The American Community Survey (ACS) has experienced a high response rate since full implementation began in 2005. Overall weighted response rates between 2005 and 2011 range from 97.3 percent in 2005 to 98.0 percent in 2009 (U.S. Census Bureau (2012)). These rates take all three modes of data collection into account (mail, telephone, and personal visit). Vacant housing unit addresses are included in these rates as they are interviews for the ACS. See U.S. Census Bureau (2009) for details.

Although these response rates are high, two to three percent of cases still did not respond. In this evaluation we want to determine whether the nonrespondents are categorically different in any way from the respondents, i.e., are the respondents representative of the nonrespondents and, consequently, of their entire sample? Then, since we assume that each yearly ACS sample is representative of the frame from which it was sampled, we can simultaneously answer the question of whether the respondents are representative of their corresponding frame as well.

The primary statistic we use in measuring representivity is the R-indicator. It is a measure of the spread of response propensities (probabilities of a sample case responding in the survey) across both respondents and nonrespondents. We also look at sample completeness ratios for comparison purposes, which are measures of the combined levels of nonresponse and under- or overcoverage.

Our analysis in this evaluation focuses on the United States and Puerto Rico. We estimate sample representivity at the national level as a whole and by various subgroups, e.g., race categories. We anticipate that the methods and results in this evaluation will serve as a springboard for future representivity research, for both the ACS and other surveys.