Abstract: Targeted advertising, advanced by behavioral tracking and data analytics, enhances consumer experience and marketing effectiveness. However, targeted advertising has raised significant privacy concerns among consumers and policymakers due to extensive data collection. Consequently, comprehending the tradeoff between match value and privacy concerns is crucial for the effective implementation of targeted advertising. In this research, we develop a structural model to empirically analyze this tradeoff, addressing a gap in the literature. We assume consumers form correlated beliefs about privacy invasion and match value from targeted ads in a Bayesian fashion, and use these beliefs to decide on ad clicks and opting out of ad tracking. Consumers update their privacy invasion beliefs based on how each received ad corresponds to their clicked ads and update their match value beliefs based on how well each ad engages them, doing so jointly due to the potential correlation between these beliefs. Leveraging the Limit Ad Tracking (LAT) policy change with iOS 10 in September 2016, which allows consumers to opt out of ad tracking, we estimate the model using panel ad impression and consumer response data from 166,144 opt-out and 166,144 opt-in consumers, across two months pre- and three months post-policy implementation. Our findings show that consumers generally dislike privacy invasion and prefer match value, with notable heterogeneity in these preferences. Consumers with higher uncertainty about privacy invasion are more likely to opt out of tracking. Upon opting out, highly privacy-sensitive consumers (about 20%) experience net benefits, while the majority face a net loss due to reduced match value. Furthermore, LAT challenges advertisers, especially smaller and lower-ranking ones, due to reduced visibility and increased customer acquisition costs. In counterfactual analyses, we propose a probabilistic targeting strategy that balances match value and privacy concern. Results suggest that it can benefit consumers, advertisers, and ad networks.

Abstract: Privacy regulations impact advertisers’ ability to target consumers based on behavioral information, leading to a shift toward targeting based on aggregate-level consumer data, such as contextual or bestseller targeting. This shift reduces ads relevance and disproportionately affects smaller advertisers in attracting new customers. To address this challenge, we propose a privacy-preserving targeting strategy. Using mobile in-app advertising as the empirical context, our algorithm, based on the SHOPPER framework, infers the app relationships from the “basket” of installed apps, and combines aggregate opt-in consumers data with ad content and relationship information. Using a Transformer model with a Deep Kernel Learning prediction head, the algorithm links consumer preferences to latent app features, generating ad sequences that closely resemble those from full targeting scenarios. This enhances consumer welfare by balancing match value and privacy and increases visibility for small advertisers.

Abstract: Targeted marketing interventions on social networks, such as referral campaigns and social advertising, are prevalent. Companies are increasingly interested in conducting network experiments using peer encouragement designs to causally quantify the heterogeneous direct effects on focal individuals (egos) and indirect effects on their connections (alters). A widely adopted practice to obtain clean estimates involves drawing random samples from the population network and excluding contaminated egos and alters (e.g., those in the treatment and control groups with a friend relationship) from the analysis. However, this approach often results in underrepresentation and undersupply, which have been documented in the literature as major technical challenges in conducting network experiments with peer encouragement designs. While underrepresentation indicates the samples’ lack of representation of the population characteristics, leading to biased inferences and limited generalizability, undersupply pertains to small sample size, resulting in low statistical power and experimental efficiency. In this research, we propose a representative sampling algorithm to improve peer encourage designs and causal inference by addressing these issues. Our method produces samples that better represent the population on individual network properties and personal characteristics and provides a larger sample size. Through simulations, we demonstrate that our method outperforms the traditional excluding approach, allowing more precise estimation of average treatment effects and heterogeneity. Additionally, the proposed method is computationally efficient and can be easily adapted for various applications evaluating social influences.

Abstract: Marketing interventions on social networks generate both direct and indirect treatment effects due to social influence, posing a challenge for standard Difference-in-Differences (DID). We propose a Bayesian semi-parametric approach (SDID-BART) that incorporates social influence within the DID framework. Our model uses Bayesian Additive Regression Trees (BART) to capture dynamic treatment effects, social influence effects, and their interactions with user characteristics, allowing for estimating direct and indirect treatment effects and their heterogeneity without assuming specific exposure models. Using data from a large-scale randomized experiment on a social network, we find that promoting an app function by adding a tab increases usage by 70.6% directly and 9.6% indirectly through social influence, indicating a downward bias in standard DID estimates. Additionally, we identify substantive heterogeneity in social influence, with untreated users and younger users reacting more positively to influence from their treated friends. Our research offers a practical tool for evaluating the effectiveness of marketing interventions on social networks and provides insights into targeting on social networks and Word-of-Mouth strategies.

Abstract: The premise of blockchain technology is that increasing public information and transparency should help people make more informed decisions. We investigated the influence of blockchain data on decision making, examining the mechanisms driving the price bubble on CryptoKitties, one of the first and most successful consumer products built on a blockchain. Observing past prices created significant bias in investing behavior. We found evidence of rational herding. Identifying the underlying mechanism confirmed that public transparency amplifies rather than reduces bias in decision making. Moreover, the tendency to herd was stronger during the market’s rise versus its collapse, indicating that investors used public information to justify beliefs in further price increases. Counterintuitively, our findings suggest that publicly disclosing all information is not sufficient for protecting investors. 

Abstract: This research sheds light on consumer motivations for participating in the sharing economy and examines downstream consequences of the uncovered motivations. We use text-mining techniques to extract Airbnb hosts’ motivations from their responses to the question “why did you start hosting.” We find that hosts are driven not only by the monetary motivation “to earn cash” but also by intrinsic motivations such as “to share beauty” and “to meet people.” Using extensive transaction-level data, we find that hosts with intrinsic motivations post more property photos and write longer property descriptions, demonstrating greater engagement with the platform. Consequently, these hosts receive higher guest satisfaction ratings. Compared to hosts who want to earn cash, hosts motivated to meet people are more likely to keep hosting and to stay active on the platform, and hosts motivated to share beauty charge higher prices. As a result, these intrinsically motivated hosts have a higher customer lifetime value compared to those with a monetary motivation. We employ a multimethod approach including text mining, Bayesian latent attrition models, and lab experiments to derive these insights. Our research provides an easy-to-implement approach to uncovering consumer motivations in practice and highlights the consequential role of these motivations for firms.

Abstract: Using text-mining, the authors develop version 1.0 of the Relevance to Marketing (R2M) Index, a dynamic index that measures the topical and timely relevance of academic marketing articles to marketing practice. The index assesses topical relevance drawing on a dictionary of marketing terms derived from 50,000 marketing articles published in practitioner outlets from 1982 to 2019. Timely relevance is based on the prevalence of academic marketing topics in practitioner publications at a given time. The authors classify topics into four quadrants based on their low/high popularity in academia and practice —“Desert,” “Academic Island,” “Executive Fields,” and “Highlands”—and score academic articles and journals: Journal of Marketing has the highest R2M score, followed by Marketing Science, Journal of Marketing Research, and Journal of Consumer Research. The index correlates with practitioner judgments of practical relevance and other relevance measures. Because the index is a work in progress, the authors discuss how to overcome current limitations and suggest correlating the index with citation counts, altmetrics, and readability measures. Marketing practitioners, authors, and journal editors can use the index to assess article relevance, and academic administrators can use it for promotion and tenure decisions (see www.R2Mindex.com). The R2M Index is thus not only a measurement instrument but also a tool for change.

Abstract: This paper quantifies the trade-off between social-economic characteristics and physical attractiveness. We use the state-of-the-art facial recognition algorithm to extract features such as eye size, lip width, and nose height from more than 460,000 photos. We find that education is the most important factor when it comes to whom to send a Like through the mobile dating app. Given the same educational attainment, men are more likely to use facial traits, and women are more likely to use height as the criterion to determine to whom to send a Like. We find that male users prefer women with the following facial cues: larger eyes, shorter distance between two eyes, smaller nose width, longer nose length, smaller upper lip, and larger lower lip. By contrast, most of the males' facial traits are not statistically significant in explaining females' Like data, except that larger upper lip and wider nose are unattractive traits.