Poster Paper: Using Administrative Data for Impact Evaluations

Saturday, November 5, 2016
Columbia Ballroom (Washington Hilton)

*Names in bold indicate Presenter

Laura Feeney, Julia Chabrier, Michelle Woodford, Jason Bauman and Geetika Mehra, Massachusetts Institute of Technology

Administrative data are an excellent source of information for research and impact evaluation with the potential to make government more effective. Compared to survey data, these data are often easier and cheaper for researchers to obtain, provide near-universal coverage of relevant populations, and are less susceptible to many sources of bias (Finkelstein and Taubman, 2015). Meyer, Mok, and Sullivan (2015) advocate for the increased use of administrative data for research because of its accuracy and completeness. They declare household surveys to be “in crisis” due to reduced participation and declining accuracy. Meyer and Mittag (2015) find that income from transfer programs are more accurately reported in administrative data than the Current Population Survey.

Despite these benefits, administrative data – especially linked administrative data – are underutilized in research and impact evaluation. Finkelstein and Taubman (2015) find that, among randomized controlled trials in the field of healthcare delivery, most involved primary data collection, only half used administrative data, and only fifteen percent relied exclusively on them.

This under-utilization may be due in part to the complex web of statutes, laws, and policies that govern the release of identified data. Even when the regulations allow for the release of data for research purposes, researchers and/or policymakers are often unclear about the process and strategies for sharing data, and policy-relevant research often fails to launch. Access to de-identified data is often much simpler, but complex and policy-relevant research often requires the linkage of multiple sources of administrative data, or linkage with survey data, which requires linking the data at the individual level.

This report provides practical guidance about obtaining and using nonpublic administrative data for randomized impact evaluations. We illustrate a set of strategies for accessing data while complying with relevant regulations, describe the ethical and legal landscape, and review common challenges and potential for bias in using administrative data for impact evaluation.

The President’s budget for FY 2017 notes the underutilization of administrative data for evaluation of government programs, and proposes several strategies to increase the availability of administrative data to researchers. In March 2016, President Obama signed a bill creating an Evidence-Based Policymaking Commission, which will identify optimal strategies for linking and releasing administrative data for research and evaluation purposes. While these efforts may prove to be transformative, understanding the current possibilities for acquiring, accessing, and analyzing administrative data could facilitate more research leading to better policies today.