American Community Survey Data

The American Community Survey (ACS) is a fairly recent product from the US Census Bureau that is replacing the Decennial long form.  The data tables (Summary File 3) that were prepared from the responses to detailed questions included on the Decennial long form will no longer be available as part of the 2010 Census since every household will only receive a short form. With the elimination of the long form, detailed characteristics of the population will be derived from the results of the data collected only through the annual American Community Survey which is sent to 250,000 households nationwide on a monthly basis. There are significant differences between the long form and the ACS and we recommend that data users exercise extreme caution when comparing ACS data with 2000 decennial data. The following are some of the more significant guidelines that should be followed when using the ACS data.

• Whereas the 2000 data derived from the long form was an estimate of the population at a specific point in time, the ACS is a period estimate; that is, the survey is issued to a much smaller sample on a monthly basis, which are then pooled into one, three, and five year periods. Due to the small sample size, one, three, and five year estimates are only available for communities greater than 65,000, three and five year estimates are available for communities greater than 20,000 population, and only five year estimates will be available beginning in 2010 for geographic areas, including census tracts and block groups, with a population of less than 20,000. When discussing the multi-year period estimates, the best way to characterize the data is to state that the data cover a period of years, rather than use the year of release. For example, for a 3 year estimate, released in 2008, the data might be characterized as, “The poverty rate in this community over the period 2005-2007 was X”.
• If comparisons are to be made between various communities, care should be taken to ensure that the same period estimates are used; 1 year data should only be compared with other 1 year data; 3 year data compared only with other 3 year data and 5 year data with other 5 year data. With the 3 and 5 year multi-year estimates, time series comparisons should only be done with non-overlapping data; for example with a 3 year estimate, the 2005, 2006, 2007 product should be compared to the 2008, 2009, 2010 product.
• While 1 year data, when available, is more current, the smaller sample size means that it is also less accurate and often highly volatile year-to-year. Which period estimate is used may ultimately be determined by programming agencies; however CDRPC considers accuracy to be more important than timeliness for most uses, and indeed using inaccurate data could have harmful and detrimental effects on local programming and analytical needs. For that reason, we recommend whenever possible that the 5 year data be used to ensure the most accurate data.
• Because of the accuracy issues, the US Census Bureau is publishing the margins of error (MOE) or ranges for all ACS estimates at the 90% confidence level.  It may be more helpful to express an estimate as a range rather than a single number with an MOE. The Census Bureau ACS guidebooks recommend using extreme caution in using any estimate where the MOE is 10% or more of the estimate. You might also consider the size of the MOE in comparison to the total population in the area under analysis.
• When using the ACS data to determine change over time, note that due to the smaller sample in the 1 year ACS estimates, the ACS estimates can show large variation from one year to the next. We recommend reviewing several years’ worth of one year estimates, combined with their MOEs. If there is overlap between the MOEs (if the year to year change in the characteristic is less than the MOE) then there is most likely no significant difference between the estimates. For example, if the result in the one year ACS estimate was “100” with a range between 93 and 107 and the following year it was “105”, then there is most likely no significant difference between the estimates.
• When comparing the ACS estimates to the 2000 Census data, make sure that the questions are comparable – a table showing those items that can be compared is available at the US Census website. As examples, the ACS definition of place of residence differs significantly from the Census, which can have considerable impact on the population data for areas that experience seasonal housing changes due to universities, large number of vacation homes, etc. Additionally, the ACS data collected up to 2005 did not include a survey of group quarters and therefore comparing estimates should be performed with caution after reviewing the US Census Bureau’s comparability charts.

Additional Resources for Using the ACS Data

In many, if not most, cases, the ACS data have margins of error that are similar to those of the 2000 Census data. Checking whether this is so requires comparing the variability of the ACS and Census data using either the standard error for quantitatively similar data or the coefficient of variation (CV), which allows comparisons of the variability of even extremely diverse data sets.

The Missouri Census Data Center has a concise general description and instructions for using the American Community Survey data.

The use of time series or cross-sectional data from the ACS, because of the variability of data collection schedules and base populations sampled, requires the calculation of the statistical significance of differences.

The State Data Center/Business Industry Data Center Network has an Excel worksheet for computing standard errors and statistical significance for the ACS data.

American Community Survey Access

Return to Index of Data and Maps