This grant funds a project led by Brian Granger to expand and enhance Jupyter Notebooks—a powerful, popular, and open-source scientific computing platform—to enhance its handling of private, proprietary, or otherwise sensitive data. Planned features to be developed and implemented include a nuanced permissions structure that can be used to ensure that only properly credentialed individuals can see or manipulate data, stronger event logging and internal telemetry, and encryption of data both at rest and in transit. To test the software, Granger will work closely with a wide variety of data holders including a U.S. company dealing with health data; a German warehouse for financial data; and NYU’s Center for Urban Science and Progress (CUSP), which collect all kinds of municipal data. Grant funds support software development along with a host of dissemination activities that will promote and test Jupyter’s expanded capabilities among researchers dealing with sensitive data.

This grant provides funds for the creation and operation of the Administrative Data Research Institute (ADRI), a national membership organization designed to provide services and set standards for the global network of Administrative Data Research Facilities (ADRFs). An ADRF is a data intermediary, i.e., an institution that facilitates researcher access to private or sensitive data owned or held by corporations and government entities. To become a member of the new ADRI, member ADRF’s would have to provably maintain high data standards and practices, particularly regarding data privacy and security. In return, the ADRI would provide member institutions with expert advice and leadership concerning the many technical, legal, political, or privacy challenges that data intermediaries face. Above all, the ADRI would help engender trust in at least three domains where it is much needed: among the member intermediaries when it comes to arrangements for linking their data; among the data producers when it comes to negotiating data use agreements with member intermediaries; and among the public and their policymakers when it comes to allowing administrative data containing sensitive information to be reused through member intermediaries for research purposes. Grant funds support initial start-up and operational costs of the new network for a period of two years.

Funds from this grant support three different large-scale and randomized field experiments, devised by economists Matthew Gentzkow at Stanford and Hunter Allcott at New York University, that aim to help us better understand the various ways social media and smartphones impact our lives. In the first, Gentzkow and Allcott study the welfare implications of Facebook usage by measuring the impact of cutting off access to that platform. In the second experiment, the team studies how limiting smartphone access to social media and other apps affects the welfare of college students. The third experiment concerns the demand for “fake news” and investigates what happens to consumer behavior when incentives to find accurate information increase. Grant funds support the fielding and analysis of all three experiments as well as documentation of methods to enable easy replication by other scholars.

Nudging consists of changing how alternatives are presented in ways designed to help choosers select what they say they would ideally want. Substantial empirical evidence shows that nudges can significantly—and sometimes substantially—modify behaviors. Yet much remains mysterious about nudging, especially with regard to heterogeneous effects and behavioral welfare economics. How is it possible to create effective choice architectures when not everyone should be nudged the same way? This grant funds a series of experiments by Eric Johnson of Columbia University. He has devised a series of experiments about what he calls “smart nudges.” In contrast to the one-size-fits-all approach, these take into account how individuals differ. One idea is to set different defaults for different people depending on their background characteristics. Another technique, called “preference checklists,” presents a prospective decision-maker with a list of criteria that other people sometimes take into account when considering similar choices. Checking off the criteria that the decision-maker thinks should apply in this case is a way of bringing to mind factors that might otherwise be forgotten or ignored. Johnson’s hypothesis is that choice architectures that incorporate smart defaults and preference checklists will be welfare-enhancing compared to traditional one-size-fits-all nudges. Grant funds support the fielding and analysis of these experiments, and the publication of two papers on the results.

In economics, the assumption called “rational expectations” holds that the value of a variable expected by agents in a model matches the expected value predicted by the model. As a description of the decision-making processes of actual, real-life people, rational expectations models seem unrealistic. The idea nevertheless dominates much of contemporary economics, not least because it vastly simplifies the modeling process. This grant funds work led by Nobel laureates Tom Sargent and Lars Hansen to develop a robust, mathematically rigorous alternative to rational expectations models. Sargent and Hansen envision agents not as utility maximizers with fully rational expectations, but rather as decision-makers using Bayesian reasoning to update their beliefs in light of uncertainty and limited information. The approach, while technically daunting, is very promising as a better description of real life. Grant funds will support the development, debugging, and dissemination of new models and of the software for analyzing them as well as a workshop, conference, and travel expenses for junior scholars participating in the project.

Women economists cluster more in certain subfields (e.g., labor, health, public economics) than in others (e.g., macro, theory, or international trade). This grant funds work by labor economists Ron Oaxaca and Eva Sierminska to analyze this gendered difference in career choices among professional economists. Rigorously analyzing why men and women make career decisions differently requires holding unchanged as many factors as possible besides gender. Focusing on career choices in graduate school allows just that. Ph.D. candidates in the same program, for example, interact with the same group of professors and advisors, expect similar sets of potential salaries, and face the same specialization options. Grant funds are supporting the creation of a public use dataset, analysis of that dataset, development of a website about the project, and the production of two research articles and a less technical policy brief reporting the results of the analysis.

According to a Sloan-funded survey, Chief Privacy Officers at major corporations like Amazon, AT&T, Comcast, Facebook, and General Electric say they are willing to share private data with independent scholars under two conditions. First, they want to take risks and institute policies that are similar to their peers. Second, to the extent that shared data helps academics gain insight into important societal issues, they want to be widely recognized for the role they played. This grant funds a project by the nonprofit Future of Privacy Forum (FPF) to launch a peer-to-peer network of private sector C-suite leaders called the Corporate-Academic Data Stewardship Research Alliance. Members will collaborate in order to encourage, recognize, and honor model data-sharing arrangements between the private sector and academe. The network will hold regular phone conferences as well as quarterly in person meetings to discuss how to overcome the legal, practical, and technical barriers that hinder cooperation with independent scholars. The network will also develop and publicize a CEO-level award honoring those corporate leaders who have made pioneering data-sharing agreements with independent scholars. Grant funds will help defray operating costs of the new network for 18 months.

This grant funds work by behavioral economists Erzo Luttmer from Dartmouth and Dmitry Taubinsky from University of California, Berkeley, who are analyzing how behavioral biases might explain the annuity puzzle—the observation that annuities, as a financial product, are not nearly so popular as economic theory and people’s stated preferences would predict them to be. Of the many possible explanations for the annuity puzzle, behavioral biases are not easy to study. Consumers who avoid annuitization might have privileged information and unobserved motivations (like making bequests, for example), or they may be systematically affected by behavioral biases that result in suboptimal choices. Teasing out what accounts for what is not just difficult, but also important for devising potential remedies. Luttmer and Taubinsky have carefully designed a series of controlled experiments where real economic incentives are at stake, but where most other complications from the real world have been abstracted away, thus allowing them to isolate how subjects’ psychological attitudes toward time and risk affect decision-making. By identifying the role behavioral mechanisms play in such decisions, this research will generate new insights about the annuity puzzle in particular, and also about the behavioral welfare economics of risk taking, saving decisions, and insurance markets more generally.

Behavioral economists assume that people make decisions based on the perceived probabilities of events. Behavioral experiments, interpretations, and the policies they inform should therefore depend on information about popular perceptions about likelihoods. This grant funds work by Paul Slovic of the University of Oregon and co-principal investigator Howard Kunreuther of the Wharton School to field surveys that will collect data on public perceptions of the probabilities for a host of important events, including nuclear war, chemical attack, opioid addiction, school shootings, as well as the mass adoption of driverless cars or e-cigarettes. Opinions about more than a hundred hazards will be elicited. In addition, Slovic and Kunreuther will conduct textual analyses based on the frequency that Google News describes a given hazard using words with high emotional valence. Last, the team will field a series of experiments designed to probe how people act on those perceptions and what can be done to help everyone make better estimates and better decisions.