Person-Related Selection Bias in Mobile Sensing Research: Robust Findings from Two Panel Studies

In psychology, mobile sensing is increasingly used to record behavior in real-life situations.However, little is known about the selectivity of samples participating in these new datacollection approaches and thus about potential risks to the validity of research findings. We thereforeinvestigated coverage error and nonresponse error as two sources of selection bias insmartphone-based data collections. Specifically, we examined whether smartphone system ownership(Android versus iOS, i.e., coverage error) and willingness to participate (nonparticipationversus intention to participate versus actual participation, i.e., nonresponse error) are associatedwith socio-demographic, socio-economic, and personality characteristics.Using two large-scale panel studies, we found replicable patterns for coverage error (N =1,218; N = 5,123) and nonresponse error (N = 1,673; N = 2,337): The ownership of Androiddevices (in comparison to iOS devices) was associated with lower levels of education, income,and extraversion. The willingness to participate in mobile sensing studies was found to be higheramong younger age groups, males, those with higher levels of openness to experience, and thosewith lower levels of neuroticism. Furthermore, different person characteristics played differentroles at different stages of the recruitment process. Taken together, the results show that some selectionbias in mobile sensing studies exists and that the effects were small to moderate in magnitudeas well as comparable to selection bias for other, more common data collection approaches,such as online surveys.