entire county would be sampled at 5-year cumulative rates of 1 in 2 or 1 in 3 households instead of at an average rate of 1 in 9 or less. The effect would be to reduce the sampling error for estimates of the entire county to the point that they could meet common standards of acceptable precision.12

Even with oversampling, however, the 5-year period estimates for very small governmental units will fall far short of common standards of precision for many population groups of interest. For example, the 90 percent confidence interval for an estimate of 15 percent poor school-age children for an oversampled area of 1,500 people, based on the assumptions in Table 2-7c, would likely range from 7 to 23 percent, which is not very informative about the extent of school-age poverty. By contrast, the 90 percent confidence interval for an estimate of 15 percent poor school-age children for an area of 50,000 people would likely range from 12 to 18 percent, which is a considerable improvement.

It is important to remember that the 2000 long-form-sample estimates were also subject to considerable sampling error for small areas. However, they were somewhat more precise than the corresponding estimates from the ACS cumulated over 5 years.

The precision of the 5-year period ACS estimates can be improved by aggregating small areas into larger units. Indeed, this is the recommended strategy for large jurisdictions—namely, to aggregate census tracts and block groups into larger subcity or subcounty areas for such purposes as planning the location of governmental service sites and services. A strategy of aggregation is not as suitable for small governmental jurisdictions, given that each typically provides its own services and is interested in estimates for its jurisdiction alone.

Small jurisdictions could ask the Census Bureau to provide estimates for, say, 8-, or 10-year periods that are more precise than the 5-year period estimates. The drawback of this approach is that lengthening the period of the estimates averages underlying patterns of variation in social and economic phenomena over longer periods and does not produce large gains in precision. For the case of a town of 1,500 people, the 90 percent confidence interval for an estimate of 15 percent poor school-age children would be reduced from 7 to 23 percent for the 5-year period estimate to 8.7 to 21.3 percent for an 8-year period estimate and to 9.3–20.7 percent for a 10-year period estimate (under the assumptions used in Table 2-7c). By comparison, the 90 percent confidence interval for the same estimate from the 2000 long-form sample would be 9.7 to 20.3 percent.

To produce reasonably precise estimates for small population groups in

12

The beneficial effects on sampling error for county estimates that result from oversampling subcounty areas are not as great when the subcounty areas are sampled at varying rates, such as 1 in 2, 1 in 3, 1 in 6, and 1 in 9.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement