This grant provides two years of operational support for the Center on Regulation and Markets, a project of the Brookings Institution that provides independent, non-partisan research on regulatory policy, applied broadly across microeconomic fields. Led by Sanjay Patnaik, the Center creates and promotes independent economic scholarship to inform regulatory policymaking, the regulatory process, and the efficient and equitable functioning of markets. Research supported by the Center addresses a number of pressing issues in regulatory economics, including financial markets, emerging technologies, consumer protection, regulatory processes, data privacy, common ownership, and how to accurately measure market power. Grant funds will allow the Center to commission four policy papers and four policy briefs on topics in regulatory economics; hold eight events aimed at disseminating research to academics, policymakers, regulators, the press and the public; and maintain and update the Center’s website as a general dissemination hub for information about the Center, its research, events, and other activities.

This grant funds work by a team led Francesca Dominici and Jose Zubizarreta to develop new methods and techniques that will increase the robustness and power of randomized controlled trials (RCTs) as a method for investigating causal relations across a diverse range of phenomena. Long-regarded as the gold standard in social scientific research, the randomized controlled trial has virtues in abundance. By randomly sorting participants into control and treatment groups, researchers using RCTs can, in theory, ensure that these groups are statistically indistinguishable. This allows them to conclude that differences later observed between these two groups must have been caused by the treatment. This works beautifully in principle. In practice, however, drawing causal inferences using RCTs can be bedeviled by a number of factors, all involving how statistical averages never tell the whole story. When the population under study is very diverse, for instance, randomly sorting participants into control and treatment groups may be insufficient to ensure the two groups are identical across all variables. In other instances, control and treatment groups may be insufficiently isolated from one another, allowing outcomes caused by the treatment to spillover into the control group. In other cases, the effect of a treatment within the treatment group may be unequally distributed. A treatment that benefits a few people greatly while leaving most people worse off, say, might appear to have a positive benefit on average, leading researchers to miss important facts about how that average benefit is generated. Dominici and her team will develop and test new statistical methods that, if successful, will help researchers design RCTs in ways that head off each of these problems and allow the design of RCTs that can be more reliably used to make causal inferences. Their results will be distributed through academic papers, talks at professional meetings, and through open-source software tools available to be downloaded by researchers.

The mathematical theory of differential privacy describes methods and practices that can be implemented that allow researchers to query datasets with sensitive information while monitoring how much each query threatens the privacy of the individuals contained in the dataset. Differentially private methods are the current cutting edge of privacy-protecting science, yet, they are often mathematically complex and difficult to implement for those not versed in them. Widespread use of these methods will require mediating institutions that lower the cost of adoption, trusted places where researchers can download easy-to-install and easy-to-use software applications that will allow them to apply differentially private firewalls to sensitive data. In response to this need, Harvard computer scientist Salil Vadhan has created OpenDP, a dedicated community of theorists, engineers, practitioners, and privacy experts that are aiming to increase adoption of differential privacy by producing an open source suite of flexible, tested, and industrial-strength software components that makes implementing differential privacy both straightforward and trustworthy. Funds from this grant will support the effort, allowing Vadhan to further develop the library of general-purpose differential privacy algorithms, attract new experts to the collaboration, form new partnerships with corporations interested in protecting sensitive data, promoting awareness of the collaboration and its tools, and holding an annual meeting of stakeholders and users from academia, government and industry.

Experiments conducted online, using subjects that participate via web interface instead of physically traveling to a lab, have significant advantages over their in-person counterparts. They are cheaper to field, for instance, and they can draw from a more diverse pool of potential test subjects, making inferences from their findings more robust. Setting up such virtual experiments, however, is not easy. Existing software support packages for online experiments have been developed as “end-to-end” platforms that are not very flexible, much less interoperable. Customizing this software to meet the eccentricities of any given experiment often requires special programming skill and patience. Funds from this grant support a project, led by Duncan Watts at the University of Pennsylvania and Abdullah Almaatouq at MIT, that seeks to make fielding online experiments easier for researchers of all kinds. Watts and Almaatouq aim to develop the first modular virtual experiment platform, one that subdivides an experimental design into independent, though interoperable, parts with standard interfaces. Friendly graphical controls will enable researchers and administrators to customize, reuse, and improve each module of an experiment without writing new code. One feature, for example, will be automatic recruiting tools that simplify the location and retention of participant panels that are large, diverse, and representative. Grant funds will allow Watts and Almaatouq to develop and launch an entire virtual environment that will facilitate running social science experiments that are faster, cheaper, more scalable, more complex, and more realistic than could take place in a physical laboratory. All code for this software environment, in addition to accompanying documentation, tutorials, and webinars, will be made freely available through a professional archive that makes it easy for experimenters to both preregister their experiments as well as to deposit their code, data, and documentation. Additional grant funds will support outreach and adoption activities designed to encourage use of the platform and to begin to build an open-source community of developers devoted to its maintenance and improvement.

This grant funds a series of experiments designed by University of California, Berkeley economist Dmitry Taubinsky, to examine the welfare effects of nonfinancial policy interventions (NPIs). NPIs, more commonly known as “nudges,” are policy interventions designed to increase the attractiveness of prosocial behaviors through means other than decreasing the financial cost of that behavior. Examples include informational labels on products, salient reminders, default options, and praise and public recognition for desired behavior. Because nudges involve motivating people using nonfinancial means, quantifying the costs and benefits of NPIs is conceptually challenging. What if some people really liked what they were doing before they were nudged? What if some people don’t like being nudged at all, regardless of whether what is suggested would be good for them? What’s needed is a larger theoretical framework for evaluating the costs and benefits of nudges, a framework that includes the larger psychological costs that nudging may impinge on those who find themselves subject to it. In a series of experiments, Taubinsky will begin to develop such a framework, focusing on three issues.  First, what is the proper way to measure whether information nudges are well targeted in the sense that they change the behavior of people making the biggest mistakes? Second, what is the proper way to measure the psychological costs and benefits of motivating behavior by leveraging shame and pride through public recognition?  And third, what is the proper way to measure the discomfort that some people experience when moral suasion and other social factors create demands to act in prosocial ways? Grant funds will allow Taubinsky to field a series of experiments on each topic, along with a detailed analysis of his findings. Three peer-reviewed articles are anticipated.