Abstract:
|
We estimate the magnitude of three sources of error in a mobile Web survey. 1390 members of the Longitudinal Internet Studies for the Social Sciences (LISS) panel were asked to answer the same questions, once using a smartphone and once using a personal computer (PC). We use the PC Web survey as a benchmark, and deviations from the benchmark are regarded as error. To estimate coverage errors in the mobile Web survey, we compared those with their own smartphones (71%) to the full sample. To estimate nonresponse errors, we compared those who responded on their smartphone (73%) to the covered sample. Finally, to estimate measurement differences, we compared how the same people responded when using their smartphone and when using their PC. We find large non-coverage error errors relative to nonresponse and measurement errors. Furthermore, the non-coverage errors were not consistently offset by the other sources of error. This suggests that limiting Web surveys to mobile Web users only is risky for general population surveys. This research is a first step towards understanding the effect of mobile-only Web surveys on data quality using a total survey error framework.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.