Detailed Comparison of Decennial and ACS data
Beginning in 2010, a long form is no longer part of the decennial census. Instead, the American Community Survey has taken it's place. The ACS is administered every year, and provides much more current data for use by government at all levels, businesses, and the public at large. Its variables are similar to those of the decennial long form. Because the ACS is drastically different in methodology from decennial censuses up to 2000, comparing data with earlier years' long form data must be done cautiously.
|(See Notes denoted below)|
2000 Decennial Census Long Form
(see Terms for definition)
|Household population plus group quarters and transient populations||Household population only until 2006 data; no transient populations|
|Smallest geography (2)||Block group||Block group|
|Sample size (3)||About 17%||About 12.5%|
|Collection date (4)||April 1||No single date – all year long|
|Data (5)||Single point in time||>65k – single pt. in time|
<65k – 3- or 5-yr average
|Shows confidence interval (6)||No (but could calculate)||Yes|
(MUST use - look for Margin of Error or MOE)
|Residence rule (7)||Usual||Current|
|Additional Demographic Details (8)||Summary File 4||See Note|
|1% PUMS (9)||Yes||Yes|
|5% PUMS (9)||Yes||Maybe|
|Special Tabulations (EEO, CTPP, etc.) (10)||Yes||Unclear – dependent on non-Census Bureau partners|
|Frequency (11)||Decennial, 1790 to 2000||Annually, since 2003 or 2006 or 2010, depending on geography size|
Be wary of comparing individual variables, even if they have the same name in both surveys, especially with respect to the reference period (i.e., time frame) each uses. For example, in the decennial census the reference period for income variables is the previous complete calendar year, so the 2000 Census reports all income figures from 1999. The ACS, because it collects data every month, instead reports income based on the previous 12 months, whatever those months may be.
- "Group quarters" includes groups such as those living in dormitories, military barracks, nursing homes and prisons. Because these populations are often specialized in terms of age or other characteristics, their absence can substantially skew the figures of the areas where they are concentrated. Prisons, for instance, often raise the population numbers of rural areas.
- Both decennial long-form and ACS data are limited to a slightly larger geography than decennial short-form data because they are calculated on the basis of samples. In both cases, the surveys don't have large enough samples to be accurate at the Census block level.
- The ACS sample size is clearly substantially smaller. Although the Census Bureau has established procedures to ensure that smaller geographies are accurately represented by over-sampling in smaller geographies and among smaller population groups, there are obvious cases where the smaller sample size affects the accuracy of the data.
- The difference in collection date shows up in a variety of ways. In the decennial long-form survey respondents are asked questions about income based on the previous complete calendar year, and about children enrolled in school based on the reference date of April 1st. Decennial long-form questions about one's "usual place of residence" state that the respondent must have lived in that place for most of the year. However, the ACS uses the survey month as the reference period, and request the respondent's "current" residence based only on whether the person has lived there most of the time for the previous two months. (See more on this in #7.) Consequently, even the same respondent's answer could be quite substantially different in the two surveys.
- In the decennial long-form survey all variables are "single point in time." In the ACS, because the sample size is smaller it is necessary to combine the pool of data collected in several years to accumulate a large enough sample to estimate variables for smaller geographies. So for geographies with more than 65,000 people, variables are still reported as single point in time. But for geographies with 20,000 to 65,000 people, variables are reported as an average of three years' worth of data. And for geographies with less than 20,000 people variables must be reported as a five-year average. Since the single-point-in-time variables cannot be compared with three- or five-year averages, geographies with more than 65,000 people will have three- and five-year averages reported in addition to their single-point-in-time figures in order for researchers to be able to compare them with smaller geographies. Likewise, geographies reported in three-year averages will also have five-year averages reported.
Users will need to understand the effect of averaging on the reported figure. An average is a measure of the "middle" of any set of data. So any averaged figure will, by its nature, ameliorate individual outlying responses. See more on the effect of averaged data here.
- Both decennial long-form data and ACS data are estimates. The long-form data simply didn't state its confidence interval alongside the estimates. The sample size of the decennial survey was large enough that it wasn't truly necessary. In the ACS, knowing the confidence interval is *truly* necessary. In some cases the CI is larger than the estimate itself, demonstrating clearly that the data is not reliable.
- As stated in #4, long-form decennial questions about one's "usual place of residence" state that the respondent must have lived in that place for most of the year. However, the ACS uses the survey month as the reference period, and request the respondent's "current" residence based only on whether the person has lived there most of the time for the previous two months. One result of this is that the decennial census could provide no information about temporary populations. Areas that attract seasonal populations, whether beach or ski communities, could not easily quantify the effect of those groups on their local areas. The ACS makes this kind of analysis possible.
- The decennial long-form data is further broken down by ethnic groups, age groups and the like in Summary File 4. These more detailed tabulations are required to have certain thresholds, for example, an ethnic group must have at least 100,000 people nationally to be included in SF4. No such additional detail has yet been planned for ACS data.
- Public Use Microdata Samples have made a great deal of detailed data available to researchers. It has not yet been determined whether 5% PUMS data will be created from ACS data.
- Special tabulations like the EEO and School District data have always been special contracts paid for by the partner agencies. Whether they each will be able to afford to pay for more frequent releases of their various special tabulations remains to be seen.
- The great advantage of ACS data is its far more frequent release. It will give planners in all levels of government, business and the general public far more current data than the decennial long form.
See detailed comparison (.doc file) for more information.