The National Academy of Sciences (NAS) was chartered by Abraham Lincoln “to give advice to the nation.” And, man, does it ever. Commissioned studies released during the first few days of September 2016 alone, for example, cover everything from clean electric power options to molybdenum-99 production, from eye health to eldercare. Funders and clients alike know the Academy’s work to be prestigious, authoritative, and impartial, but slow, inefficient, and expensive. Internal studies of NAS operations conducted sporadically over the years have resulted in only modest modifications. Now the incoming president, Marcia McNutt, wants to do more than that. A former editor of Science magazine and the first woman ever elected to lead the Academies, she is committed to comprehensive reform of how the NAS functions. Her first step is commissioning an outside management study by a distinguished but independent panel. The National Academy of Public Administration has agreed to carry out the project. Funds from this grant provide partial support for this independent management study.

Empirical economists tell tales of woe about the difficulties of accessing, processing, linking, and analyzing administrative data. Many have tried to address such impediments independently in the course of this or that research project. A piecemeal approach, however, is less effective than what might be accomplished by working together. This grant supports a project by a team of empirical economists and technologists at Stanford University to build and staff a new Stanford Data Core that will reduce social scientists’ struggles, and enhance their triumphs, with administrative data. Led by principal investigator Mark Cullen, the team has already gained access to over 230 administrative datasets, more than 100 server racks, and petabytes of data storage. The team will begin by harmonizing, documenting, cleaning, and adding to these datasets and then moving computations on them to the cloud in collaboration with Google. After the data have been pulled together, the team will test this new computational environment through the launch of four pilot research projects covering topics in economics from economic opportunity to contract labor markets. Though interesting in themselves, the projects will primarily serve as useful test cases to measure the functioning of the new computational environment. Finally, the project team is particularly keen on finding, sharing, and standardizing solutions to the legal challenges that encumber research on administrative data. Working with university lawyers at Stanford, the team will model what routine nondisclosure and data use agreements can and should look like. They will then promote this resource and these contracts to the wider scientific community.

The importance of macroeconomic statistics compiled by the government, such as Gross Domestic Product (GDP) and the Consumer Price Index (CPI), is difficult to overstate. The data used to calculate these statistics are collected through surveys fielded by a host of differing government agencies, including the Bureau of Labor Statistics, the Bureau of Economic Analysis, and the Census Bureau. These surveys have significant and well-known methodological limitations leading to inaccuracies, substantial lag times, sampling distortions, and the need for (often significant) revisions. The worrying methodological basis of many economic statistics stands at odds with the increasingly high-quality data available about the economy. Vast improvements in the ability of retailers, for instance, to electronically track transactions provide a wealth of data for price and quantity measurement that is several orders of magnitude richer than currently captured in government surveys. This grant funds a pilot project by a team led by John Haltiwanger at the University of Maryland, College Park, to develop new, more accurate, and more timely methods to calculate portions of GDP and CPI using administrative data collected by retail firms. Partnering with several large retailers, the team will compile a large set of administrative data bearing on retail prices and quantities produced and sold, document how these data can be acquired and harvested, use the data to calculate portions of CPI and GDP, and then issue a report comparing and contrasting these calculations along several dimensions with current methodologies. Though it is too much to hope that such a project will, by itself, change the way government economic statistics are calculated, this project is an important proof of concept demonstrating one potential path to what all agree is badly needed reform.

This grant funds a dozen conferences to be held over two years that are designed as community-building exercises for a new Administrative Data Research Network (ADRN). The ADRN is a linked network of Administrative Data Research Facilities, research centers devoted to facilitating the use of administrative data by researchers by lowering transaction costs and increasing the reproducibility of research conducted on administrative data. Conference topics will focus on common difficulties encountered when working with administrative datasets, including securing data access, privacy and anonymity, data ethics, documenting and versioning data, data use agreements, and archiving. Conference outreach will target not only researchers, but key officials in government and industry who have access to administrative data. If successful, these conferences will serve as important gatherings to build consensus around standards, priorities, and best practices in the growing community of researchers working with administrative data.

Though considered by most economists to be the gold standard method for testing hypotheses, randomized controlled trials (RCTs) are nonetheless difficult to design, implement, or interpret in the field. That is why the Abdul Latif Jameel Poverty Action Lab (J-PAL) was founded at MIT in 2003. J-PAL has been a staunch champion of RCT methods, providing funds and expert guidance in the design and implementation of RCTs. Initially focused on developmental economics, the J-PAL network expanded its focus to the U.S. with the 2013 launch of J-PAL North America. The goal of this new network is to promote rigorous study of economic issues in the United States and its neighbors, both by catalyzing high-quality research directly and by strengthening the capacity of institutions and individuals to conduct and understand such work. This grant provides three years of core operating support for J-PAL North America’s activities. Supported activities include salary support for core J-PAL staff, publication costs, training expenses, and development of workshops and toolkits for researchers wishing to field RCTs in North America.

Suppose you, as a researcher, have succeeded in wrangling important and sensitive data from a government or corporate source, data that is valuable to the research community. Because your data are sensitive, however, you want to share only with appropriately trained and responsible scholars who can be trusted to treat the data ethically. Suppose now a request comes in from someone who wants to study the data. What do you do? Not every researcher is savvy about the technical, privacy, or legal compliance issues related to sensitive administrative data. You could investigate the individual and draft an agreement for them to sign. But starting from scratch to answer each new request, with all the associated inefficiencies, is time consuming and costly. Wouldn’t it be much better if you could begin instead by asking your data seeker for some standard researcher certification, a kind of Good Housekeeping Seal of Approval that would indicate what kind of training and track record they have? This grant funds a project by Maggie Levenstein, executive director of the Interuniversity Consortium for Political and Social Research (ICPSR), to design, test, implement, and promote just such a researcher credentialing system for use by her own and other institutions holding sensitive administrative data. The widespread adoption of such a system could significantly decrease the transaction costs associated with access to administrative data, increase the analysis of important though sensitive datasets, and promote responsible training and research protocols concerning preregistration, anonymization, reproducibility, and other research practices.

One goal of the Foundation’s grantmaking in economics is to lower barriers that impede obtaining and using naturally occurring datasets for sound and reproducible research. One strategy is to identify, strengthen, and, in some cases, create intermediaries who can manage the relationships between data generators and academic researchers. We refer to such an intermediary as an Administrative Data Research Facility (ADRF). These facilities will, in turn, work more effectively if they can be linked together into a network that facilitates the sharing of standards, best practices, and data among ADRFs. Such networks, however, require a dedicated coordinator to ensure their proper functioning. This grant to the University of Pennsylvania provides two years of (partial) salary support for a network coordinator devoted to servicing the growing needs of the ADRF community. Although this network organizer will work with various ADRFs, he or she will be initially hosted by the University of Pennsylvania’s Actionable Intelligence for Social Policy (AISP) project, headed by Dennis Culhane. AISP has emerged as a leader in developing practical procedures and protocols for conducting research that uses administrative data and is thus well positioned within the community to successfully host this important position.

Better management of the economy requires better macroeconomic models that can be used to predict the consequences of this or that economic policy choice, proposal, or regime. Better economic models, in turn, require the ability to compare, contrast, and evaluate the predictions of various models across a variety of scenarios. How do professional economists, financial regulators, and central bankers compare one macroeconomic model to another? As it turns out, they don’t. Not only are there no widely accepted ways to compare models, there is little agreement about what criteria make one model better than another. Enter economist Volker Wieland, who is approaching this problem head-on. Collaborating with a wide network of economists in Europe and the U.S., Wieland has developed what he calls the Macro Model Database (MMD), a software platform that can upload, store, and run different macroeconomic models, allowing researchers to confront models with the same historical or synthetic scenarios and compare the predictions subsequently produced. Funds from this grant support a project to make the Macro Model Database fully open source, documenting and disseminating the underlying code to any expert interested in using it. By making it easier for researchers to upload and explore models, Wieland and his team (including Stanford economist John Taylor) plan to double the number of models available on its platform. Additional grant funds support the establishment of a research network based at the Center for Economic Policy Research (CEPR) in London to further promote the MMD’s use within the research community.