Following several years of testing and evaluation, the American Community Survey (ACS) was launched in 2005 as a replacement for the census “long form,” used to collect detailed social, economic, and housing data from a sample of the U.S. population as part of the decennial census. During the first year of the ACS implementation, the Census Bureau collected data only from households. In 2006 a sample of group quarters (GQs)—such as correctional facilities, nursing homes, and college dorms—was added to more closely mirror the design of the census long-form sample. With some exceptions, residents of group quarters are asked the same questions as household members. These include the questions about personal characteristics formerly included in the census long form—for example, disability, veteran status, and employment. Questions about housing characteristics are not asked, although the type of GQ facility is noted.
The design of the ACS relies on monthly samples that are cumulated to produce multiyear estimates based on 1, 3, and 5 years of data. The data published by the Census Bureau for a geographic area depend on the area’s size. Estimates will be available based on data collected over a 1-year period only for the largest geographic areas (with a population over 65,000). For the smallest geographic areas (with a population under 20,000), the estimates are based on data cumulated over 5 years.
The multiyear averaging approach enables the Census Bureau to produce estimates that are intended to be robust enough to release for small areas, such as the smallest governmental units and census block groups. However, the sparseness of the GQ representation in the monthly samples affects the quality of the estimates in many small areas that have large GQ populations relative to the total population. For example, some counties have group quarters within their administrative boundaries but have no facilities represented in the sample. The Census Bureau, concerned about the adverse effects of group quarters on the 5-year estimates for small geographic areas, asked the National Research Council, through a panel of the Committee on National Statistics, to review and evaluate the statistical methods used for measuring the GQ population.
Below we summarize the panel’s main recommendations addressing improvements in the sample design, sample allocation, weighting, and estimation procedures to assist the Census Bureau’s work in the very near term, while further research is conducted to address the underlying question of the relative importance and costs of the GQ data collection in the context of the overall ACS design. The recommendations in this interim report are limited in their ability to address the fundamental issues related to the effects of the current ACS GQ data collection approach on small-area estimates. The recommendations herein are focused instead on identifying opportunities to evaluate the extent to which the GQ data could be improved, assuming that the goals related to this subset of the sample in the ACS remain the same or are only minimally modified. The panel’s final report will consider the overall design of the current survey in light of data user needs.
Sampling Frame Development and Maintenance
The sampling frame for the ACS is based on the Census Bureau’s Master Address File (MAF), which is an inventory of addresses in the United States, including housing units and group quarters. However, the procedures developed to maintain and update the MAF are focused on housing units and are less adequate for group quarters, a situation that affects the representativeness of the sampling frame and increases data collection costs because of the additional work necessary to clean and improve the sample.
The Census Bureau should explore opportunities for developing and maintaining a better inventory of group quarters, and possibly reducing the amount of work that is invested in updating cases after they have been assigned to a field representative. For example, the ACS office could take advantage of the successful partnerships already in place between other Census Bureau divisions and state demographic offices and other local entities. These relationships could be expanded to better meet ACS sampling frame needs. Collaborations with other federal statistical agencies and organizations that also collect data about residents of various group quarters could be explored as a source of updates to the sampling frame.
Sample Allocation and Selection
The current ACS sample design is suitable for producing estimates of the characteristics of the household population for small-area geographies, but it is not optimized for substate estimates of the GQ population. Even though the Census Bureau does not intend to release detailed characteristics of GQ residents for small geographic areas, GQ data are included in the total population estimates. The Census Bureau should evaluate the current sample allocation rates at the state and substate levels, as well as the subsampling rates in large group quarters, to determine whether there are opportunities for a more efficient design.
Weighting and Estimation
Similar to problems with the sample design, the current group quarters weighting and estimation procedures for group quarters populations are not optimized for small-area estimates. The ACS substate samples are highly variable, particularly by GQ type, and some geographic entities with known GQ facilities have none represented in the sample. If the sample for a smaller area does not include group quarters, the state-level nonresponse adjustment factors and population controls will disproportionately increase the weights of group quarters in other areas. When the 5-year data are released for smallareas, in many cases the estimates will not reflect local reality. The Census Bureau should:
Conduct an evaluation to better understand the quality of the estimates affected by group quarters and the effects of the population controls on these estimates.
Evaluate alternative approaches for producing estimates for areas in which the quality of the direct estimates is in question.
Clearly label data tables and other data products that may be affected by the presence of group quarters in a geographic area.