Combining Information from Multiple Modes to Reduce Nonresponse Bias

Over 3,000 subjects were recruited in 3 U.S. regions for a randomized experiment of an online weight management intervention. Participants were sent invitations to web survey reassessments after 3, 6, and 12 months. High and increasing nonresponse to the three follow-up surveys created the potential for nonresponse bias in key program outcomes. A subsample of the nonrespondents at the one-year follow-up was selected for a nonresponse study. This subsample was then randomly assigned to a short telephone or mail survey. This was done in order to evaluate cost efficiency, differential effectiveness of mode combinations in reducing nonresponse bias, and measurement differences by mode. The responses from the nonresponse study were then to be added to the baseline measures and used in an imputation model. Differences between the telephone and mail survey reports posed an added methodological problem, allowing further exploration of sensitivity of the results not just to nonresponse, but also to the mode used in the second stage through comparison of different imputation models. Implications are discussed for cost, nonresponse bias, measurement differences, and post-imputation variance estimates