Key Areas of Research
Nondisclosure agreements and externalities from silence
PNAS (Proceedings of the National Academy of Sciences)
How do contractual restrictions on worker voice affect information flows about employers? We develop a framework in which the legal risk from violating a nondisclosure agreement (NDA) reduces the willingness of workers to share negative information, making it more difficult for high-road employers to differentiate themselves to workers. Empirical support for these ideas comes from studying the relationship between NDA use and the content of Glassdoor reviews after three states prohibited employers from using NDAs to conceal unlawful conduct. By curtailing the flow of negative information, NDAs impose negative externalities on workers who value such information and on competing employers who are less able to stand out.
Jason Sockin, Cornell, Aaron Sojourner, UpJohn, Evan Starr, UMD
Data and methods for identifying artificial intelligence-related patents
Research Policy
This paper evaluates existing approaches to identifying artificial intelligence (AI)-related patents and introduces a novel, scalable framework to improve classification performance. Motivated by the growing reliance on patent data in innovation research, we assess widely used methods and document their substantial performance limitations. To address these challenges, we develop a CPC-informed label-refinement framework inspired by positive-unlabeled (PU) learning to construct high-quality training data. Using this refined dataset, we train a range of machine learning, deep learning, and transformer-based models. Our models substantially outperform existing approaches, including those employed by the USPTO. We further demonstrate the empirical value of more accurate AI patent identification through two applications. First, we find that the release of ChatGPT increased the market valuation of AI patents. Second, we show that firms increased the allocation of innovative effort toward AI technologies following its release. To facilitate future research, we release our training data, source code, and a novel dataset of patent-level predictions (AIPat), which will be continuously updated to reflect the evolving nature of AI innovation.
Tianjun Wu (Nanjing University),
Chao Min (Nanjing University),
Waverly W. Ding (University of Maryland),
Guolong Wang (University of International Business and Economics),
Kunpeng Zhang (University of Maryland)
Clause and Effect: Theory and Field Experimental Evidence on Noncompete Clauses
Quarterly Journal of Economics
We study worker noncompete clauses in a large field experiment with two finance firms. Across ~14,000 job offers to freelance recruiters on short-term contracts, we randomize wages and the presence, salience, and duration of noncompetes (all contracts also included a nondisclosure agreement). Removing a noncompete increases mobility between competing employers by 36–52% and raises workers’ total earnings from the two firms by 12–17%. We find no evidence—rejecting even small effects—that removing noncompetes generates secret leakage. We also find no evidence that workers choose noncompete jobs for higher pay. Many workers appear unaware of noncompetes before firms’ post-employment communication. The results align with a model of inattention and uncertainty about enforcement.
Bo Cowgill (Columbia), Brandon Freiberg (INSEAD), Evan Starr (UMD)
From Hotspots to Help-Spots: An AI-enabled, Community-centered Risk-resource Mapping
This project will focus on Baltimore City and/or Prince George’s County, with the final study site selected in consultation with researchers and community partners. The project’s central contribution will be the integration of two complementary geospatial frameworks: risk terrain and resource terrain. The risk terrain layer will identify places where firearm violence risk may be elevated based on spatial, environmental, social, and temporal indicators, such as prior shootings, vacant properties, limited lighting, transportation corridors, housing instability, alcohol outlets, commercial activity nodes, and other built-environment or neighborhood conditions (Brantingham & Brantingham, 1995; Caplan & Kennedy, 2016; Kennedy et al., 2011). The resource terrain layer will identify places where prevention capacity already exists or could be strengthened, including community violence intervention programs, schools, libraries, recreation centers, faith institutions, small businesses, health providers, youth-serving programs, neighborhood associations, and trusted informal leaders.
Principal Investigator: Kunpeng Zhang
Understanding Firms’ AI Efforts and Their Economic Impact
The Review of Corporate Finance Studies
Our research asks a central question for businesses investing heavily in artificial intelligence: Is AI actually making firms more productive, and if so, how? Using 15 years of data on AI-skilled employment at U.S. public firms, we find that firms increasing their AI investments experienced significantly faster productivity growth from 2018 to 2024—roughly one percentage point of additional productivity growth per year for a one-standard-deviation increase in AI investment. Importantly, these gains do not appear immediately; they build gradually over several years.
The key mechanism is organization capital: durable, firm-specific knowledge embedded in a company’s systems, processes, data, and workflows. AI appears to create the most value when firms use it not simply to automate individual tasks, but to build reusable organizational capabilities—for example, systems that codify proprietary knowledge, improve decision-making, or redesign business processes. Productivity gains are concentrated in AI jobs that build this organization capital, and are especially strong at firms that began with less of it.
Tania Babina, Associate Professor of Finance, UMD
From Hypothesis Testing Towards Inference to Best Explanation: A PEEBI Testimonial Structure for Abductive Studies in Strategy
Strategic Management Journal
Though scholars employ abduction, there is no agreed-upon structure to reporting their findings. Moreover, the traditional hypo-deductive reasoning structure does not align with the epistemology of abduction. We propose an abductive testimonial structure, termed PEEBI, which consists of five sections in which the authors take prior knowledge and theories, establish the context and observations that are worthy of interest, identify and evaluate candidate explanations for the observed patterns, determine the best explanation and their reasoning for accepting it, and abstract the best explanation to a more generalizable theoretical contribution. Consistent with abductive epistemology, PEEBI advances knowledge in a modest, stepwise fashion. PEEBI foregrounds transparency and communication of the author's judgment and elevates the role of the readers by giving them information to better make judgments.
Sandeep Devanatha Pillai (Tulane), Brent Goldfarb, David A. Kirsch, Evan Starr, and Seojin Kim (Drexel)
An Unattainable Emotional Advantage? Supervisors' Perceptual Biases in the Relationship Between Emotional Understanding Ability and Performance Ratings for Female Employees
Journal of Organizational Behavior, July 2027
Empathy is increasingly valued at workplaces as it facilitates interpersonal relationships, collaboration, and team functioning. Accordingly, organizations increasingly observe and recognize employees' emotional understanding (EU) ability—a key foundation of empathy—as an important dimension of individual merit in performance evaluations. We note that, however, supervisors' gender stereotypes may constrain whether EU ability is recognized properly, particularly for female employees. Drawing on dual-process models of social cognition, we theorized that the female empathy stereotype—the belief that women are inherently more empathetic than men—leads supervisors to attribute female employees' EU ability to gender-typical traits rather than to individual merit. As a result, the supervisors' female empathy stereotype weakens the link between observed emotional intelligence (EI) ability and job performance ratings. Across a video vignette experiment (Study 1) and a field study (Study 2), we found that the relationship between supervisors' perceptions of EU ability and their ratings of job performance was consistently weaker for female than for male employees. We conducted another vignette experiment (Study 3) to test three-way interactions with two moderators (i.e., employee gender and raters' female empathy stereotype) and found that supervisors' female empathy stereotype attenuated the relationship for female employees but not for males. Our findings extend research on gender stereotypes, empathy, and EI by demonstrating how a widely shared gender stereotype can discount female employees' valuable emotional abilities.
Hanbo Shim (TU Arlington) | Myeong-Gu Seo (UMD) | Joo Hun Han (KAIST) | Sirkwoo Jin (Merrick College)
Behavioral Intervention Construal: A Framework for Understanding Inferences from Behavioral Interventions
Organization Science, August 2026
Managers and policymakers frequently use behavioral interventions—including incentives and messaging campaigns—to influence people’s behavior. They often choose an intervention by asking whether it will make a desired behavior easier, cheaper, or more attractive. Our research suggests they should also ask a second question: “What could this intervention unintentionally communicate?” For instance, interventions can signal that an organization is self-interested or trying to control people’s choices.
The paper introduces a framework for understanding the inferences people draw from behavioral interventions. By identifying these inferences, the framework helps leaders better predict when interventions will succeed or fail, while also broadening evaluation to outcomes beyond behavior, such as trust and satisfaction. Finally, we provide a publicly available AI tool that managers and policymakers can use before launching an intervention to anticipate the inferences people might draw and revise its design accordingly.
Joseph Reiff, Assistant Professor of Marketing, Robert H. Smith School of Business; Jon Bogard, Assistant Professor of Organizational Behavior, Olin Business School
The Relative Effects of Design Thinking Versus After-Action Review on Team Performance: An Experiential/Episodic Team Learning Perspective
Journal of Applied Psychology, October 2025
Learning is key to teamwork success. In a field experiment, a design thinking intervention was shown to promote effective team learning, leading to greater team performance and financial savings. Adopting a learning orientation -- emphasizing hypothesis development and testing foir improving work efficiency - was shown to be is key to teamwork success.
Jingqiu Chen (Shanghai Jiao Tong University), Dana Vashdi (University of Haifa), Qingyue Fan (Shanghai Jiao Tong University), Peter Bamberger (Tel Aviv University), and Gilad Chen (University of Maryland)
Prompt Adaptation as a Dynamic Complement in Generative AI Systems
Information Systems Research, April 2026
As generative-AI models become more powerful, organizations will only realize a portion of that improvement unless users learn to adjust how they interact with the models — prompt-adaptation becomes a critical skill for unlocking full value.
To capture the full benefits of new AI technologies, companies should invest not only in the latest models and infrastructure, but also in user training and workflow design — enabling teams to use the new technologies effectively.
Eaman Jahani (UMD), Benjamin S. Manning (MIT), Joe Zhang (Stanford), Hong-Yi TuYe (MIT), Mohammed Alsobay (Microsoft), Christos Nicolaides (University of Cyprus), Siddharth Suri (Microsoft), David Holtz (Columbia)