Page 47
8
Conclusion
The thought pieces, discussant papers, and floor discussions all contributed to a productive interchange of ideas, with creative suggestions, not so much on how to resolve but rather on how to frame and provide perspectives to the many issues raised. Many of the questions raised have an empirical aspect, therefore answers will have to wait until more data are collected. The presenters provided key benefits in addressing the questions raised about alternative approaches and the underlying assumptions, illustrative methodologies used in comparable situations, and the advantages and disadvantages of these different methods.
One area of particular interest involved fund allocation programs, for which the ACS will provide increased opportunities for timely allocation of public funds at low levels of geographic aggregation. Participants argued for the importance of educating legislators about the unintended negative aspects of interactions of funding formulas and the distributional attributes of the estimates used in these formulas.
Participants generally agreed on the importance of the ACS, which under current plans (and assuming that the proposed budgets are realized) will become the main vehicle for collection of long-form-type data on a continuing basis. This potential role underscored the important issue as to who sets the policies for access to the ACS. That is, who will decide on the content of the ACS questionnaire, either as supplements, special subject modules, or as a series of screener questions to target specific groups, and at what cost? Will the ACS be treated similarly to other census surveys, that is, subject to Census Bureau constraints on data sharing, especially addresses of target groups of
Page 48
interest to other agencies, due to concerns about privacy and confidentiality? In effect, will the ACS have all of the characteristics of “census confidential” and the attendant problems of restricted access to microdata by outside agencies?
The question of how often to redesign household surveys on the basis of the updated ACS address list is also a policy matter, though it will have methodological features as well. Quality assurance and timing of the associated master address file updates are also important elements and therefore will require further attention.
In his concluding remarks as chair, John Rolph urged that this workshop be the first step in a continuing process of addressing the many statistical issues raised relative to undertaking the ACS. He said that the Committee on National Statistics would like to be as helpful as possible as the process of designing and fielding the ACS unfolds. Such activities might include committee-sponsored workshops focused on specific issues that need to be addressed at particular points in the design process. He invited the workshop participants to send in their ideas and suggestions for further activities.
Although the workshop was successful in generating valuable initial ideas and discussion on how to address some interesting and difficult methodological problems raised by the ACS, it is important to note that not all important methodological problems were raised. Several issues that were not raised or mentioned only briefly were: (1) additional approaches to combining information from the ACS, household surveys, and administrative records that could also be examined, especially variance component models; (2) methods to treat undercoverage in the ACS, particularly methods for using demographic analysis to address undercoverage in the ACS, and also to use the ACS to improve demographic analysis; (3) further examination of the issues raised through use of incompatible definitions in the ACS, the decennial census, and household surveys; (4) the development of estimates that (a) sum to estimates at higher levels of geographic aggregation and (b) more closely approximate direct estimates at higher levels of aggregation, along with the release of direct estimates at higher levels of aggregation—in the event that aggregate estimates are not constrained to (approximately) equal direct estimates (and also the release of direct estimates at lower levels of aggregation for analysis purposes); (5) the evaluation of the quality of the estimates from the ACS, especially given that no long form is planned after the 2000 census and therefore external evaluation opportunities will be very limited; (6) the need for specific ideas for developing models for borrowing strength from household surveys and administrative records to assist the ACS in the estimation of various outputs; (7) weighting ACS output to produce estimates that are consistent with the recognized estimates from existing household surveys; (8) using information from the ACS to develop models to effectively reduce the sample size requirements of existing household surveys and still produce estimates of
Page 49
the same quality; and (9) coordinating estimates between the ACS and the decennial census short-form in 2010, 2020, and so on for statistics that can be generated from the short-form items alone. These issues were at most touched on at the workshop.
In addition, as noted above, some of the issues that were discussed cannot be fully addressed until the full ACS is collected, and they may require more research. Examples include (1) the development of time-series combination-of-information models; (2) problems raised through the planned addition of questions to the ACS questionnaire, especially the effect on the quality of the information collected for the other questions; and (3) the formation of the model needed to calibrate the long form to the ACS.
This workshop helped to identify a number of interesting and important problems, many of which will have much broader application than the ACS. The workshop succeeded in making the attendees aware of these problems, and raising for discussion promising avenues for their solution.