Published Papers:
Banerjee, A., & Urminsky, O. (2022). What you are getting and what you will be getting: Testing whether verb tense affects intertemporal choices. Journal of Experimental Psychology: General.
https://doi.org/10.1037/xge0001192
Prior research has shown that the way information is communicated can impact decisions, consistent with some forms of the Sapir-Whorf hypothesis that language shapes thought. In particular, language structure—specifically the form of verb tense in that language—can predict savings behaviors among speakers of different languages. We test the causal effect of language structure encountered during financial decision-making, by manipulating the verb tense (within a single language) used to communicate intertemporal tradeoffs. We find that verb tense can significantly shift choices between options, owing to tense-based inferences about timing. However, the spontaneous use of verb tense when making choices occurs only in the complete absence of other timing cues and is eliminated if even ambiguous or nondiagnostic time cues are present, although prompted timing inferences persist. We test between multiple competing accounts for how verb tense differentially impacts timing inferences and choices. We find evidence for a cue-based account, such that the presence of other cues blocks the spontaneous use of verb tense in making intertemporal decisions, consistent with the “Good Enough” proposal in psycholinguistics.
(Data publicly available here)
“The Language That Drives Engagement: A Systematic Large-scale Analysis of Headline Experiments. ”
Banerjee, A., & Urminsky, O. (2024). The language that drives engagement: A systematic large-scale analysis of headline experiments. Marketing Science.
https://doi.org/10.1287/mksc.2021.0018
We use a large-scale data set of thousands of field experiments conducted on Upworthy.com, an online media platform, to investigate the cognitive, motivational, affective, and grammatical factors implementable in messages that increase engagement with online content. We map from textual cues measured with text-analysis tools to constructs implied to be relevant by a broad range of prior research literatures. We validate the constructs with human judgment and then test which constructs causally impact click-through to articles when implemented in headlines. Our findings suggest that the use of textual cues identified in previous research and industry advice does impact the effectiveness of headlines overall, but the prior research and industry advice does not always provide useful guidance as to the direction of the effects. We identify which textual characteristics make headlines most effective at motivating engagement in our online news setting.
(Materials publicly available here)
Papers under Revision/Review:
Banerjee, A. & Urminsky, O.
(Invited third round at Journal of Consumer Psychology)
Language is pervasive and hence a common factor in people’s decision making. Prior research has mostly studied the effects of comprehensible language, language that communicates a literal meaning to consumers – on behavior and attitudes. In this paper, we investigate the potential for language that is incomprehensible to a given consumer to nevertheless impact willingness to pay and choice. In particular, we propose that potentially meaningfully incomprehensible language can convey associations beyond literal meaning. We demonstrate that adding text in a foreign language unreadable to the consumer to a known native language description of foreign food significantly increases perceptions of authenticity, uniqueness, and quality, resulting in higher valuations and greater likelihood of choice, while holding the country of origin constant. Thus, we show that, contrary to prior accounts, an incomprehensible cue creates consumer value by instilling feelings of intangible experiences and that those feelings impact decisions. We test our framework using secondary field data as well as experiments, including with consequential choices.
(Data publicly available here)
“Confident judgments of (mis)information veracity are more, rather than less, accurate."
Banerjee, A., & Rocklage, M., Mosleh, M., Rand, D.
(Under second round review at PNAS Nexus)
Confidence is increasingly viewed as a barrier to recognizing misinformation, as people who are more confident tend to be more likely to believe and share false news. Here, we challenge this view by distinguishing between two types of confidence: general confidence in one’s abilities, and specific confidence in a particular judgment. Using a large, pre-registered study in which participants judge the accuracy of news posts, we demonstrate a striking dissociationbetween these two forms of confidence. While higher general confidence is associated with worse discernment of true versus false headlines (in line with past work), we find that higher confidence in specific judgments is associated with better truth discernment - a finding that we observe among both Democrats and Republicans. These results call for a reevaluation of the role of confidence in the detection of misinformation. Focusing on confidence in specific judgments, rather than general abilities, could play an important role in helping to mitigate belief in misinformation.
(Materials publicly available here)
Banerjee, A., & Urminsky, O.
(Submitted to PNAS)
Negative language is widely believed to boost engagement online—but does it always work? Across six studies combining over 150,000 headline field experiments and five pre-registered experiments (N = 8,369), we find that the effects of negative tone and emotion vary widely and depend critically on the reader’s goals. In large-scale headline A/B tests across 398 news platforms from around the world, negatively worded headlines showed no average effect on engagement, but revealed stark heterogeneity across platforms. Experiments testing this heterogeneity do not support differences in content or reader’s site-specific expectations, but instead identified a motivational distinction: whether readers seek credibility or enjoyment from news. Crucially, this motivational orientation appears to be a characteristic of the audiences that different platforms attract—rather than a structural feature of the platforms themselves. In controlled experiments, we show that under credibility-seeking goals, negative language triggers inferences of manipulative intent, suppressing engagement even for societally important topics. By contrast, under enjoyment goals, the same negative language may increase appeal to readers. In political contexts, we further find that while negative emotion heightens affective polarization, negative tone—when paired with credibility goals—can reduce partisan favoritism. These findings challenge the prevailing assumption that “negativity sells,” and instead show that the effects depend on goal alignment. In high-stakes domains like health, science, and politics, emotional negativity in headlines can backfire—undermining trust, engagement, and civic discourse. Goal-sensitive communication strategies are essential in a fragmented digital environment.
(Materials publicly available here)
Working Papers:
“Linguistic Amplifiers of Misinformation: How Toxicity and Confidence Signal the Spread Low-Quality News Online”
Banerjee, A., Aghamohammadi, Z., Rocklage, M., Rand, D., Mosleh, M. (In Preparation for PNAS)
As misinformation proliferates online, understanding the linguistic features that fuel its spread is critical. Analyzing 27 million posts across eight social media platforms, this study finds that two linguistic cues—toxicity and confidence—are more common in posts linking to lower-quality news sources. These same features also increase engagement, suggesting they play a central role in amplifying misinformation.Toxicity, marked by hostile or inflammatory language, attracts attention through outrage. Confidence, conveyed through assertive and certain language, enhances perceived authority. Together, these cues not only characterize misinformation-prone content but also help it reach wider audiences. This work contributes to the growing literature on language and misinformation by linking specific linguistic markers to both content quality and engagement. It underscores the role of language in shaping what people see and share online, and highlights the need for platform interventions that consider how toxicity and confidence reinforce the spread of low-quality information.
Ongoing Projects:
“Does Big-Data Correlational Analysis Predict Causal Effects of Language on Decisions?”
Banerjee, A. & Urminsky, O.
A substantial research literature has used large-scale correlational text analysis to determine how characteristics of language encountered by people predicts (and presumably causes) their subsequent behavior, such as online engagement or donation decisions. These approaches have high external validity, by analyzing large-scale real- world data and behavior, but are causal interpretations of the findings internally valid? Using a novel large-scale dataset containing thousands of headline experiments, we compare the results of correlational analyses (excluding the experimental variation) with causal analyses (solely using experimental variation) to test what factors increase the likelihood of click-through in an online news setting. We find that not only does experimental data provide higher statistical power, but the correlational findings differ in magnitude, and sometimes even in direction, from the causal findings. Our results suggest that big-data correlational analyses may provide poor predictions of causal effects, underscoring the need for marketers to conduct experiments.
“Choosing Oleander Over Zanthoxylum: How Consumer Inferences of Chemicalness from Linguistic Cues in Non- Comprehended Ingredients Influence Product Choice”
Banerjee, A., Chen, S., Urminsky, O.
We explore how consumers integrate incomprehensible but meaningful linguistic cues into decision making, finding that they use linguistic cues to categorize unknown ingredients as chemical-sounding or natural-sounding. Our findings reveal that consumers are more likely to favor products with natural-seeming names, often perceiving them as less harmful and more desirable, despite lacking a full understanding of the ingredients. This research highlights the significant impact of associations drawn from incomprehensible labels on product choice and consumer decision-making.
“Asymmetric Variety Seeking in Hierarchical Choices”
Banerjee, A.*, & Winet, Y.*
People frequently seek variety when choosing hedonic experiences for themselves (McAlister & Pessemier, 1982; Ratner, Kahn & Kahneman, 1999; Simonson, 1990). They also face multiple choices that contain other choices. This research explores how the structure of decision-making hierarchies impacts consumer variety-seeking behavior. When making hierarchical choices—such as selecting a restaurant before choosing a specific dish — consumers tend to seek more variety at higher levels (e.g., restaurant) compared to lower levels (e.g.,dish). This effect occurs in a variety of domains, in a consequential decision, and with both familiar and unfamiliar option sets. We also rule out alternative explanations, showing this effect cannot be explained by a sequence effect or a desire to have a prototypical experience. This work builds on extant work on choice architecture showing how understanding the impact of small changes in the structure of decisions can push consumers to make better decisions, which has important implications for improving consumer well-being.
“Information Mis-Creation: The Role of Story-Telling and Curiosity Gaps”
Banerjee, A., Wadhwani, R., Loretizo, J., Aribarg, A.
How content is written—not just what it says—shapes the way people engage with it online. Across Reddit data and a large lab study, we show that story-like writing and curiosity gaps (i.e., missing information that prompts questions) influence how users generate speculative or polarizing content in response. In political discussions, story-like writing and curiosity gaps each increase speculation, but their combination can reduce it by providing a more complete narrative. In short-form or headline-driven posts, however, curiosity gaps fuel speculation by inviting users to fill in missing details. In non-political forums, these same features spark speculation without increasing disagreement. These findings suggest that speculation and polarization don’t only result from misinformation exposure—they can also emerge from the structure of otherwise accurate content. By highlighting how linguistic features drive reader inference, our research underscores the need for platform-level interventions that address not only misinformation itself, but also the ways people fill in its gaps.