Effects of nonresponse and coverage problems on survey estimates of physical activity

To what extent do survey nonresponse and coverage problems bias estimates of physical activity? Research has focused on social desirability as a cause of observed bias but there are relatively few studies of nonresponse and coverage. I analyze data from a survey designed to allow estimation of nonresponse bias using a two-phase sampling design that resamples nonrespondents from an initial wave of telephone and IVR (interactive voice response) interviewing for follow-up face-to-face interviews. Both initial and nonresponse interview waves included measures of physical activity frequency and duration. Estimates are compared between first-round respondents and those from follow-up interviews of nonrespondents, accounting for mode and other design elements. Telephone, but not IVR, interviews were found to include bias from two sources. Findings suggest that coverage is a cause of bias in the measure of frequency of physical activity but nonresponse may bias the measure of physical activity duration.