American Community Survey (ACS)

For certain group quarters (GQ) characteristics and GQ types, the item non-response rates and coverage rates do not meet the Census Bureau's quality standards. The GQ coverage rate of 79.6 percent was below the standard of 80 percent. Most notably, the coverage rate for college dormitories was artificially low because data were collected throughout the entire year, including the summer months when many dormitories were vacant. This, in turn, lowers the coverage for the GQ population as a whole. Missing data for a particular question is called item nonresponse. It will occur when a respondent fails to provide an answer to a question or when an invalid answer is provided. The Census Bureau uses allocation methods to correct for item nonresponse and thus, allocation rates are used to measure the magnitude of item nonresponse. The Census Bureau has established a standard states that users should be made aware of items with allocation rates exceeding 20 percent. Several GQ population characteristics in the subject tables did not meet this standard. For example, item allocation rates for characteristics such as year of entry and income were high for certain GQ types and for the GQ population as a whole. A separate document [XLS] provides item allocation rates for GQ characteristics that did not meet the Census Bureau's quality standard. High item allocation rates and low coverage rates can adversely impact final estimates for the GQ population and introduce bias if the characteristics of nonrespondents differ from those of respondents. If the GQ population makes up a large proportion of the total population for an area, the estimates for the total population can likewise be affected.

Selected sections of the iterated SPPs by country of birth show a '0,' 'N,' '-,' or '**,' for estimates and margins of error that are not applicable and should have been represented with an '(X)'. This affects the native iteration and the region, subregion, and country of birth iterations for "Place of Birth, Citizenship Status and Year of Entry" and "World Region of Birth of Foreign Born." Also affected is the total foreign-born iteration for "Place of Birth, Citizenship Status and Year of Entry." For example, in the total foreign-born table, the estimate of natives under "Place of Birth, Citizenship Status and Year of Entry" was reported as '0' with a margin of error of '+/-267' when both measures should have received an '(X).' The corrected SPPs will be released prior to the release of 3-year estimates in December 2008.

The review of the 2007 ACS tabulations revealed large discrepancies between the final 2007 ACS estimate of total population and the 2007 population estimates from the Population Estimates Program for a set of places in Orange County, CA. These same areas showed a similar pattern of differences in the 2005 and 2006 ACS estimates. An additional adjustment to the weights was introduced in the weighting for Orange County to remove the source of the discrepancy. The adjustment was aimed at forcing the weighted distribution of housing units in the ACS sample to match the distribution of housing units on our sampling frame. The weighting adjustment was applied at the census tract level. As a result, it has eliminated most of the large discrepancies originally found for places in Orange County. This modification was also made to the 2006 ACS weighting methodology.

The review of the 2007 operational data discovered evidence that suggests a high incidence of misclassification of uninhabitable units as vacant units. The effect of misclassification was almost entirely removed through a modification in the weighting methodology for Orleans and St. Bernard Parishes. The effect of the weighting adjustment was to down-weight units that had the vacancy status of "Other Vacant". This modification resulted in more consistent and accurate ACS estimates of the number of vacant units and "persons per household" in these two parishes. This modification was also made to the 2006 ACS weighting methodology.

The weighting for the group quarters population is controlled at the state level. For that reason, users should expect that the ACS estimate for the group quarters population to be consistent with the Population Estimates Program only at the state level. At the county level, the ACS estimate will typically be different from the Population Estimates Program due to sampling variability. Like estimates of total population, data users who require an estimate of the group quarters population at the county level should use the official estimates from the Population Estimates Program.

The weighting for the group quarters (GQ) population is controlled at the state level, but not at sub-state levels. For this reason, users may observe greater fluctuations in year-to-year ACS estimates of the GQ population at sub-state levels than at state levels. The causes of these fluctuations typically are the result of either GQs that have closed or where the current population of the GQ is significantly different than the expected population as reflected on the sampling frame. Substantial changes in the ACS GQ estimates can impact ACS estimates of total population characteristics for areas where either the GQ population is a substantial proportion of the total population or where the GQ population may have very different characteristics than the total population as a whole. Users can assess the impact that year-to-year changes in sub-state GQ total population estimates have on the changes in total ACS population estimates by accessing Table B26001 on American Fact Finder. Users should also use their local knowledge to help determine whether the year-to-year change in the ACS estimate represents a real change in the GQ population or may be the result of the lack of adequate population controls for sub-state areas.

In Abilene, TX, the Census Bureau has identified a large increase in the percent of Hispanics reporting their race as White Alone and a large decrease in the percent of Hispanics reporting their race as Some Other Race. This increase is likely the result of a problem in the data collection process and probably does not represent real change. Data users should use caution when interpreting the data for Hispanic respondents who reported their race as White or Some Other Race.

The review of the 2006 ACS tabulations revealed large discrepancies between the final 2006 ACS estimate of total population and the 2006 intercensal population estimates for a set of places in Orange County, CA. These same areas showed a similar pattern of differences in the 2005 ACS tabulated estimates. An additional adjustment to the weights was introduced in the weighting for Orange County to remove the source of the discrepancy. The adjustment was aimed at forcing the weighted distribution of housing units in the ACS sample to match the distribution of housing units on our sampling frame. The weighting adjustment was applied at the census tract level. As a result, it has eliminated most of the large discrepancies originally found for places in Orange County.

The review of the 2006 operational data discovered evidence that suggests a high incidence of misclassification of uninhabitable units as vacant units. The effect of misclassification was almost entirely removed through a modification in the weighting methodology for Orleans and St. Bernard Parishes. The effect of the weighting adjustment was to down-weight units that had the vacancy status of "Other Vacant". This modification resulted in more consistent and accurate ACS estimates of the number of vacant units and "persons per household" in these two parishes.

As a result of our continuous effort to improve the quality of the estimates the 2006 weighting methodology was modified in order to ensure total consistency between the estimates of householders, households, and occupied housing units. This has not been the case prior to 2006. With this data release, the estimates of occupied housing units, households, and householders will all be equal. In addition, the modification also reduces the differences between estimates of spouses and married-couple households and between the estimates of unmarried partners and unmarried-partner households. See the Changes to the Estimation (pdf) for detailed information regarding which tables are most affected and the impact on the estimates had this methodology been applied in 2005.

The weighting for the group quarters population is controlled at the state level. For that reason, users should expect that the ACS estimate for the group quarters population to be consistent with the Population Estimates Program only at the state level. For sub-state areas, the ACS estimate will typically be different from the Population Estimates Program due to sampling variability. Like estimates of total population, data users who require an estimate of the group quarters population at the county level should use the official estimates from the Population Estimates Program.

Here is how we determine which sub-state geographic areas meet the ACS population threshold of 65,000 for publishing estimates for that area: Wherever possible, we use the most current estimate of the total resident population from official Census Bureau Population Estimates Program (PEP). This number includes both household and group quarters population. PEP provides this estimate for counties, incorporated places, and sub-county areas (e.g. townships), which have a functioning government. If the PEP estimate for any of these areas is 65,000 or more, then we publish ACS data products for that area. However, since the ACS sample is still a sample of the housing unit population, the ACS estimates will often be lower than the PEP estimates. In some cases, the ACS estimate of total (housing unit) population will be less than 65,000. This is an indication that the total resident population for that geographic area is over 65,000 (although probably very close to 65,000). There are 52 geographic areas where this occurs, and they are shown in the table below.

Data Release Rules

Even with the population size thresholds described earlier, in certain geographic areas some very detailed tables might include estimates whose reliability is unacceptable. Data release rules, based on the statistical reliability of the survey estimates, will be used starting with the 2005 ACS data released in the summer of 2006. These release rules apply only to the single-year and three-year data products.

The main data release rule for the ACS tables works as follows. Every base table consists of a series of estimates. If more than half the estimates are not statistically different from 0 (at a 90 percent confidence level), then the table fails. Each estimate is subject to sampling variability that can be summarized by its standard error. Dividing the standard error by the estimate yields the coefficient of variation (CV) for each of the estimates. (If the estimate is 0, a CV of 100 percent is assigned.) To implement this requirement for each table at a given geographic area, CVs are calculated for each of the table's estimates, and the median CV value is determined. If 13-11 the median CV value for the table is less than or equal to 61 percent, the table passes for that geographic area; if it is greater than 61 percent, the table fails. Tables that are too sparse will fail this test. In that case, the table will not be published for that geographic area. Whenever a table fails, a simpler table that collapses some of the detailed lines together can be substituted for the original, more detailed table. The data release rules are then applied to the simpler table. If it passes, the simpler table is released. If it fails, none of the estimates for that particular table is released for this geographic area. These release rules are applied to single-year period estimates and multi-year period estimates based on three years of sample data. No data release rules are applied to the estimates based on five years of sample data.

For more information go to the Design and Methodology document, the link to it is http://www.census.gov/acs/www/Downloads/tp67.pdf

Update October 2, 2008: Partial ACS data products are restored back to AFF. Detailed tables and reference maps for 2000-2004 and data profiles for 2002-2004 are now available. We will continue to restore the data and will provide more updates as the files become available.

On August 26, 2008, all ACS data products for the years 2000, 2001, 2002, 2003, and 2004 were removed from the American FactFinder (AFF) and placed on a special FTP site (http://www2.census.gov/acs/downloads/Core_Tables/). However, there were several problems that users discovered with the files on this site, and, based on the reports from users of these problems, on September 9, 2008 the Census Bureau decided to pull this site down until the problems can be fixed. We expect resolution soon, and we will make the FTP site available at that time.

In the meantime, for years 2002, 2003, and 2004, Detailed Tables, Profiles, and Ranking Tables are available for downloading on:

• http://www2.census.gov/acs2002/
• http://www2.census.gov/acs2003/
• http://www2.census.gov/acs2004/

For years between 2000 and 2003, Profiles, Multiyear Profiles, and Change Profiles are available at links below. In addition, ranking tables for years 2003, 2002, 2001, and 2000 are also available on the ACS Website - located on: http://www.census.gov/acs/www/Products/Ranking/index.html.

• 2003 Data Profiles: http://www.census.gov/acs/www/Products/Profiles/Single/2003/ACS/
• 2003 Multiyear Profiles: http://www.census.gov/acs/www/Products/Profiles/Chg/2003/ACS/
• 2002 Data Profiles: http://www.census.gov/acs/www/Products/Profiles/Single/2002/ACS/
• 2001-2002 Change Profiles: http://www.census.gov/acs/www/Products/Profiles/Chg/2002/0102/
• 2000-2002 Change Profiles: http://www.census.gov/acs/www/Products/Profiles/Chg/2002/0002/
• 2001 Data Profiles: http://www.census.gov/acs/www/Products/Profiles/Single/2001/ACS/
• 2000-2001 Change Profiles: http://www.census.gov/acs/www/Products/Profiles/Chg/2001/ACS/
• 2000 Data Profiles: http://www.census.gov/acs/www/Products/Profiles/Single/2000/ACS/