Erika Check Hayden at Nature: “The data contained in tax returns, health and welfare records could be a gold mine for scientists — but only if they can protect people’s identities….In 2011, six US economists tackled a question at the heart of education policy: how much does great teaching help children in the long run?
They started with the records of more than 11,500 Tennessee schoolchildren who, as part of an experiment in the 1980s, had been randomly assigned to high- and average-quality teachers between the ages of five and eight. Then they gauged the children’s earnings as adults from federal tax returns filed in the 2000s. The analysis showed that the benefits of a good early education last for decades: each year of better teaching in childhood boosted an individual’s annual earnings by some 3.5% on average. Other data showed the same individuals besting their peers on measures such as university attendance, retirement savings, marriage rates and home ownership.
The economists’ work was widely hailed in education-policy circles, and US President Barack Obama cited it in his 2012 State of the Union address when he called for more investment in teacher training.
But for many social scientists, the most impressive thing was that the authors had been able to examine US federal tax returns: a closely guarded data set that was then available to researchers only with tight restrictions. This has made the study an emblem for both the challenges and the enormous potential power of ‘administrative data’ — information collected during routine provision of services, including tax returns, records of welfare benefits, data on visits to doctors and hospitals, and criminal records. Unlike Internet searches, social-media posts and the rest of the digital trails that people establish in their daily lives, administrative data cover entire populations with minimal self-selection effects: in the US census, for example, everyone sampled is required by law to respond and tell the truth.
This puts administrative data sets at the frontier of social science, says John Friedman, an economist at Brown University in Providence, Rhode Island, and one of the lead authors of the education study “They allow researchers to not just get at old questions in a new way,” he says, “but to come at problems that were completely impossible before.”….
But there is also concern that the rush to use these data could pose new threats to citizens’ privacy. “The types of protections that we’re used to thinking about have been based on the twin pillars of anonymity and informed consent, and neither of those hold in this new world,” says Julia Lane, an economist at New York University. In 2013, for instance, researchers showed that they could uncover the identities of supposedly anonymous participants in a genetic study simply by cross-referencing their data with publicly available genealogical information.
Many people are looking for ways to address these concerns without inhibiting research. Suggested solutions include policy measures, such as an international code of conduct for data privacy, and technical methods that allow the use of the data while protecting privacy. Crucially, notes Lane, although preserving privacy sometimes complicates researchers’ lives, it is necessary to uphold the public trust that makes the work possible.
“Difficulty in access is a feature, not a bug,” she says. “It should be hard to get access to data, but it’s very important that such access be made possible.” Many nations collect administrative data on a massive scale, but only a few, notably in northern Europe, have so far made it easy for researchers to use those data.
In Denmark, for instance, every newborn child is assigned a unique identification number that tracks his or her lifelong interactions with the country’s free health-care system and almost every other government service. In 2002, researchers used data gathered through this identification system to retrospectively analyse the vaccination and health status of almost every child born in the country from 1991 to 1998 — 537,000 in all. At the time, it was the largest study ever to disprove the now-debunked link between measles vaccination and autism.
Other countries have begun to catch up. In 2012, for instance, Britain launched the unified UK Data Service to facilitate research access to data from the country’s census and other surveys. A year later, the service added a new Administrative Data Research Network, which has centres in England, Scotland, Northern Ireland and Wales to provide secure environments for researchers to access anonymized administrative data.
In the United States, the Census Bureau has been expanding its network of Research Data Centers, which currently includes 19 sites around the country at which researchers with the appropriate permissions can access confidential data from the bureau itself, as well as from other agencies. “We’re trying to explore all the available ways that we can expand access to these rich data sets,” says Ron Jarmin, the bureau’s assistant director for research and methodology.
In January, a group of federal agencies, foundations and universities created the Institute for Research on Innovation and Science at the University of Michigan in Ann Arbor to combine university and government data and measure the impact of research spending on economic outcomes. And in July, the US House of Representatives passed a bipartisan bill to study whether the federal government should provide a central clearing house of statistical administrative data.
Yet vast swathes of administrative data are still inaccessible, says George Alter, director of the Inter-university Consortium for Political and Social Research based at the University of Michigan, which serves as a data repository for approximately 760 institutions. “Health systems, social-welfare systems, financial transactions, business records — those things are just not available in most cases because of privacy concerns,” says Alter. “This is a big drag on research.”…
Many researchers argue, however, that there are legitimate scientific uses for such data. Jarmin says that the Census Bureau is exploring the use of data from credit-card companies to monitor economic activity. And researchers funded by the US National Science Foundation are studying how to use public Twitter posts to keep track of trends in phenomena such as unemployment.
….Computer scientists and cryptographers are experimenting with technological solutions. One, called differential privacy, adds a small amount of distortion to a data set, so that querying the data gives a roughly accurate result without revealing the identity of the individuals involved. The US Census Bureau uses this approach for its OnTheMap project, which tracks workers’ daily commutes. ….In any case, although synthetic data potentially solve the privacy problem, there are some research applications that cannot tolerate any noise in the data. A good example is the work showing the effect of neighbourhood on earning potential3, which was carried out by Raj Chetty, an economist at Harvard University in Cambridge, Massachusetts. Chetty needed to track specific individuals to show that the areas in which children live their early lives correlate with their ability to earn more or less than their parents. In subsequent studies5, Chetty and his colleagues showed that moving children from resource-poor to resource-rich neighbourhoods can boost their earnings in adulthood, proving a causal link.
Secure multiparty computation is a technique that attempts to address this issue by allowing multiple data holders to analyse parts of the total data set, without revealing the underlying data to each other. Only the results of the analyses are shared….(More)”
