphasize the importance of researchers reporting results conditioned on specific prompt templates and exercising caution when drawing parallels between human behavior and LLMs.

Estimating consumer preferences for new products in the absence of historical data is an important but challenging problem in marketing, especially in product categories where brand is a key driver of choice. In these settings, measurable product attributes do not explain choice patterns well, which makes questions like predicting sales and identifying target markets for a new product intractable. To address this ''new product introduction problem,'' I develop a scalable framework that enriches structural demand models with large language models (LLMs) to predict how consumers would value new brands. After estimating brand preferences from choice data using a structural model, I use an LLM to generate predictions of these brand utilities from text descriptions of the brand and consumer. My main result is that LLMs attain unprecedented performance at predicting preferences for brands excluded from the training sample. Conventional models based on text embeddings return predictions that are essentially uncorrelated with the actual utilities. In comparison, my LLM-based model attains a 30% lower mean squared error and a correlation of 0.52 i.e. for the first time, informative predictions can be made for consumer preferences of new brands. I also show how to combine causal estimates of the price effect obtained via instrumental variables methods with these LLM predictions to enable pricing-related counterfactuals. Combining the powerful generalization abilities of LLMs with principled economic modeling, my framework enables counterfactual predictions that flexibly accommodate consumer heterogeneity and take into account economic effects like substitution by consumers and price adjustments by firms. Consequently, the framework is useful for downstream decisions like optimizing the positioning and pricing of a new product and identifying promising target markets. More broadly, these results illustrate how new kinds of questions can be answered by using the capabilities of modern LLMs to systematically combine the richness of qualitative data with the precision of quantitative data.