research
The corresponding author is denoted by *. Video summaries 🎬 generated with NotebookLM.
manuscripts in revision
current
- JBRWhy does pharmaceutical detailing increase over time? Indication expansion versus inter-firm competitionSungjoon Nam, Minki Kim, and Jaewon Yoo*Revise & Resubmit (3rd round, minor revision), Journal of Business Research (SSCI/NSTC Mgmt. II Top Journal List)
The pharmaceutical industry is well known for their enormous marketing expenditures, particularly on direct-to-physician promotions, also known as detailing. Whereas previous studies have typically concluded that firms should start with a high level of detailing and decrease those activities over time, we find that actual detailing activities do not follow the predicted down- sloping pattern but rather the opposite. Our unique data set (IMS Spectra data) shows that firms’ detailing activities increased over time for approximately 40% of 464 drugs. This study sheds light on that discrepancy and the underlying reasons for it. We find that firms show increasing patterns of detailing activities when it takes longer to learn about a drug’s efficacy and when uncertainty concerning future sales is higher. Of interest to policymakers, our results reveal that increases in detailing activity are not necessarily driven by excessive competition among pharmaceutical firms. Rather, increases in detailing activities are associated with the approval of drugs for new indications, which help to inform physicians’ decisions about prescriptions.
@article{nam_detailing, kind = {manuscripts in revision}, title = {Why does pharmaceutical detailing increase over time? Indication expansion versus inter-firm competition}, author = {Nam, Sungjoon and Kim, Minki and Yoo, Jaewon}, year = {current}, journal = {Revise & Resubmit (3rd round, minor revision), <i>Journal of Business Research (SSCI/NSTC Mgmt. II Top Journal List)</i>}, }Generated with NotebookLM - JNMInteractive Blind Lead-In Ads and Purchase Conversion: The Roles of Completion and Creative Design FeaturesRevise & Resubmit (2nd round), Journal of Interactive Marketing (SSCI/NSTC Mgmt. II Top Journal List)
- Presented: Darden-Cambridge Judge-HKU FBE Entrepreneurship and Innovation Research Conference; NYU-Temple Conference on Digital, Mobile Marketing, and Social Media Analytics (by invitation only)
- Grant: Principal investigator, National Science and Technology Council, Taiwan (NSTC), "Curiosity Evocation and Resolution in Ads: Empirical Investigation in Large-Scale Randomized Controlled Trials" (2022-2024)
Interactive display advertising increasingly employs delayed product disclosure and multi-stage formats that require active user progression, yet little is known about how such designs affect purchase behavior when exposure to the reveal depends on users’ interaction. This article studies interactive blind lead-in (BLI) ads using a large-scale randomized field experiment with a major book retailer, in which more than 109,000 customers were randomly assigned to either an interactive BLI ad with delayed reveal or a standard static ad with immediate product disclosure over an 80-day period. Findings indicate assignment to the interactive BLI format does not increase average purchase conversion, either immediately or within 30 days. Using random assignment as an instrument, however, results show that completing the BLI sequence increases long-term purchase conversion among compliers, indicating that delayed-reveal interactivity is effective only when users meaningfully interact with the ad. Completion varies systematically across ads in ways consistent with a Cost-of-Progress framework. Stronger early informational cues increase completion, whereas greater textual and visual complexity later in the sequence reduces completion, and social cues and higher user-ad fit raise completion. A supplementary study shows that the downstream impact of interaction with a BLI ad depends on the overall hedonic experience.
@article{yoon_bli, kind = {manuscripts in revision}, title = {Interactive Blind Lead-In Ads and Purchase Conversion: The Roles of Completion and Creative Design Features}, author = {Yoon, Seokjoon and Yoo, Jaewon and Kim, Minki and Kim, Wonjoon and Lee, Min Sok}, year = {current}, journal = {Revise & Resubmit (2nd round), <i>Journal of Interactive Marketing (SSCI/NSTC Mgmt. II Top Journal List)</i>}, presentation = {Darden-Cambridge Judge-HKU FBE Entrepreneurship and Innovation Research Conference; NYU-Temple Conference on Digital, Mobile Marketing, and Social Media Analytics (by invitation only)}, grant = {Principal investigator, National Science and Technology Council, Taiwan (NSTC), "Curiosity Evocation and Resolution in Ads: Empirical Investigation in Large-Scale Randomized Controlled Trials" (2022-2024)}, } - JMHelping Hands to Bear the Burden of Choice: Recommendations from Similar, but Not Close, Others Reduce Choice DifficultyJaewon Yoo, Wonjoon Kim*, and Joshua AckermanRevision invited, Journal of Marketing (UTD24, FT50)
The introduction of the Internet generated two significant developments for consumers—the emergence of social network services and relentlessly increasing product variety. However, recent research has highlighted the negative consequences of extensive choice variety and how it overloads cognition. Three experimental studies and a pilot study show that social recommendations appearing in the choice context mitigate assortment-driven choice difficulty, thereby eliminating choice overload effects. Interestingly, this effect is moderated by the similarity, but not by the felt social distance, between the recommendation provider and the chooser. In other words, consumers who are burdened by choice difficulty are more likely to choose the alternative recommended by similar others, which allows them to reduce the difficulty of evaluating other options available in the assortment—an activity that can produce dissatisfaction and choice deferral.
@article{yoo_helpinghands, kind = {manuscripts in revision}, title = {Helping Hands to Bear the Burden of Choice: Recommendations from Similar, but Not Close, Others Reduce Choice Difficulty}, author = {Yoo, Jaewon and Kim, Wonjoon and Ackerman, Joshua}, year = {current}, journal = {Revision invited, <i>Journal of Marketing (UTD24, FT50)</i>}, }
manuscripts under review
current
- [Title and authors withheld; under blind review]
- [Title and authors withheld; under blind review]
working papers
current
- Structure-Aware Robust Counterfactual Explanations via Conditional Gaussian Network Classifiers
- Award: 2025 Best Master's Thesis Award (Honorable Award), Algorithms and Computation Theory Society of Taiwan. Awarded to Zhan-Yi Liao's master's thesis, co-advised with Po-An Chen; presented at CMCT 2026.
Counterfactual explanation (CE) is a core technique in explainable artificial intelligence (XAI), widely used to interpret model decisions and suggest actionable alternatives. This work presents a structure-aware and robustness-oriented counterfactual search method based on the conditional Gaussian network classifier (CGNC). The CGNC has a generative structure that encodes conditional dependencies and potential causal relations among features through a directed acyclic graph (DAG). This structure naturally embeds feature relationships into the search process, eliminating the need for additional constraints to ensure consistency with the model’s structural assumptions. We adopt a convergence-guaranteed cutting-set procedure as an adversarial optimization framework, which iteratively approximates solutions that satisfy global robustness conditions. To address the nonconvex quadratic structure induced by feature dependencies, we apply piecewise McCormick relaxation to reformulate the problem as a mixed-integer linear program (MILP), ensuring global optimality. Experimental results show that our method achieves strong robustness, with direct global optimization of the original formulation providing especially stable and efficient results. The proposed framework is extensible to more complex constraint settings, laying the groundwork for future advances in counterfactual reasoning under nonconvex quadratic formulations.
@article{yoo_robustce, kind = {working papers}, title = {Structure-Aware Robust Counterfactual Explanations via Conditional Gaussian Network Classifiers}, author = {Liao, Zhan-Yi and Yoo, Jaewon and Yang, Hao-Tsung and Chen, Po-An}, year = {current}, } - From What Ifs to Insights: Counterfactuals in Causal Inference vs. Explainable AI
Counterfactuals play a pivotal role in the two distinct data science fields of causal inference (CI) and explainable artificial intelligence (XAI). While the core idea behind counterfactuals remains the same in both fields—the examination of what would have happened under different circumstances—there are key differences in how they are used and interpreted. We introduce a formal definition that encompasses the multi-faceted concept of the counterfactual in CI and XAI. We then discuss how counterfactuals are used, evaluated, generated, and operationalized in CI vs. XAI, highlighting conceptual and practical differences. By comparing and contrasting the two, we hope to identify opportunities for cross fertilization across CI and XAI.
@article{shmueli_counterfactual, kind = {working papers}, title = {From What Ifs to Insights: Counterfactuals in Causal Inference vs. Explainable AI}, author = {Shmueli, Galit and Martens, David and Yoo, Jaewon and Greene, Travis}, year = {current}, eprint = {2505.13324}, archiveprefix = {arXiv}, primaryclass = {stat.ML}, } - Exploring Apps with Mixed Ratings: Examining How Users Find Regulatory Fit in Making Purchase DecisionEdgar A. Duron, Soumya Ray, and Jaewon Yoo
- Presented by Edgar A. Duron: ISR Paper Development Workshop at PACIS (2024); TSWIM (2024); DIGIT Workshop at ICIS (2024); MIS Quarterly Author Workshop (2026).
App stores present a unique challenge for users because they must make quick decisions based on limited information. This speedy decision making process is not favorable for apps with mixed reputations (e.g., having many 5-star ratings but also many 1-star ratings). We seek to understand, in part, why users might still be attracted to purchase such apps, despite their bi-polar ratings distributions. Regulatory focus theory, which is gaining traction in information systems research, helps explain how users make decisions on such fast-paced and uncertain online platforms, by distinguishing between the choices of people who are promotion focused versus being prevention focused. We thus expect that users’ regulatory focus (promotion or prevention) shapes which app they finally choose to purchase. But we further propose that their choice is contingent on whether they explored the app and how well it matched their regulatory focus. This match is called regulatory fit in earlier theory development, but its utility seems to have been lost in empirical studies; however, modern platforms like app stores give new grounds to examine and reinvigorate this theoretical concept. We test our proposed model using a randomized experiment of users on a mock app store. Our results demonstrate that by inducing a prevention-focused mindset, users become more willing to consider apps with mixed ratings, but only if they chose to explore them prior to making the purchase decision. Our renewed perspective of regulatory fit highlights the nuances of the exploration process itself, which has not yet been closely examined by researchers. Our findings extend how regulatory focus theory could be applied in information systems, and offer practical insights for app developers who seek to survive on competitive digital platforms.
@article{duron_exploring, kind = {working papers}, title = {Exploring Apps with Mixed Ratings: Examining How Users Find Regulatory Fit in Making Purchase Decision}, author = {Duron, Edgar A. and Ray, Soumya and Yoo, Jaewon}, year = {current}, } - Navigating the New Retail Landscape: Mobile Scan-and-Go and the Impact of Mobile Payment Adoption
- Award: Winner for the Best Paper Award in KMA Doctoral Dissertation Competition at The 1st Joint Conference on Distribution, Marketing, Advertising, and Consumer Behavior (DMAC)
- Presented: The 11th Yale-MIT China India Consumer Insights Conference (Accepted as a recipient of the doctoral student support from Yale School of Management); 2022 Information Systems Research (ISR) Author Development Workshop; The 3rd CEIBS Marketing Symposium
- Grant: Principal investigator, National Science and Technology Council, Taiwan (NSTC), "Mobile Scan-and-Go and the Future of Offline Retail: How Reduced Transaction Costs Affect Consumer Behavior" (2023-2025)
In today’s dynamic retail environment, mobile Scan-and-Go services are providing a new way for consumers to shop in-store. To better understand this new retailing practice, we collaborated with a premier book retailer in Asia, which has a well-established mobile Scan-and-Go service. Our findings, based on detailed transactional data, reveal that these users tend to consume a more diverse range of products compared to non-users. More importantly, we show that mobile payment adoption within the context of Scan-and-Go further amplifies the consumption variety of those users. Utilizing a difference-in-differences approach coupled with propensity score matching, we find that mobile payment adoption leads to a significant reduction in the normalized Herfindahl-Hirschman Index (HHI) across multiple dimensions of consumption. This reduction in HHI indicates that users are diversifying their consumption by exploring a broader range of categories. Additionally, we observe a positive impact on overall spending. Grounded in transaction cost theory, our research underscores that the reduction in transaction costs is a key mechanism driving these changes. Our study contributes to the understanding of payment methods and retail technologies, providing novel insights into the strategic use of technological advancements to enhance consumer behavior and bolster emerging sales channels. Our findings offer strategic implications for retailers aiming to leverage technology to thrive in a rapidly evolving market.
@article{yoo_navigating, kind = {working papers}, title = {Navigating the New Retail Landscape: Mobile Scan-and-Go and the Impact of Mobile Payment Adoption}, author = {Yoo, Jaewon and Chen, Yuxin and Kim, Minki and Kim, Wonjoon}, year = {current}, ssrn = {https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4947618}, presentation = {The 11th Yale-MIT China India Consumer Insights Conference (Accepted as a recipient of the doctoral student support from Yale School of Management); 2022 Information Systems Research (ISR) Author Development Workshop; The 3rd CEIBS Marketing Symposium} }
work-in-progress (correct authoring will be determined later)
current
- Improving Causal Identification with Explainable AI-Assisted Confounder Discovery
Credible causal inference hinges on two distinct tasks; i.e., identification, which secures exogenous variation in treatment through designs such as regression discontinuities or instrumental variables, and adjustment, which conditions on the correct set of covariates so that the identifying assumptions (e.g., conditional ignorability, conditional parallel trends) plausibly hold. While recent machine‑learning pipelines automate covariate selection, they remain vulnerable to over‑adjustment (collider bias) and under‑adjustment (residual confounding) when covariate roles are ambiguous. We propose Explainable AI‑assisted Confounder Discovery (XCD) which combines flexible machine learners, explainable AI feature attributions, and a targeted conditional‑dependence test to distinguish genuine confounders from bad controls. The resulting adjustment set satisfies the back‑door criterion under mild assumptions, yielding bias‑robust effect estimates without inflating variance by including irrelevant controls. We revisit Dube et al. (2020) and demonstrate that XCD can effectively isolate confounders and reduces bias compared with double‑Lasso, PC/FCI, and naively adjusting for all pre‑treatment variables. XCD thus offers a transparent, scalable, and empirically grounded solution for confounder selection in modern high‑dimensional causal analysis.
- Can AI Dream Up a Control Group? Generating Synthetic Respondents to Reconstruct Untreated Potential Outcomes
The estimation of causal effects from marketing interventions is often hampered by the absence of feasible randomized control trials (RCTs), forcing researchers to infer counterfactuals from observational data. While the conventional synthetic control method offers a powerful solution by re-weighting real control units, its efficacy is contingent on the availability of a suitable donor pool. This research investigates whether generative AI can create high-fidelity "synthetic respondents" to serve as a valid counterfactual for treated populations. We ask: Can generative models, trained on pre-intervention data, produce synthetic control groups that are more accurate and broadly applicable than those derived from traditional weighting methods? This question is central to the future of scalable marketing research and causal testing. We answer this question in a three-step validation framework. First, using a real-world marketing dataset from a completed RCT, we fine-tune a large language model (LLM) on the rich, high-dimensional pre-intervention data of the treated units. Second, we task the model with generating a complete set of synthetic respondents, simulating the counterfactual scenario where no treatment was received. Finally, we rigorously evaluate the validity of this approach by: (1) comparing the outcomes of our AI-generated control group with the actual, real-world control group from the original experiment, and (2) comparing the causal effect estimated using our synthetic control with the "ground truth" effect from the RCT. This study aims to make a significant methodological contribution by providing a framework for validating the use of generative AI in causal settings. The findings will offer practitioners early insights into leveraging generative AI for more robust and scalable causal analysis, especially in scenarios where traditional methods fall short.
publications
2018
- AERCompromise Effect and Consideration Set Size in Consumer Decision-MakingJaewon Yoo, Hyunsik Park, and Wonjoon Kim*Applied Economics Letters, 2018
The compromise effect dictates that a decision-maker chooses a middle option over an extreme one given a set of choice alternatives since choosing an intermediate option is easier to justify, less likely to be criticized, and is consistent with loss aversion. Our experiment is designed to identify whether the connection between the extremeness of the options and the size of the consideration sets is economically and statistically significant and thus would have important behavioural implications. Specifically, we compare decision-making under small and large consideration sets where the extremeness of the comprising choice options is high, as opposed to low. The results demonstrate that an increase in consideration set size leads to weaker compromise effect (i.e. boundary condition) but when composed of high extremeness, strengthens the compromise effect.
@article{yoo2018compromise, kind = {publications}, title = {Compromise Effect and Consideration Set Size in Consumer Decision-Making}, author = {Yoo, Jaewon and Park, Hyunsik and Kim, Wonjoon}, journal = {Applied Economics Letters}, volume = {25}, number = {8}, pages = {513--517}, year = {2018}, publisher = {Taylor \& Francis}, }Generated with NotebookLM - JKTISThe Effect of Alliance Activity on Patent Litigation: In the Case of Printed ElectronicsJournal of Korea Technology Innovation Society, 2018
Patent litigation has been considered as a tool to protect and facilitate innovation. Ironically, yet, the misguided uses of patent litigation as a strategic tool for vigilance against competitors are acting as a hindrance for innovation. Previous studies show that the better the quality of a patent, the higher the chance of the patent being litigated. Therefore, it is particularly important for the innovating firms to take strategic precautions to minimize the risk of patent litigation. This study investigates the moderating role of firms’ past alliance experiences on the relationship between patent quality and patent litigation from the perspective of a defendant. A unique dataset on patents, infringement lawsuits, and firm performances in the printed electronics industry confirms that firms’ previous alliance experiences mitigate the impact of patent quality on infringement litigation. For instance, the results confirm that the presence of past alliance experience reduces the litigation rate by 33% for firms with median-quality patents. This paper makes two major contributions. First, it contributes to the literature on alliance experience by confirming its role as a reputation in mitigating future litigations. Second, this paper contributes to the literature on patent litigation by identifying a unique moderator, ie, alliance experience, on the linkage between patent quality and litigation. An innovating firm is likely to become an alleged infringer under a false accusation. Therefore, this paper focuses on firms that partake in infringement lawsuits unwillingly. Despite the importance, to the best of our knowledge, this is the first study to investigate patent litigations from the perspective of defendants.
@article{kang2018effect, kind = {publications}, title = {The Effect of Alliance Activity on Patent Litigation: In the Case of Printed Electronics}, author = {Kang, Minjeong and Yoo, Jaewon and Kim, Wonjoon and Kim, Namil}, journal = {Journal of Korea Technology Innovation Society}, volume = {21}, number = {1}, pages = {265--299}, year = {2018}, publisher = {Korea Technology Innovation Society}, }Generated with NotebookLM