The following HTML text is provided to enhance online
readability. Many aspects of typography translate only awkwardly to HTML.
Please use the page image
as the authoritative form to ensure accuracy.
Using the American Community Survey: Benefits and Challenges
3-D.1 Using the ACS 1-Year PUMS Files
Transportation planners are concerned that the ACS yearly PUMS product will contain only about 3 million person records. This reduction from 14 million persons means that the sampling error of estimates from the ACS 1-year PUMS will be much larger than those of estimates from the 2000 long-form-sample 5 percent PUMS, and estimates from the long-form-sample 5 percent PUMS are already subject to about 1.8 times more sampling error than estimates from the full long-form sample.
As an example, for a PUMA with 50,000 workers, an estimate from the 2000 long-form sample 5 percent PUMS that 15 percent of workers carpooled to get to work would have a 90 percent margin of error of approximately plus or minus 1.6 percent—1.83 times the margin of error of about ±0.9 percent for the full long-form sample (see the fourth row in Table 2-7b). This margin of error equates to a coefficient of variation of 6.5 percent. However, a corresponding estimate from the ACS 1-year PUMS would have a 90 percent margin of error of at least plus or minus 3.6 percent based simply on the difference in the number of records. This margin of error equates to a coefficient of variation of 14.5 percent, which does not meet accepted standards for precision. Moreover, the weights in the ACS PUMS will be more variable than those in the 2000 long-form-sample PUMS due to the subsampling for CAPI follow-up in the ACS. Consequently, estimates from the ACS PUMS will likely be even less precise compared with estimates from the 2000 long-form-sample PUMS than indicated above.14
A possible solution for the smaller size of the ACS PUMS is to combine two or more PUMS. While transportation modelers will not likely want to fully analyze each new PUMS release because of the time and resources that would require, the availability of an annual PUMS will make it possible to periodically check and recalibrate their models. Similarly, the availability of updated ACS 5-year period estimates will make it possible to reestimate control totals for the models at the county and TAZ levels more often than once a decade.
The current scheme for selecting the ACS PUMS files draws an equal-probability systematic sample of all ACS housing unit records and their household members in each state, with the records sorted by several characteristics (see the 2005 PUMS accuracy statement at http://factfinder.census.gov/home/en/acs_pums_2005.html). A different selection scheme would retain a higher proportion of the CAPI cases so as to equalize the weights of CAPI and non-CAPI cases, yielding a PUMS that would produce more precise estimates than the current PUMS. This scheme could be extended toward equalizing the weights of all sampled housing unit records within PUMAs.