Loading

The Leadership Conference on Civil and Human Rights

The Nation's Premier Civil and Human Rights Coalition
The Leadership Conference on Civil and Human Rights & The Leadership Conference Education Fund
The Nation's Premier Civil and Human Rights Coalition

Census 2000 Education Kit

How Do We Know There Is An Undercount?

The Undercount

In 1940, as the country was preparing to go to war, the War Department asked the Census Bureau to provide them with estimates of the numbers of young men they could expect to sign up for the draft. The Bureau gave them their best numbers; however, when the draft came, more men registered than predicted by the Bureau based on the most recent decennial census. At that time, the military was still segregated, and it became apparent that the numbers from the census for young Black men were far less accurate than those for young White men. While the Bureau had known for some time that people were missed in the census this was the first evidence they had that some populations (young Black men in this instance) were more likely to be missed than other populations. With this evidence, the Bureau began a serious study to develop better measures of the undercount.

Measuring the Undercount

Using information from birth and death records, past censuses, and information on immigration, the Bureau developed a system of measuring the undercount called Demographic Analysis. This method has been used to gage the accuracy in the census since 1940; however, it can only give national undercount numbers. Most people want to know what the undercount is for their state, city, or community by age group, race, and gender.

To satisfy the desire for these more specific undercount numbers, in 1970, the Bureau began experimenting with the use of sample surveys to measure the undercount. The 1970 experiment was too small to provide many numbers, but did show promise. A more elaborate experiment was conducted following the 1980 census, but again, the size of the sample survey was not sufficient for providing undercount estimates for small areas. A number of cities sued the Commerce Department trying to force the use of the 1980 estimates, but the courts consistently ruled that if the Census Bureau did not find the adjusted numbers sufficiently accurate, it was not required to do so.

Nevertheless, the Bureau was committed to providing estimates sufficient to correct the census counts, and in 1987 proposed a survey of 300,000 households which could be used to adjust the 1990 census to correct undercounts and overcounts. The survey size was large enough to ensure accuracy at all geographic levels. That plan was canceled by political officials in the Department of Commerce. In response to a lawsuit filed by the City of New York and dozens of other states and localities, the Department of Commerce let that survey go forward, but with only 150,000 households. That survey was known as the Post Enumeration Survey (PES), and provided the basis for the adjusted numbers that were produced for the 1990 census. These adjusted numbers represent our most elaborate understanding of the undercount.

Checking Accuracy in the 2000 Census

The Bureau's plan for 2000 includes a similar procedure to check for missing households and people who were not reported on a census form. Like the 1990 Post Enumeration Survey, the Bureau will conduct a very large, "quality check" sample survey.

The Bureau will send its best enumerators out to interview (in person or by phone) a large sample (perhaps as large as 750,000) of households in selected census blocks in every type of community, spread across every State in the country and the District o f Columbia to make a final check on undercounts and overcounts. They will collect information on who was included in the initial count and who was not in those households. That information will allow the Census Bureau to measure the extent of the undercount (people who were missed), and any overcounts (people counted more than once), for all types of households in all types of communities, specific to each State.

This type of statistical method is well known to quality assurance managers in the private sector. The survey will provide the basis for a correction of undercounts and overcounts before the final numbers are tallied and published.

The Bureau will then use the results of that "quality check" survey to project rates of undercount and overcount for every census block in the country, by comparing household characteristics collected in the initial count in that block with undercount/overcount rates for households with the same characteristics in the second survey.

The projected rates of undercount (and overcount) will be very precise, by race, gender, age, owner/renter status (a proxy for income level), and type of geographic area (i.e. rural, small city, large city, etc.) within each State. Scientific research shows that the rates of undercounting (and overcounting) are similar for people with similar demographic characteristics who live in similar types of geographic areas.

The Bureau will then correct the initial census counts for each block to eliminate the undercounting and overcounting it has projected for that block. (In the case of overcounting, the Bureau will not subtract actual people from the counts who may have re turned a form by mail or responded in person. The correction involves an adjustment of the initial counts for that block, to reflect the fact that too many people were included the first time.) The result is a final census number, for all levels of geography (blocks, tracts, cities, counties, states, etc.) that combines several counting methods (direct contact, estimation, and a quality check survey) to produce numbers that most accurately reflect the demographic composition and geographic location of the population.

To ensure the accuracy of the 2000 census, it is critical that the Census Bureau be permitted to conduct a quality check on the 2000 census and that the check be of sufficient size to ensure accuracy at every geographic level.

Our Coalition

Our Members