RE: [CTPP] Interesting Results from the ACS

Long post from Elaine Murakami, Federal Highway Administration ACS 2000-2002 Change profiles WARNING: It is not appropriate to compare decennial Census 2000 numbers directly with the ACS 2002 results. We agree (with Phil Salopek) that it is not wise to compare the "current" ACS results directly with decennial 2000, since it appears that methodological differences are making a significant difference in variables like travel time. It is BETTER to compare ACS 2000 figures with ACS 2002 to determine if there has been significant change. The Census Bureau issued a press release on Feb 25, 2004 related to data from the American Community Survey. Right now, the ACS is not in "full implementation" (expected to begin in July 2004). However, an annual sample (1203 counties of 3141 counties in the U.S. are included) is surveyed over 12 months and the results are available for large geographic units (states and some counties). http://www.census.gov/acs/www/Products/Tables/Tables1.htm Under full implementation, all counties will be surveyed. To access the 2000-2002 Change Profiles available for States and large Metropolitan Areas, go to http://www.census.gov/acs/www/Products/Profiles/Chg/2002/0002/index.htm Once you have selected a geography, select the 3rd page of the profile to find tables on means of transportation to work and average commute time. We have extracted State level geography for average travel time, number of workers and percent of workers using transit for their usual commute and it is attached as a spreadsheet. Recently, on the CTPP and the State Data Center listservs there have been discussions about the differences between the two surveys, particularly when people use decennial 2000 data, and then ACS 2002 data, without understanding that the ACS 2000 number may differ significantly. For example: % Carpool to work, US Total 2000 decennial 12.2% 2000 ACS 11.2% 2002 ACS 10.4% It would be misleading to say that carpooling declines by 1.8% between 2000 and 2002. It is probably better to say that carpooling declined by .5% (after assumptions about errors are incorporated). Similar differences occur when comparing average commute times. Average Travel Time to Work, US Total 2000 decennial 25.5 minutes 2000 ACS 24.4 minutes 2002 ACS 24.4 minutes It would be misleading to say that average travel times have declined by 1 minute, when it is more accurate to say that the average travel time remained the same. So far, we believe that the key factors that contribute to the difference between decennial and ACS results include: a. Seasonality (12 months instead of "April 1") (different jobs and workers, especially in summer) b. ACS does not currently include group quarters population. c. ACS has only one-third of counties of the country represented. d. Differences in Non-response follow-up e. Data collection period (ACS is very long, compared to very short in decennial) FHWA research conducted by Westat, using 1999-2001 ACS data, showed that average travel times were significantly different between the ACS and the decennial. Generally speaking, average travel times in Census 2000 are from .8 min to 1.8 minutes LONGER than the ACS reported times. Pima County ACS 25.2 Decennial 26.7 * San Francisco 33.3 34.6 * Broward, FL 28.7 30.5 * Lake, IL 34.6 34.0 * Flathead, MT 20.6 21.4 Bronx, NY 48.0 49.2 * * = significantly diff When examined spatially (by tract) and by mode, there was no pattern found in our limited research to explain the difference. We did not compare by month or season, but some of us suspect some seasonality differences. Another factor may be that the decennial is more likely to capture longer distance commutes. We have not yet been able to compare this, but I hope the NCHRP project on the ACS for transportation planning will explore this issue. Other researchers believe that the use of better-trained, experienced field interviewers is resulting in overall higher quality data in the ACS. Reporting at small geographic units Because the ACS surveys a small sample continuously, for reporting at small geographic units (smaller than 250,000 population), samples over multiple years will averaged. (Note from Elaine—we really don’t know what this means, but are working with the CB to get a better handle on it). For counties or places (e.g. incorporated cities) with population between 65,000 and 250,000, a three-year average can be reported. For areas with less than 65,000, it will require a five-year average before reporting. Last words (at least for today) There are many questions about ACS yet to be answered. The Census Bureau plans to release a series of reports comparing the 1999-2001 ACS test to the decennial census 2000 this Spring. Fiscal year 2005 will be first year when the Census Bureau requests a budget to cover a full 12-months of data collection costs for the ACS. The ACS will require 5 years of data collection before small area reporting can occur. The Census Bureau has promised the transportation community that "small area reporting" includes Block Group and TAZ reporting. However, any cuts to sample size would seriously impact this ability for small area tabultion.