Wisconsin v. City of New York

The manner in which the census is taken has changed relatively little, despite advances in the sciences of statistics and demographics. Various technologies have been employed to speed the gathering and analysis of census data, but the system still relies on voluntary information and, in some cases, door-to-door canvassing by state authorities. Beginning with the 1970 census, demographic and statistical experts and federal census officials agreed that the traditional method of taking the census resulted in a population count that was too low. Many people either mistrusted census authorities or could see no reason to participate in the census, resulting in low counts for such populations.

Many suggestions were made to address this chronic undercount and make the census more accurate. One of the foremost among these was known as dual system estimation (DSE), a method of census taking which combined sampling, estimation, and traditional census taking techniques to arrive at a more accurate result than traditional methods alone. To illustrate the methods a DSE census would employ, the Court made use of the following analogy:

Imagine that one wanted to use DSE in order to determine the number of pumpkins in a large pumpkin patch. First, one would choose a particular section of the patch as the representative subset to which the `recapture' phase will be applied. Let us assume here that it is a section exactly one-tenth the size of the entire patch that is selected. Then, at the next step--the `capture' stage--one would conduct a fairly quick count of the entire patch, making sure to record both the number of pumpkins counted in the entire patch and the number of pumpkins counted in the entire section. Let us imagine that this stage results in a count of 10,000 pumpkins for the entire patch and 1,000 pumpkins for the selected section. Next, at the `recapture' stage, one would perform an exacting count of the number of pumpkins in the selected section. Let us assume that we now count 1,100 pumpkins in this section. By comparing results of the `capture' phase and the results of the `recapture' phase for the selected section, it is possible to estimate that approximately 100 pumpkins actually in the patch were missed for every 1,000 counted at the `capture' phase. Extrapolating this data to the count for the entire patch, one would conclude that the actual number of pumpkins in the patch is 11,000.

It was widely believed that the use of a DSE count would greatly improve the accuracy of the census.