In the social sciences and academic research, traditional survey modes like face-to-face or telephone surveys are increasingly being replaced by surveys conducted via online panels. As no interviewers are involved, these panels offer a distinctively cheaper and faster possibility of data collection than the traditional modes. However, concerns prevail that they cannot meet the high quality standards needed for scientific research. Commercial online panels rarely use probability-based sampling strategies to recruit their sample members, which is a prerequisite for population-based inference. Some comparative studies have shown that the samples of such non-probability online panels lack representativeness of the general population and lead to less accurate data than traditional probability-based offline surveys. This presentation contributes to this line of inquiry by assessing the representativeness of a series of non-probability and probability web surveys conducted in Germany. In line with other studies, our results indicate that online panel data based on probability samples is more representative of the target population than data from non-probability online panels. I go on to consider the potential utility of non-probability surveys when used in conjunction with probability surveys for population-based inference. A Bayesian modeling approach that combines both types of surveys is presented and demonstrated on the same set of German web surveys noted above. Preliminary results show that the combined modeling approach yields better mean-squared error properties compared to the probability survey alone. I conclude the presentation with a general discussion of the implications of these results for survey practice.
Presented by:
Joseph Sakshaug, University of Manchester
Date & time:
March 20, 2017 4:00 pm
Venue:
2N2.4.16 - ISER Seminar room
External seminars home