By

Nancy Kirkendall

First, I’d like to thank the authors for their interesting papers. I enjoyed reading them. The American Community Survey (ACS) is an innovative alternative to the long form of the decennial census with great potential for enabling improvements to other surveys and programs within the Federal Statistical System. The papers in this session address the possibility from the point of view of the Current Population Survey (CPS), the National Crime Victimization Survey (NCVS) and from a broader perspective – use by local transportation planners.

The ACS was designed to replace the long form in 2010. For this purpose it will:

• streamline census processing, as there will only be the short form

• eliminate differential response rates on short and long form.
• help to reduce the big decennial bulge in both the Census Bureau's burden budget and dollar budget by collecting long form data in a rolling fashion over the decade.
• result in an improved Master Address File (MAF) for use in the 2010 census.
• result in more highly skilled interviewing staff available throughout the country for surveys and censuses conducted by the Census Bureau.
• provide current demographic data through the decade for purposes of survey weighting, distributing federal funds to local areas, and supporting local planning activities.

This session focuses on potential benefits of the ACS to other surveys and programs. Some of the benefits I noted above are also benefits to other Federal surveys -- and were pointed out in this set of papers. In particular, other surveys will benefit from more highly skilled interviewing staff; and better data for weighting and estimation.

Ultimately the MAF should also be of great benefit to other surveys, although these papers are not making that claim. Though the surveys will ultimately benefit, the benefit may not be achieved until the next major sample redesign activity following the 2010 census. Still, the availability through the decade of an up-to-date Master Address File for households has great potential.

The most exciting innovation related to the availability of the ACS was pointed out by both Shail Butani (BLS) and Denise Lewis (BJS) -- the use of ACS data in conjunction with data from other surveys to prepare small area estimates. These two applications are very good examples because they are in very different situations. The ACS currently has a set of unemployment questions that might be useful in conjunction with CPS data for small area estimation. So BLS can consider doing preliminary analysis soon. On the other hand, at present there are no crime-related questions on the ACS, so the first challenge for BJS is to convince the Census Bureau to add a question (or questions) about crime.

The other potential benefit to ongoing surveys, is the possibility of using the ACS to provide screening information, or information for oversampling. Again, for some surveys, no change is required to serve this purpose. For example, if a survey needs to target households with kids, or with kids of a certain age group, the data are available on the ACS. For other applications, for example the crime surveys, an additional question or questions would need to be added to the ACS.

From the BJS perspective, both of these possible applications require adding questions to the ACS. This, of course, leads to the question of how likely it is that changes can be made?

In past decades, in preparation for the decennial census, an interagency working group, chaired by OMB, advised the Census Bureau concerning the content of the long form. The Census Bureau is currently required to deliver the topics to be included in the long form to Congress on April 1 of xx07, and to deliver the questions themselves to Congress on April 1 of xx08.

With the ACS, of course, it is too early to know whether the content of the ACS will be required to be delivered to Congress with this same schedule. It is also too early to know whether ACS will receive the same level of Congressional attention. Recall that it was Congressional attention that convinced the Census Bureau to minimize the length of the long form questionnaire in 2000.

For the 2000 Census the content of the long form was limited to only those topics and questions required by law. Most likely with the ACS, this Congressional pressure to keep the long form short will be eased – but, personally, I think that it will still be important to keep the ACS as short as possible – while maximizing the utility of the information.

As agencies consider developing the questions they would like to include on the ACS, they should also be reviewing their mandate to make small area estimates -- to assemble and develop solid arguments for needing the data. These arguments could be key to getting new questions added to the ACS.

Two interesting themes of this conference are small are estimation and cognitive methods. I would like to link these two ideas here. Agencies that are interested in using ACS data in conjunction with their own survey data for small area estimation should develop a small number of very good questions to be asked on the ACS. As part of this activity, they should also make sure that data from these questions are highly correlated with a broader set of very good questions on their on-going surveys. Making sure the questions are ‘very good’ requires the application of cognitive methods and testing. This work will stand agencies in good stead when the debate about the content of the ACS begins in 2002, or so.

In John Bailar’s talk yesterday, he stressed the importance of asking the right questions in ways they can be answered. Now is the right time for agencies to improve their basic questionnaires, and to develop a carefully crafted set of questions that might be included in the ACS and used in conjunction with the basic data to develop small area estimates.

Shail Butani also pointed out some of the BLS concerns about the quality and comparability of data from the ACS. Concerns she noted include differences due to mode effects (interviewer versus self-administered), differences in handling non-response, and the use of a rolling time period. There are good questions, and they should be examined with the intention of continuing to improve both the ACS and the CPS. I will cite one example Shail used to illustrate the benefits of adaptive interviewing – possible only in an interviewer administered mode. When asked the question "Did you work for pay or profit last week?" respondents typically said something like "just my regular job". In this case, the interviewer can help the person decide how to respond. It is not clear how this question would be answered in a self-administered questionnaire.

Daniel Hess presented a very interesting comparison of local data with ACS data via special transformations. In the front of the paper they claim that they are "focusing on the impact of the ACS on data for transportation planning and policy analysis. An analytical approach is taken to document the validity of estimates from the ACS." I do not agree that they are documenting the validity of the estimates from the ACS, however the paper does provide a useful and informative analysis.

When differences are found between the ACS and the local data it is not clear which one is more likely to be in error. In this case, it is the local data for which there are no metadata. Still the comparisons are interesting and as more comparisons are made, the comparative approach may identify real problems that can be addressed and solved. If no problems are found, comparisons can be used to convince users of the value of the ACS. I would have been particularly interested in a similar comparison for the Portland area of the local data and the long form data in 1990. That would have helped me to better understand the nature of the differences highlighted in this paper.

One of the discrepancies found in the current comparison is within the highest and lowest income groups. The fact that the ACS uses a rolling time period for collecting income rather than the time period associated with the tax year may be one reason for these differences. I am sure that the Census Bureau is continuing to evaluate the way the ACS collects income data.

The BTS paper provides a list of variables typically needed for travel demand forecasting. It would be useful to apply cognitive studies and testing to fine tune a set of questions that could be used to solicit this information. No only could this help improve ACS questions, it could provide a valuable resource for local agencies that develop and implement local planning surveys.

One hopes that the timely and consistent ACS data will prove to be useful to local transportation planning groups -- and hence to minimize the need for, or at least the size of, special local surveys. This would be yet another benefit of the ACS.