Boundary information about census blocks, precincts, and districts can be critically valuable for research in a range of social science disciplines, including economics, political science, and finance. Recent research in household finance, for example, has used such data to investigate how foreclosures affect neighborhoods. In most countries, the boundary coordinates of geographical regions, such as voting precincts, are readily available for all to use. In many U.S. states, however, such information is astonishingly difficult to collect. This grant supports an initiative by Michael McDonald, one of the nationХs foremost experts on fine-grained political geography, to collect detailed data on the geospatial boundaries of U.S. electoral precincts and districts, to make that data available to researchers and the public, and to develop open, easy-to-use software that will facilitate the use and analysis of this data by researchers.

This grant funds a project by Stanford economist Liran Einav to study the factors influencing researchersХ decisions about when and how to share data. Topics to be studied include researchersХ incentives for data sharing, the perceived privacy risk if data are shared, the impact of data-sharing decisions on scientific progress, and the relevant policy implications. On the privacy issue, specifically, Einav will develop and train a machine learning algorithm to estimate the vulnerability of different datasets to privacy violations, a potentially useful tool in and of itself, as it would bring more objectivity to risk assessment than current processes, through which researchers and review boards try to estimate these risks subjectively. Using the algorithmic risk assessment and many other field-specific variables as inputs, Einav and his team will then develop econometric models of researchersХ data-sharing and data-hoarding decisions. The regression coefficients calculated from this estimation should permit causal insights about the relative importance of different factors in those decisions, as well as generate predictions of the effect of various policy changes on scientific progress. This research therefore stands to inform and empower the many existing efforts to encourage better data practices among academics.