# An Example of Sample Size Determination: Non-Response Follow Up in the 2000 Census The U.S.

Census Bureau planned to introduce sampling into the process of conducting the 2000 Census. The plan was abandoned after the Supreme Court ruled against the use of estimation for reapportionment purposes. The plan, however, was the following. The nation is divide up into geographic areas of approximately equal numbers of housing units, called Census blocks. After questionnaire mailbacks were complete, a SRSWOR of size nNR from the NNR non-responding housing units of each geographic area would be chosen. Interviewers would go tot the sample of nonresponding housing units and collect their census information. A census “count” for that geographic region would then be estimated as ˆ C = C r + N NR y NR , s , where Cr is the actual count from the responding housing units, and y NR , s is the average household size for the sample of nonresponding housing units. Suppose that each geographic area had about 40,000 housing units. How large a sample should be chosen from each area so that the margin of error is about 0.3% of the population, if the initial response rate is 65%? If the initial response rate is 85%? Suppose the geographic area was made of 10,000 housing units. Do you think a smaller or large sampling fraction would be required to achieve the same percentage-wise margin of error?