Drop-out in a Longitudinal Survey of Amazon Mechanical Turk Workers with Low Back Pain: an Observational Study (Preprint)

BACKGROUND Online surveys of internet panels such as Amazon’s Mechanical Turk (MTurk) are common in health research. Non-response in longitudinal studies can limit inferences about change over time. OBJECTIVE We (1) describe the patterns of survey responses and non-response among MTurk members with back pain, (2) identify factors associated with survey response over time, (3) assess the impact of non-response on sample characteristics, and (4) assess how well inverse probability weighting can account for differences in sample composition. METHODS We surveyed MTurk adults who identified as having back pain. We report participation trends over three survey waves and use stepwise logistic regression to identify factors related to survey participation in successive waves. RESULTS A total of 1,678 adults participated in Wave 1. Of those, 983 (59%) participated in Wave 2 and 703 (42%) in Wave 3. Participants who did not drop out took less time to complete prior surveys (30 minutes vs. 35 minutes in Wave 1, p<0.005; 24 minutes vs. 26 minutes in Wave 2, p=0.019) and reported having fewer health conditions (6 vs. 7, p<0.005). In multivariate models, higher odds of participation were associated with less time to complete the baseline survey, older age, not being Hispanic, not having a bachelor’s degree, being divorced or never married, having less pain interference and intensity, and having more health conditions. Weighted analysis showed slight differences in sample demographics and conditions, and larger differences in pain assessments, particularly for those who responded to Wave 2. CONCLUSIONS Longitudinal studies on MTurk have large, differential dropouts between waves. This study provided information about the types of individuals who are more likely to drop out over time which can help researchers prepare for future surveys.