Morrison Center for Marketing and Data Analytics

Yu Ding - A Novel Crowdsourcing Approach for Improving the Information Ecosystem

"Given the explosion of news transmitted to, and shared by, consumers across different media, the veracity of information is of critical importance. However, the scale of existing fact-checking organizations is limited, hence resulting in a scant proportion of news articles being fact-checked. We address the challenge of scaling up fact-checking operations in the domain of science-related articles by proposing and testing a novel crowdsourcing solution. A key challenge with asking lay consumers to rate the veracity of scientific news articles is that they are likely to be biased by their prior beliefs. Using articles that have been rated for veracity by scientists as a starting point, we overcome this bias by proposing the use of crowdsourced similarity ratings rather than veracity ratings. We find that asking lay consumers to rate the similarity between scientist-rated and unrated articles provides an unbiased, effective, and efficient way to scale up veracity ratings of scientific articles. Our proposed method (human similarity-judgments) outperforms algorithm-rated similarity (e.g., by TF-IDF and by Word Embedding) to predict more accurately an article’s scientific veracity. Our alternative is also superior to previous approaches for judging veracity—such as utilizing the semantic markers of false news previously detected by previous research. We further compute a “transitivity index” to identify consumers likely to be more accurate at making similarity judgments and show how veracity predictions can be improved by paying close attention to consumer segments recruited for the similarity-judgment task. We demonstrate that our method can predict the scientific veracity of articles with over 95% accuracy and that both type-I and type-II errors are minimized. Lastly, we find preliminary evidence that the advantage of using similarity judgments comes from the fact that consumers are more likely to distance themselves from the arguments being rated. We also discuss the limitation of this method (e.g., topic level granularity) and further research (e.g., increase trust by involving consumers in fact-checking)."

Tesary Lin - Choice Architecture, Privacy Valuations, and Selection Bias in Consumer Data

"How much do consumers value their personal data? How do their privacy valuations change under the influence of choice architecture? How does this influence affect the efficiency of data collection, not only by changing the quantity of data collected, but also by altering its representativeness? To answer these questions, we run a large-scale choice experiment to elicit consumers’ valuation for their Facebook data while randomizing two common choice frames: default and price anchor. An opt-out default decreases valuations by 14.4% compared to opt- in, while a $0-$50 price anchor decreases valuation by 51.1% compared to a $50-$100 anchor. Moreover, in some consumer segments, the susceptibility to frame influence is negatively correlated with consumers’ valuation absent frames. Our results suggest that conventional frame optimization practices with a goal to maximize the supply of data can have opposite effects on its representativeness: A bias exacerbating effect emerges when consumers who value their data less are also easier to be influenced by frames. On the other hand, such a volume- maximizing frame may also mitigate the bias by getting a high percentage of consumers into the sample data, thereby improving its coverage."

Nils Wernerfelt - Estimating the Value of Offsite Data to Advertisers on Meta

"We study the extent to which advertisers benefit from data that are shared across applications. These types of data are viewed as highly valuable for digital advertisers today. Meanwhile, product changes and privacy regulation threaten the ability of advertisers to use such data. We focus on one of the most common ways advertisers use offsite data and run a large-scale study with hundreds of thousands of advertisers on Meta. Within campaigns, we experimentally estimate both the effectiveness of advertising under business as usual, which uses offsite data, as well as how that would change under a loss of offsite data. Using recently developed deconvolution techniques, we flexibly estimate the underlying distribution of treatment effects across our sample. We find a median cost per incremental customer using business as usual targeting techniques of $43.88 that under the median loss in effectiveness would rise to $60.19, a 37% increase. Similarly, analyzing purchasing behavior six months after our experiment, ads delivered with offsite data generate substantially more long-term customers per dollar, with a comparable delta in costs. Further, there is evidence that small scale advertisers and those in CPG, Retail, and E-commerce are especially affected. Taken together, our results suggest a substantial benefit of offsite data across a wide range of advertisers, an important input into policy in this space."

Click Here to View Paper

Ryan Hamilton - In Lieu of Gifts: Understanding and Overcoming Givers’ Reduced Generosity Toward Charitable Gift Requests

"Charitable organizations are increasing their reliance on fundraisers created by individuals on behalf of a charity. Many such fundraisers feature requests for donations in lieu of gifts for celebratory occasions such as weddings and baby showers. Despite the donation being requested as a gift, givers tend to be markedly less generous when giving a charitable donation gift relative to giving typical gifts, including gifts of cash. Across six studies, including a study utilizing secondary data from a gift-registry organization, this research examines why givers are less generous when presented with requests for a charitable gift compared to other gift requests. We find that givers’ reduced generosity is driven by their desire to give a gift that offers a practical benefit to the recipient, rather than by consideration of the recipients’ desires. This motivation toward practicality is especially pronounced for giving associated with rites-of-passage occasions, such as weddings. Understanding this motivation is important, because such occasions are often utilized to request charitable donations in lieu of gifts. We also identify practical methods to reduce the generosity gap by increasing the efficacy of charitable gift requests."

Kristin Donnelly - Once and Again: Repeated Viewing Affects Judgments of Spontaneity and Preparation

"Reality is fleeting, and any moment can only be experienced once. Re-watching a video, on the other hand, allows people to repeatedly observe the exact same moment. We propose that people apply their understanding of repetition in the real world to the replay context, failing to fully distinguish behavior that they merely observe again (through video replay) from that behavior being performed again in the exact same way. Eleven experiments (N = 10,602) support this idea across a broad assortment of stimuli that includes auditions, dances, commercials, public speeches, and potential evidence in a murder trial. We demonstrate that re-watching makes a videotaped behavior appear more rehearsed and less spontaneous, as if the actor in the replayed video were simply precisely repeating the same set of actions. We rule out alternate explanations including repetition increasing accuracy of judgments, mere exposure leading to a positivity bias, and experimenter demand effects. These findings build on an influential literature showing that incidental video features like perspective or slow motion can meaningfully change how people judge the action of the video. Video re-watching may inadvertently shape judgments in contexts ranging from mundane to consequential. To understand how a video is going to influence its viewer, one will need to consider not only its content, but also how often it is viewed."

Ada Aka - Inferring Consideration Sets: A Computational Model of Naturalistic Memory-Based Decision Making

"Memory plays a crucial role in everyday decision making and especially in consumer decision making. We propose a new computational framework to study how people retrieve and choose between hundreds of common choice items stored in memory during such decisions. Our approach combines established theories of consideration set formation and memory search, with techniques from natural language processing (which use text data to derive representations and associations for choice items) and recommender systems (which provide algorithms for capturing individual-specific preferences and retrieval tendencies for such items). We show that our framework successfully describes the items that are retrieved from memory even when memory processes are not directly observed. It also captures the effects of situational variables and individual differences on memory. Thus, it provides data-driven insights into the core cognitive mechanisms at play in memory-based decision making. We demonstrate the power of this approach in three sets of studies, each with several different types of naturalistic decision prompts. In doing so, we show how established theories in marketing and psychology can be combined with new computational techniques to explain complex everyday decisions."

Greg Lewis  "Online Search and Product Rankings: A Double Index Approach"

"We develop a flexible yet tractable model of consumer search and choice, and apply it to the problem of product rankings optimization by online retail platforms. In the model, products are characterized by an observable search index, which governs what consumers search; and a utility index, which governs which of the searched options is purchased. We show that this framework generalizes several commonly used search models. We then consider how platforms should assign products to search ranks. To optimize consumer surplus, platforms should promote “diamonds in the rough,” products whose utility index exceeds their search index. By contrast, to maximize static profits, the platform should favor high-margin products, creating a tension between the two objectives. We develop computationally tractable algorithms for estimating consumer preferences and optimizing rankings, and we provide approximate optimality guarantees in the latter case. When we apply our approach to data from Expedia, our suggested ranking achieves both higher consumer surplus and higher revenues than is achieved by the Expedia ranking, and also dominates ranking the products in order of utility."

Hortense Fong, "A Theory-Based Explainable Deep Learning Architecture for Music Emotion"

Music is used to evoke emotion throughout the customer journey. This paper develops a theory-based, explainable deep learning convolutional neural network (CNN) classifier—MusicEmoCNN—to predict the time-varying emotional response to music. To develop a theory-based CNN, we first transform the raw music data into a format—mel spectrogram—that accounts for human auditory response. Next, we design and construct novel CNN filters for higher-order music features that are based on the physics of sound waves and associated with perceptual features of music, like consonance and dissonance, which are known to impact emotion. The key advantage of our theory-based filters is that we can connect how the predicted emotional response (valence and arousal) is related to human-interpretable features of music. Our model outperforms traditional machine learning models and performs comparably to state-of-the-art black-box CNN models. Finally, we illustrate an application involving digital advertising. Motivated by YouTube’s mid-roll advertising, we first conduct a lab experiment in which we exogenously place ads at different times within content videos and find that ads placed in emotionally similar contexts are more memorable in terms of higher brand recall rates. For a given ad, we then use the model's predictions to identify emotionally similar contexts in content videos.

Stephanie Chen, "Getting Credit and Minimizing Blame: How Social Affiliations Influence Perceptions of Charitable Giving (and Not Giving)"

While prior research on charitable giving has focused on exploring how much credit consumers and organizations get for charitable giving, less attention has been focused on the negative reputational consequences that may accompany the decision not to give. Based on accounts that suggest that social categories and affiliations mark interpersonal obligations, I explore whether perceptions of these obligations influence how the act of donating or not donating is viewed. An asymmetry occurs in which consumers do not get more credit for donating to people within their social groups (vs. outside of their social groups) but are blamed more for not donating to those within their social groups (vs. outside of their social group) because they are seen as breaking an obligation. The asymmetry is attenuated when the charitable cause is focused on providing benefits rather than removing harm, consistent with the view that there is a greater obligation to avoid harm than to provide benefits. The results highlight the importance of understanding perceptions of consumers both when they donate and when they do not, and suggest that feelings of obligation may be a barrier to efficient giving.

Tong Guo, "Adoption of Bio-Engineered Food"

We study the early-stage adoption of impossible meat, a novel food technology that synthesizes meat substitutes by closely simulating the texture, flavor, and appearance of real meat. Unlike traditional veggie meat that targets vegans and vegetarians, impossible meat attempts to attract meat lovers with minimal taste differentiation. How would this apparently artificial food product gain its success while the market demands less bio-engineered and more organic food? We document the early-stage adoption pattern of impossible meat by overcoming the common challenges in understanding adoption of bio-engineered food: 1) the new technology usually does not reach consumers before first being adopted by intermediaries, for whom the data is often hard to obtain; 2) the documentation of the marketing antecedent of adoption decisions is incomplete and endogenous, making it hard to causally identify the driving factors behind the technology adoption. Focusing on the key players in their US market debut between 2015-2019, we construct a novel location-specific adoption metric that accurately measures the decisions of local intermediaries. To explain the adoption pattern, we analyze a comprehensive set of media corpus using Natural Language Processing. We find that local news focusing on health and financing of the new technology positively increases the regional adoption of impossible meat, whereas discussion on sustainability and consumer WOM has little impact.

Sam Maglio, "How Long Do People Use Their Stuff?"

Research in marketing, psychology, and behavioral economics has long examined what people consume, when people consume, and why people consume. But what about how long people consume? For things that aren’t swallowed, snuffed, or savored in an instant, there exists a clear duration of time over which people acquire a continued stream of utility. As the owners of it, they get to decide how long that consumption period lasts. This research asks what goes into that decision. A series of longitudinal studies and supplemental experiments—across a variety of duration measures and products—offers three answers. First, while incentives like nudges and norms increase choosing, they subsequently reduce consuming. Second, certain personality traits make people more prone to keep or discard their belongings. Third, people can consider planned ownership for new acquisitions, which colors how they treat that stuff now and later.

Joachim Vosgerau, "Big Data - Big Biases"

Big data provide incredibly accurate descriptive statistics, and — as input to sophisticated AI methods — allow for accurate predictions of complex phenomena, causing big data’s “aura of truth, objectivity, and accuracy” (Boyd & Crawford 2012). The aim of this research project is to demonstrate that this “aura of truth” is dangerous. Because larger sample sizes (e.g., big data) vastly improve descriptive statistics, decision-makers erroneously believe that they also improve statistical and causal inferences, rendering decision-makers more susceptible to data-inferential biases. Specifically, we demonstrate that the larger is the size of a data sample, the more likely are decision-makers to interpret correlations as indicative of causation. 

Contrary to the popular belief that “With enough data, the numbers speak for themselves,” our research highlights that data do not speak for themselves, but human decision-makers interpret data patterns, and because they are human, they make mistakes in their inferential judgments.