Administrative data

Administrative data is data generated as a byproduct of running a government program or institution, tax filings, benefit enrollment records, school attendance, hospital admissions, unemployment insurance claims, rather than data collected specifically for research through a survey.

Because administrative data covers everyone who interacts with a program, not a sample, it can avoid the sampling and response rate problems that surveys face, and it often spans long time periods at low marginal cost. But it is structured around an agency's operational needs, not a researcher's question, so fields can be inconsistent, sparsely documented, or missing exactly what a study needs. Researchers frequently combine records across agencies, or with survey data, through record linkage, and increasingly use administrative data to build finer-grained estimates through small area estimation. Some of it is also released publicly as open data, though most is not, due to privacy and legal restrictions on the underlying records.

Administrative data underpins much of modern program evaluation because it is cheaper and often more complete than new data collection, and it captures real behavior, a benefit actually paid, a re-arrest that occurred, rather than a self-reported survey answer. The main pitfall is that definitions and data quality can shift whenever a program's rules or systems change, which can look like a genuine trend in the data when it is really a measurement artifact.