Statistical Adjustment Wouldn't Work
October 2, 2001
An in-depth analysis of the Census Bureau's attempts to make up for the problems of undercounting in the Census by using statistical estimates concludes the process would not correct the undercount of severely undercounted neighborhoods.
In a two-part report, the Congressionally Appointed Members of the U.S. Census Monitoring Board -- created by Congress to oversee all aspects of the 2000 Census --looked at the Census 2000 Accuracy and Coverage Evaluation (A.C.E.), a program to statistically adjust the Census counts to include those who were not counted.
- The report indicates that the outcomes (adjusted estimates of population) of statistical adjustment are based on unreliable and changeable assumptions.
- The way in which the United States population is sub-divided (into post-strata) to measure for the undercount has many limitations and changing only one assumption could dramatically affect statistically adjusted census counts.
- For instance, the difference between a statistically adjusted count for the Southern region produced by the A.C.E. was over half-a-million less than another equally valid estimate produced by the Board using a different post-strata.
- The Board's analysis also strongly suggests that the statistical adjustment process is not sensitive enough to local communities and neighborhoods, and may actually add people to the wrong places at both the regional level and at the neighborhood level.
- While most severely undercounted neighborhoods wouldn't have their counts corrected, many overcounted neighborhoods would actually gain population under the A.C.E.
A. Mark Neuman, co-chair for the Congressional Members, says statistical adjustment may not be the best method to address the differential undercount: "Severely undercounted neighborhoods, especially those in historically undercounted and minority communities, will remain severely undercounted. Statistical adjustment doesn't fix that."
Source: "Changing an Assumption: Measuring the Undercount in Census 2000 with an Alternative Post-Strata Creates Different Results," and "Statistical Adjustment Fails to Correct the Census for Severely Undercounted Neighborhoods: An Analysis of Synthetic Estimation in Blockclusters," Report to Congress, September 29, 2001, Congressional Members of the U.S. Census Monitoring Board.
Browse more articles on Government Issues