The American Community Survey is designed to replace the census long form as the primary source of information on social and economic trends in the U.S. population. Its primary advantage over the long form is that it will produce information every year rather than just every ten years.

When to Use Data from the ACS

Use the 2000 Census instead of the ACS:
- when you need information on the entire population of an area, including group quarters such as dormitories, prisons, shelters, and nursing homes;
    -OR-
- when the indicated confidence intervals for ACS data are unacceptably wide. (That is most likely to be the case for small segments of the population in cities, counties, and other sub-state areas.)

Use the Census Bureau's population estimates instead of the ACS:
- when you need up-to-date information about the total population of counties, cities, townships, or villages;
    -OR-
- when you need up-to-date information about age, race, sex, and Hispanic origin at the state or county level.

(The ACS is controlled to the Census Bureau's estimates of household population by age, race, sex, and Hispanic origin. However, the methodologies for imputing, controlling, and weighting responses can cause ACS figures to deviate from the population estimates in ways that should not be considered meaningful.)

Use the ACS if:
- you need up-to-date information on the PERCENTAGE of people or housing units with various characteristics such as educational attainment, place of birth, occupation, etc.;
    -AND-
- your question involves only large geographic areas or large segments of the population;
    -AND-
- you can use information that excludes the portion of the population living in group quarters such as dormitories, prisons, shelters, and nursing homes.

Use the ACS with caution:
- when you need information about small geographic areas or small segments of the population;
    -OR-
- when you need information about the NUMBER of people or housing units with various characteristics such as educational attainment, place of birth, occupation, etc.

(In addition to being affected by the limitations inherent in a sample survey, which are reflected in the confidence intervals that are published with ACS data, such figures are affected by limitations of the population estimates to which they are controlled and limitations of methodologies for imputing, controlling, and weighting responses.)

Residents of group quarters are being included in the ACS beginning with data collected in 2006 and released in 2007.

Specific Limitations of the ACS Prior to 2005

The ACS has been conducted on a small scale since 2000. However, its sample was limited in geographic scope as well as overall size until 2005. Only 36 of Michigan’s 83 counties were included: 24 out of 26 metropolitan counties, 7 out of 23 micropolitan counties, and 5 out of 34 undesignated counties. In order to produce state totals, figures for the remaining counties were estimated based on similar counties in Michigan and other states. (Click here for a list of Michigan counties included in the ACS prior to 2005 .)

The ACS was also limited to people living in “households.” People living in “group quarters” such as prisons, group homes, nursing homes, and dormitories were not included in the ACS until 2006. The exclusion of students living in dormitories has a particularly large effect upon data involving young adults, school enrollment, educational attainment, and income. The exclusion of people living in nursing homes has a large effect upon data involving the elderly population and people with disabilities. Poverty figures are not significantly affected, however, since poverty status is not determined for people in institutions or dormitories even when income data are available.

Specific Limitations of the ACS for 2005

In 2005, the ACS was expanded to its full sample size for households in every county. However, the 2005 survey still excluded people in group quarters. This affects most data items, as discussed above, with particularly serious effects upon data regarding young adults, school enrollment, educational attainment, income, the elderly population, and disability status.

Click here for a comparability chart indicating which ACS data items are comparable to data items from decennial censuses.

Statistical Reliability

The statistical reliability of survey data is affected by the size of the sample and other factors. Data from the ACS are normally published with a “90 percent confidence interval” or a “margin of error.” These indicate the likely range within which the true value of the characteristic being measured is likely to fall. If the survey were to be conducted again for the entire population rather than just a sample, the new “true” values would be expected to fall somewhere within the designated confidence intervals or margins of error 90 percent of the time.

It is important to consider these measures of statistical reliability when comparing data from different geographic areas or different points in time. Generally speaking, the difference between the figures being compared is not “statistically significant” if their confidence intervals or margins of error overlap.  If the confidence intervals or margins of error do not overlap, then the difference between the figures is “statistically significant at the 90 percent level”—in other words, there is at least a 90 percent chance that the higher figure would still exceed the lower figure if the survey were to be repeated for the entire population.

It should also be noted that differences observed between 2005 and prior years can reflect the geographic expansion of the sample as well as sampling error and actual changes that have occurred over time.

The Census Bureau has produced a document containing useful guidelines for calculating confidence intervals for ACS data .

Using Multiple-Year Data from the ACS

Although the level of sampling error was greatly reduced when the size of the ACS sample was increased in 2005, it is still advisable to combine ACS data from several years whenever feasible.

It is valid to compare multi-year figures that are statistically independent of one another, such as figures for different geographic areas or figures for ranges of years that do not overlap.

HOWEVER, IT IS GENERALLY NOT VALID TO COMPARE MULTIPLE-YEAR DATA FOR OVERLAPPING TIME PERIODS, SUCH AS "MOVING AVERAGES." Such comparisons are equivalent to comparing the year dropping out of the moving average to the year coming into the moving average, and they should be based on the standard errors for those individual years rather than the standard errors for the multi-year averages.

The Census Bureau has prepared a document containing useful  guidelines for calculating standard errors for multi-year data.

Further Information on the Nature and Limitations of ACS Data

• The Library of Michigan has prepared a Media Advisory to highlight specific issues in interpreting data from the ACS.
• The Census Bureau has prepared a document on definitions and comparability of individual data items.
• Further information about the ACS and its limitations is available from the ACS page of the Census Bureau’s website.

Updated 8/16/2006

Library of Michigan, Department of History, Arts and Libraries