March 8, 2002

Ms. Madeleine Clayton

Departmental Paperwork Clearance Officer

Department of Commerce, Room 6086

14th and Constitution Ave., NW

Washington, DC 20230

RE: The American Community Survey; Proposed Collection; Comment Request

Dear Ms. Clayton:

As requested in the Federal Register of January 16, 2002, we are submitting comments on data collection methods for the American Community Survey (ACS) as proposed by the U.S. Census Bureau. Our agency – the Metropolitan Transportation Commission (MTC) – is the transportation planning and financing agency for the nine-county San Francisco Bay Area region. We also function as a Regional and Affiliate Data Center in the Census Bureau’s data center program, and serve as a clearinghouse for census-related data in the Bay Area and northern California.

The prospect of receiving “long form”-like ACS data on an annual basis is very alluring to the local user community. This data would be beneficial in representing socio-economic trends in our communities, and in updating our databases and analysis systems used in preparing short range and long range plans and predictions. On the other hand, the ACS will have about one-third larger sampling error than the standard decennial census long form for most statistics. The larger sampling error, as estimated by the Census Bureau, and the non-sampling errors (for example, due to non-response and non-response follow-up issues) will detract from the overall usefulness and utility of ACS data.

The Census Bureau has invested a considerable amount of time and effort in preparing for an eventual conversion to the American Community Survey. Data from the 1996-98 ACS tests are fully available, and data from the 1999-2001 “comparison sites” are partially available. We are expecting the “very small area” (census tract-level) data from the ACS 1999-2001 “comparison sites” in Fall 2002.

We are very concerned about the data quality (sampling errors and non-sampling errors) related to ACS data at the “very small area” level. It is premature to judge the accuracy and bias of ACS small area data until this data is released in 2002, and is fully analyzed by Census Bureau and independent statisticians and analysts.

In the long run, the merit of ACS data and data quality may outweigh the expected increase in sampling error, and the unknown element of non-sampling error. Detailed research must continue and must be conclusive before any final decision on the elimination of the Census 2010 “long form” is made.

Is the ACS necessary for the “proper performance of the functions” of the Census Bureau?

It is clear from our local user community perspective that there is an overwhelming demand for decennial census “long form” data. It is also abundantly clear that any elimination of the decennial census long form must be replaced by an equivalent or superior data program. We believe that the American Community Survey – as a replacement for the decennial census long form – has the promise and prospect of being a superior data program that will meet the needs of the local user community. Again, one of our principal concerns is the increased sampling errors and the unknown degree of non-sampling errors that will hamper the usefulness of ACS data, especially at the very small area level.

Will the ACS data have “practical utility”?

The practical utility of ACS data can be assessed in terms of the new data that would be made available, and familiar data that would be re-cast in an unfamiliar light. The “new data” would be the annual ACS data provided for areas of 65,000+ population. The “familiar data” would be the five-year accumulation of ACS data used in preparing data tabulations at the census tract level.

At the Metropolitan Transportation Commission, the ACS annual data should be used in tracking “big picture” trends in household auto ownership, poverty, disability, means of transportation to work, average commute duration and overall commute patterns. This data would also be used by regional planning agencies in year-to-year tracking of population, housing, income, labor force, and workers-by-place-of-work. The data should be used in updating databases and analysis systems used in preparing the long-range transportation plan for our region, and in providing regional socio-economic and transportation indicators of interest to our decision-makers, the media and the general public.

The practical utility of ACS five-year data accumulation will be difficult to gauge before the “comparison sites” data for 1999-2001 are fully analyzed. There will be major challenges for statisticians and educators to re-train local planners and data users in terms of answering such questions as: What does a five-year average of household income really indicate? Or, how do we interpret five-year average data for housing units and population characteristics?

Are the Census Bureau’s estimate of respondent burden accurate?

The Census Bureau’s estimate of respondent completion times (38 minutes per household, 15 minutes per person in group quarters, and 10 minutes per household in the re-interview sample) appear reasonable.

It would be useful for the Census Bureau to test a web-based version of the ACS form, to determine if respondent burden could be lessened.

What are ways to enhance the quality, utility and clarity of ACS data?

It will be necessary for the Census Bureau to develop a comprehensive set of American Community Survey products that will meet the data needs of the local user community. Data products should be defined in terms of annual products, five-year accumulation products, and perhaps intermediate years-of-accumulation (two-year, three-year, four-year) products.

In terms of ACS annual products, we have the following recommendations:

                 1. Summary File #3-like product, not just “profile pages” similar to the upcoming Demographic Profile #2, 3, and 4 products intended for Census 2000.

                 2. Annual Public Use Microdata Sample (PUMS) product, similar to the annual ACS PUMS data released for the 1996-1998 ACS pilot tests. Given the smaller sample size of the annual ACS, the annual PUMS may need to be compiled at the “super-PUMA” level (areas of greater than 400,000 total population). In addition, the inclusion of a month and year of response variable would help the data user begin to understand the seasonality of responses.

                 3. Census Transportation Planning Package (CTPP) annual product, similar to the CTPP 2000 package currently under development. This annual CTPP could include valuable information such as county-to-county and place-to-place commuters by means of transportation to work.

                 4. User-defined community districts in large cities. According to Census 2000 there are 166 cities in the U.S. with a total population greater than 130,000. It would be useful for the Census Bureau to develop a program that allows large cities the opportunity to split their communities into districts of 65,000-or-more total population, and gain the benefit of ACS data at a smaller geographic level, while still meeting the 65,000+ population threshold for annual ACS data. For example, San Jose, California is the eleventh largest city in the United States with a population approaching 900 thousand. It would be useful to split the City of San Jose into about 10 to 13 community districts and to receive annual ACS data at this community district level.

                 5. Alternative geographic areas such as congressional districts, state legislative districts, Public Use Microdata Areas (PUMAS, 5-percent sample) would typically have populations exceeding the 65,000+ population threshold, and would greatly benefit from having annual ACS data summaries. In addition, the Census Bureau should consider a cost-reimbursable data program that allows data users to create their own user-defined area of 65,000-or-more total population. 

In terms of five-year products, we have the following recommendations:

                 1. Summary File #3-like data product, similar to the annual “SF-3” dataset, but available at the finest level of sample geography: census tracts, block groups, and travel analysis zones (TAZ.) Higher geographic summary levels (congressional districts, state legislative districts, PUMAs, sub-city district) would also be needed for the five-year products.

                 2. Public Use Microdata Sample. This PUMS files would include the standard “five-percent” PUMS data at the 100,000+ total population Public Use Microdata Area (PUMA) level. Again, it will be most useful to include the month and year of response as a new variable in the ACS PUMS products.

                 3. Census Transportation Planning Package. The transportation community, including metropolitan planning organizations, local county and city public works departments, the state department of transportation, and local transit operators, will need a CTPP dataset similar to the CTPP 2000 currently under development. The ACS version of CTPP will need to provide the zone-to-zone, tract-to-tract, place-to-place, and county-to-county commute pattern data that local transportation data users depend on.

In addition to data products, the Census Bureau will need to partner with academic programs in statistics, urban and regional planning, geography, sociology, demography, etc., to develop training and education programs to ease the transition of the user community from the decennial census to the continuous measurement (ACS) paradigm. This may take the form of continuing education programs or seminars and workshops co-sponsored by state data centers and regional data centers. The Census Bureau should also work with professional associations such as the American Planning Association and the Transportation Research Board in terms of research and education needs related to the ACS.

What are ways to minimize the burden of the ACS on respondents?

As stated previously, it would be useful for the Census Bureau to test a web-based version of the American Community Survey. In the 2000 decennial census, approximately 64,000 households answered their short forms on the Web. It would be useful for the Census Bureau to calculate the respondent burden and data quality from these web-based short form responses, and to develop pilot tests for the ACS “long form” before full scale implementation of the ACS begins in January 2003.

For households without computers, the Census Bureau may want to pilot test the use of public access internet terminals in libraries, civic centers or shopping malls. The Census Bureau should also recognize the benefits of web-based surveys and computer-assisted personal interviews (CAPI) in possibly reducing the error inherent in optical character recognition (OCR) of census forms.

Another possible mechanism for reducing the burden on ACS respondents is to reduce the length of the questionnaire. This should not be recommended for the 2003-2007 versions of the American Community Survey, but the Census Bureau should be working with local data constituencies and Congress to determine the extent and schedule of any changes to the ACS questionnaire in 2008 or beyond.

In conclusion, we want to thank the Census Bureau for this opportunity to comment on the transition from the decennial census long form to the American Community Survey. We believe there is a strong local demand for continuous data from the ACS, but we are concerned about data quality, including the one-third higher level of standard errors in the ACS, and an unknown degree of non-sampling errors. Research needs to be conclusive in terms of the extent of standard errors at the very small geographic areas; non-sampling errors such as partial and complete non-responses; and corrective actions needed to improve the quality of ACS data. Until these issues are resolved, it may be premature to cancel the Census 2010 long form.

Sincerely,

Charles L. Purvis, AICP

Senior Transportation Planner/Analyst

Metropolitan Transportation Commission

101 Eighth Street

Oakland, California 94607