ing the respondent’s inclusion probabilities is achieved” (p. 89). Peytchev contends that this fact has inarguably contributed to the widespread interpretation of the response rate as a summary measure of a survey’s representativeness. However, response rates can be misleading as measures of survey representativeness. The fact that response rates have fallen (as documented in Chapter 1) means only that the potential for nonresponse bias has increased, not necessarily that nonresponse bias has become more of a problem. That is because nonresponse bias is a function of both the nonresponse rate and the difference between respondents and nonrespondents on the statistic of interest, so high nonresponse rates could yield low nonresponse errors if the difference between respondents and nonrespondents is quite small or, in survey methodology terms, if nonresponse in the survey is ignorable and the data can be used to make valid inferences about the target population.
Moreover, it would be a relatively simple matter to overcome the problem of bias in the estimates brought about by nonresponse if there were a linear relationship between response rates and nonresponse bias across surveys. If it were so, one could theoretically reduce nonresponse bias by taking actions to increase response rates, and more effort, cost, training, and management control of the survey operation would solve the problem. This is not the case, however, as shown by Curtin et al. (2000), Groves et al. (2006), and Groves and Peytcheva (2008). The 2008 Groves and Peytcheva compilation of the results of 59 specialized studies found very little correlation between nonresponse rate and their measures of bias. Likewise, there is no proof that efforts to enhance response rates within the context of a survey will automatically reduce nonresponse bias on survey estimates (Curtin et al., 2000; Keeter et al., 2000; Merkle and Edelman, 2002; Groves, 2006).
Recommendation 2-1: Research is needed on the relationship between nonresponse rates and nonresponse bias and on the variables that determine when such a relationship is likely.
It is possible that extraordinary efforts to secure responses from a reluctant population may even increase bias on some survey estimates (Merkle et al., 1998). A 2010 study by Fricker and Tourangeau suggested that efforts to increase response can lead to quality degradation. The authors examined nonresponse and data quality in two national household surveys: the Current Population Survey (CPS) and the American Time Use Survey (ATUS). Response propensity models were developed for each survey. Data quality was measured through such indirect indicators of response error as item nonresponse rates, rounded value reports, and interview–response inconsistencies. When there was evidence of covariation between