Ms. Taeuber cautioned the audience about overly simplified direct comparisons of survey data, such as from the American Community Survey or the decennial census, with administrative data. There are crucial differences in concepts and data collection methods among data sets. As such, estimates of population characteristics from surveys such as the decennial census and the American Community Survey will differ. For example, Ms. Taeuber headed a study of differences in 1990 Census long-form estimates for Maryland of poor children and poor families, and counts of AFDC recipients and cases. The most important factor in the differences was the sampling error. Sampling error is present in all surveys but not in administrative records (this accounted for most of the apparent discrepancy among children except in the city of Baltimore). The other two factors of significance were undercounting people in the 1990 census and differences in the definition and reporting of "income" between the two data sets.