In recent decades, the census has had two components. An intensive effort was made to collect “short form” information about all residents (information that was often provided by one person for all household members). The short form questionnaire has been the source for “100% count” items required by law, including address, age, race and Hispanic origin, relationship to other household members, and housing tenure (owner/renter). These data have been used most quickly and prominently in reapportionment and redistricting, and they have been made publicly available in aggregate form at the level of census blocks (though methods of imputation and substitution are used to mask people’s identities, especially at that level).
In the past, “long form” questionnaires were sent to a one-in-six sample of persons, with considerable effort to achieve a representative sample. This questionnaire contained all additional detailed census questions such as income, education, labor force participation, source of income, citizenship and immigration status, ancestry, county of birth, housing rent, home value, and other items. Tabulations of these “sample count” data have routinely been made available at the level of block groups and larger geographic and political units within a year of the release of full count data. Public Use Microdata (PUMS) files for a 5% sample of persons have been released about another year later (with the limitation that geographic location was disclosed only at the level of the PUMA (Public Use Microdata Area), containing 80,000 to 100,000 residents).
The long-form decennial questionnaire has been replaced by an annual national survey called the American Community Survey (ACS). In 1996, ACS began with modest samples in pilot sites. It has now expanded to national scope. Since 2005 the ACS interviews samples of housing units in all counties in the United States (including the District of Columbia) and in all of the municipalities in Puerto Rico. Each year approximately 3 million housing-unit addresses in the United States and Puerto Rico are selected, and interviews are completed for about 2 million addresses.
In 2006 the ACS began interviewing residents in group quarters (GQ) facilities. Samples for GQ facilities and persons are selected independently within each state, including the District of Columbia and Puerto Rico. Each year the ACS includes approximately 2.5% of the expected number of residents in group quarters.
Scholars will be interested in comparing analyses of 2000 PUMS data with ACS microdata from later years. Because of differences between these datasets, such as the timing of interviews or the time frame for which some questions are asked, such comparisons will need to be made with caution.
Since 2005, the number of sampled cases in each year has been sufficient to create estimates of population characteristics at the level of major cities and counties with more than 65,000 residents. Because the sample is limited and the sampling frame is based on the 2000 census and subsequent population estimation techniques, there have been anomalies in reported results – for example, population counts for a city may differ substantially between ACS and the annual population estimates disseminated by the Census Bureau. The Census Bureau has introduced a routine procedure of accompanying all estimates with the lower and upper bound of the estimated confidence interval. The width of these confidence intervals has surprised users and introduced ambiguity about what counts can be reliably studied or compared over time.
In principle, the samples drawn every year are independent of one another, and can be examined separately at the national level (e.g., in the case of public use microdata samples) or be pooled across years to yield better estimates and tighter confidence intervals for particular estimates. For this reason, the Census Bureau projects that it will be possible to use the ACS to report population information for ever smaller geographic units.