March 9, 2001

For Immediate Release

Contacts: John Chambers
301-713-6672

DESPITE DECLARATIONS OF INCREASED ACCURACY CENSUS 2000 FILLED WITH ERRORS
Errors Match or Exceed 1990 Levels And May Contribute to Reduced Differential Undercount

Washington, D.C. (March 9, 2001) -- After a preliminary analysis of the Census Bureau's quality control check of the 2000 census, the Accuracy and Coverage Evaluation (A.C.E.), the Presidential Members of the U.S. Census Monitoring Board found that the 2000 count contained 44 million errors and uncertain cases - 9 million more than the 1990 count.

The Census Bureau announced last week that a net of 3.3 million people were missed in the 2000 Census. "How can we accept 36 million errors and 8 million questionable cases as the most accurate census ever but refuse to correct 3 million errors that scientific methods identify with confidence?" asked Everett Ehrlich, Census Monitoring Board Member and former Undersecretary of Commerce.

As defined by the Census Bureau, census error consists of three components -- the numbers of omissions, erroneous inclusions and non-data defined persons. Omissions are people who are missed in the census and erroneous inclusions are people who were counted twice, counted in the wrong place, or shouldn't have been counted at all, such as fictitious people, children born after April 1st and people who died before April 1st. Non-data defined people, sometimes referred to as "whole-person imputations," are computer-generated estimates of the number or characteristics (such as age, sex or race) of people in a household. There were 5.7 million of these in the 2000 Census and 2.2 million in 1990. Uncertain cases include counts where there is a high probability of error, or where there is so little information that we can not tell if it is erroneous.

Re-instated possible duplications refer to the 2.3 million persons for whom the Census Bureau could not definitively determine whether or not they were counted twice.

Errors and Uncertain Cases in the Census, 1990 and 2000   Omissions (errors) Erronious Inclusions (error) Non-Data Defined Persons (uncertain) Re-instated Possible Duplications (uncertain) Net Undercount Total Error and Uncertain Cases 1990 Census 19.9 million 13.0 million 2.2 million ----- 4.0 million 35.1 million 2000 Census 23.7 million 12.5 million 5.7 million 2.3 million 3.2 million 44.2 million

The Census Bureau would not be able to say that Census 2000 is the "most accurate census in history" without reviewing the results of the A.C.E. Moreover, the A.C.E. tells us how many people were missed and how to adjust the census accordingly. In an operation as large as the decennial census, there are bound to be problems. The 2000 Census unfortunately includes a massive number of errors. The following chart depicts the errors and uncertain cases in the 2000 count.

"Should we accept the Census Bureau's relative definition of accuracy when we know that the 2000 count contained as many errors as 1990," asked Gilbert F. Casellas, Presidential Co-Chair of the Monitoring Board. "If you're one of the millions of Americans not included in the final count, the answer is no."

It is possible that the same person could be included in more than one error or uncertain case category. However, despite possible offsetting of errors, the vast numbers clearly indicate a massive census error rate. Furthermore, the effects of errors in the census are differential by race as evidenced in the chart below.

Errors and Uncertain Cases in the 2000 Census, by Race   Omissions Erroneous Inclusions Non-Data Defined Persons Re-instated Possible Duplications Net Undercount Whites, non-Hispanic 6.9% 4.0% 1.5% 0.8% 0.6% Blacks, non-Hispanic 13.0% 6.6% 3.3% 0.9% 2.2% Hispanics 12.6% 5.1% 3.8% 0.9% 2.8%

Dr. Eugene P. Ericksen, a decennial census expert and professor of statistics at Temple University added, "The results from the 2000 Census show us that the differential undercount was reduced, not eliminated, but doesn't tell us how. It is very possible, for example, that the level of omission in the 2000 Census was the same as or greater than the corresponding 1990 level. The reduction in the differential undercount could very well have been achieved by increasing numbers of erroneous inclusions and non-data defined people. Moreover, there are additional questions such as how did the so-called re-instated possible duplications contribute to reducing the differential undercount? It is likely that these re-instated possible duplications contributed to increased error in the raw census count. For example, why is the racial differential undercount less in the rural South than in the rest of the country? Does the fact that many Blacks in the rural South were counted twice offset the missing of others who were harder to count?"

"Now that we have information about census error, we still need to know how many people were missed, how many people were counted twice, or how many people were included by mistake. Until we have the answers to these and other key questions, we can not determine the accuracy of Census 2000," added Casellas. In 1990, 8.4 million people were missed and 4.4 million were counted twice, or incorrectly included for a net undercount of 4 million.