In the last decade, local governments have greatly expanded their use of administrative records for management of programs as statistical files to evaluate the results of program choices, to determine priorities among needs, to challenge anecdotal evidence used to make policy, and to make strategic plans.

As results from the Census 2000 long form and the American Community Survey are released, more analysts are making comparisons with administrative data.

We expect estimates from the two surveys to differ from administrative records. It isn’t that the results from one data set are “right” and the results from the other data set are “wrong.” Both have weaknesses and strengths, and the data are collected in different ways, for different purposes, and have different types of errors.

The paper examines reasons for differences, including data collection methods, sources of error, confidentiality, and differences in universes, coverage, time periods, and questions. Even when concepts seem that they should be similar, such as the number of poor children and the number of children receiving public assistance, it is comparing the proverbial apples and oranges and ending up with kumquats.