Aug. 28, 2006

The first American Community Survey (ACS) tables for 2005 were released by the U.S. Census Bureau on Aug. 15 (demographic and social characteristics), with additional releases scheduled for coming weeks:

Aug. 29: Economic statistics, including data related to income, poverty and employment

Oct. 3:  Housing data

Nov. 14:  Additional cross-tabulations by race, ethnicity, and ancestry

Several aspects of the new data require special caution on the part of news media and other data users, according to Kenneth Darga, Michigan's state demographer.

(1) Omission of people living in group quarters

The new figures do not include people living in "group quarters" such as college dormitories, prisons, barracks, shelters and nursing homes.  Although every table downloaded from American Factfinder includes a note to this effect, several data items within the tables have been labeled by the Census Bureau as "total population" when, in fact, they pertain to just the "household population" due to the omission of people in group quarters.  Several news reports have erroneously treated ACS data as representative of the total population.  Although only 2.5 percent of Michigan's population lives in group quarters, the percentage is as high as 15 percent in individual counties and 60 percent in individual communities.

Every reference to "total population" in ACS tables for 2005 and prior years should be interpreted as "household population."

Until group quarters are incorporated in the survey, it will not be valid to compare most data from the ACS to similar data from the census. 

A list of data items that are significantly affected by the omission of group quarters has been posted at www.michigan.gov/census, along with a (much shorter) list of data items that are not seriously affected.  Affected data items can be compared to ACS data from other years, but they should not be compared to decennial census data.

(2) Spurious deviations from population estimates

The Census Bureau's annual population estimates should be used instead of ACS data by all data users who need information about an area's total population, its household population or the distribution of its population by age, race, sex or Hispanic origin.

In particular, ACS data should not be used as an indicator of total population for Detroit or other individual communities in Wayne County.

A sample survey cannot tell us the number of people in a given category until its results have been weighted to some measure of the total population.  The census "long form" was weighted to overall census counts by age, race, sex and Hispanic origin.  Similarly, the ACS is weighted to the Census Bureau's annual estimates of household population by age, race, sex and Hispanic origin.

Despite being controlled to population estimates, ACS statistics sometimes differ from those estimates due to certain features of the weighting methodology.  For example, the Census Bureau computes weighting factors for entire counties or groups of counties rather than computing separate factors for individual cities.  Unfortunately, survey coverage can differ considerably between large cities and suburbs within the same county, or among the diverse communities in multi-county weighting areas.  Whenever that happens, population levels are overstated in some communities and understated in others.

Detroit is a good example of a city whose population is understated by the ACS.  Detroit's population was understated by 3.6 percent in 2005 and 4.5 percent in 2004 relative to the Census Bureau's annual population estimates.  Even the survey conducted in 2000 understated Detroit's population by 2.9 percent relative to the annual estimate and 3.3 percent relative to the 2000 Census.

One possible reason why cities tend to have lower survey coverage than their suburbs is that cities have more people in dwellings that are difficult to include in a survey.  Motel rooms and vehicles are classified as housing units only when the decennial census finds that they are occupied by people with no usual home elsewhere.  Any list of such dwellings that was compiled in a census might not be of much value in a subsequent sample survey — a van that was occupied on census day is not likely to be parked in the same place when a survey enumerator looks for it several years later, and a motel room that used to be occupied by someone with no other residence might be occupied by a tourist.

Another possible reason for lower survey coverage in big cities is that the Census Bureau can identify two-unit housing that is converted to single-unit housing more easily than it can identify single-unit housing that is converted to multiple-unit housing.  Units that are eliminated can be identified when survey forms are returned as undeliverable or when enumerators fail to find a unit that was included in the sample.  New apartments that are carved out of single-family homes, on the other hand, may not be identified until the Census Bureau canvasses neighborhoods in preparation for the next census.

(3) Faulty interpretation of changes in "moving averages " of ACS and CPS data

When working with data from the American Community Survey or Current Population Survey, it is generally advisable to combine figures from several years whenever feasible in order to reduce the effect of sampling error.  However, it is important to remember that the associated margins of error and confidence intervals are pertinent only for comparing figures that are statistically independent from one another, such as data for different geographic areas or for ranges of years that do not overlap. 

Although multi-year figures that overlap ("moving averages") are suitable for graphic depictions of long-term trends, they are not suitable for identifying year-to-year changes.  The change from one such figure to the next merely reflects the difference between the year that dropped out of the moving average and the year that was added.  There is generally no reason to suppose that the difference between a pair of non-adjacent years is more pertinent than the difference between the two most recent years, and the two differences can easily be in opposite directions.  Moreover, while it is valid to compare individual years using their own margins of error, it is not statistically valid to compare overlapping multi-year averages.  

Additional information can be found on the American Community Survey page of the Library of Michigan's census Web site, www.michigan.gov/census, or by contacting State Demographer Kenneth Darga at DargaK@michigan.gov.   

****

Michigan's census data and demography functions are part of the Library of Michigan, an agency of the Department of History, Arts and Libraries.  Dedicated to enriching quality of life and strengthening the economy by providing access to information, preserving and promoting Michigan's heritage and fostering cultural creativity, the department also includes the Mackinac Island State Park Commission, the Michigan Council for Arts and Cultural Affairs, the Michigan Film Office and the Michigan Historical Center.  To learn more, visit www.michigan.gov/hal.

# # #