| Index | Last Week | Contact DQ News | Latest Issue |
| Featured Articles | Data & Information | Science & Technology | Education |
| Economics | Business | Law | Medicine |
Census Bureau Details Census 2000 Sampling Concepts
An article in the May-June issue of American Scientist provides details of the U.S. Bureau of the Census's sampling plans for the 2000 Census. Written by Tommy Wright, the Bureau's chief of its Statistical Research Branch, the report, "Sampling and Census 2000, The Concepts," explains how the Census Bureau reached the conclusion that "sampling" is the best way to reduce the expected census undercount in 2000.
To estimate human populations, demographers combine records, basically using a simple equation: population = (births - deaths) + (immigrants - emigrants). There is evidence that the most recent U.S. decennial censuses have resulted in a net undercounting of the population. Extensive research conducted during the 1990 decennial census period suggested some characteristics of the people who were missed, including complex housing arrangements, informal housing arrangements, mobile populations, fear of government and of census takers, and limited knowledge of English. [Editor's Note: This knowledge could have been employed in 1990 to increase the response rate by using deterministic methods (e.g., targeted advertising campaigns designed by top notch advertising firms.)]
In each of the most recent censuses, questionnaires were either mailed or hand-delivered to every known residential address with a request to complete and return the questionnaires. In 1990 about 500,000 enumerators were hired to go door-to-door to collect information from the 25.9 of households who did not return a questionnaire. According to the Census Bureau, the conventional counting attempt in 2000 will resemble the 1990 Census.
The Census 2000 plan will integrate counts from conventional techniques with a nationwide probability sample of 25,000 blocks which is totally independent of the data from the first sample. Census enumerators will attempt to visit each and attempt to list every person who was a resident on Census Day. For each of the 25,000 sample blocks, the Census Bureau will compare the first measure with the independent second measure, looking for matches of housing units and persons. The two measures will be statistically matched to yield a single set of data for all areas. The concept is to multiple the first measure (mostly based on counting) by the second measure (based on sampling) and divide this product by the numer of matches, leading to an improved count, which the Census Bureau calls the one-number census. [The article's author assumes the Census Bureau will be able to accurately sample 25,000 census blocks]. The article appears on page 245.