Assessment of Nonresponse Bias in the Non-Pharmaceutical Interventions for Influenza Telephone Survey

The recent decline in telephone survey response rates compels researchers to assess the potential for nonresponse bias in most telephone survey data. One challenge to analyzing nonresponse bias is that limited information about nonrespondents is available for most surveys. Census data can usually provide demographic figures for the target population, but other reliable sources must be used to provide benchmarks on how demographic characteristics may be related to the survey measures of interest. Accepting this challenge, we assess nonresponse bias in a community health survey conducted in four counties in North Carolina. This survey focused primarily on residents' knowledge, attitudes, and practices with respect to the role of non-pharmaceutical steps for preventing the spread of influenza. We use Census data to assess how survey respondents and nonrespondents differ on demographic characteristics. Other sources, including the Behavioral Risk Factors Surveillance System and National Health Interview Survey, indicate how these demographic differences may be related to key survey measures. The key measures include steps taken to avoid contracting influenza, incidence of an influenza-like illness, and number of daily social interactions. As two examples, higher representation among women versus men likely contributed to an overestimate of the proportion of residents receiving a flu shot; higher representation among residents over age 65 versus younger residents likely contributed to an underestimate of the average number of social interactions. We discuss the potential impact of telephone survey nonresponse on community health surveys and steps researchers can take to both prevent and assess nonresponse bias in surveys.