tion partners (see Section 7-A above) could be a valuable resource to help educate users about the meaning and value of the various indicators.
This network could also help the Census Bureau determine what additional indicators to consider adding to the web site. For example, for the 2005 ACS, it could be very useful to provide all four quality measures for individual public use microdata areas (PUMAs) to help users track basic data quality for substate areas.
Looking ahead, it would be very useful for the Census Bureau to periodically issue reports that highlight patterns of basic quality measures over time for geographic areas and population groups of interest—for example, whether (and which) item nonresponse rates are increasing or decreasing and for which areas and groups. Similarly, it would be useful for the Census Bureau to analyze unit and item nonresponse rates separately by data collection mode (mail, computer-assisted telephone interviewing, CATI, computer-assisted personal interviewing, CAPI) to see if there are patterns by geographic location or such characteristics as education level, family structure, and others. It will also be important for the Census Bureau and its network of user education partners to determine the most useful set of quality measures for the 3-year and 5-year period estimates for small areas in addition to those provided for 1-year period estimates for larger areas.
Recommendation 7-5: The Census Bureau, in collaboration with user education partners, should carry out research on ways to facilitate understanding of the quality measures provided on the ACS web site. The Census Bureau and its partners should also consider what additional quality indicators—for example, some of the indicators presented at a finer level of geographic detail—would be useful to provide for the 2005 ACS and subsequent 1-year period estimates and what indicators to provide for the 3-year and 5-year period ACS estimates when those become available.
The Census Bureau provides a measure of sampling error for each sample-based estimate that is released in tabular form from the ACS. This measure is developed using a repeated replication method (see U.S. Census Bureau, 2006:Ch. 12). The published measure of sampling error is the margin of error around the estimate (plus or minus) at the 90 percent confidence level (1.65 times the standard error), not the commonly accepted 95 percent level, which is 1.96 times the standard error (see Box 2-5 for explanations of these terms).