(a) Relevance & Research Question
Online surveys are fast, efficient, and especially cheap. Self-administered surveys might even produce higher data quality compared to personal or phone interviews, e.g. for lower social desirability bias. However, web surveys also face specific problems such as the absence of a human interviewer. Among other things, this limits interview supervision, possibly causing higher item nonresponse and increasing satisficing behavior.
Generally, survey results largely differ depending on collection mode. For instance, recency effects are more likely in oral interviews, which also tend to produce more extreme answers and acquiescence bias, while primacy effects are associated with visual questionnaire presentation. Furthermore, mixed-mode surveys have received growing attention. Since mode effects are of particular relevance there, we will also cover this aspect.
Against this background, analyzing mode effects between traditional modes and web surveys, we show to what extent the latter can be a suitable alternative.
(b) Methods & Data
In the first step, data from three different GLES components are compared, including face-to-face, CATI, and online surveys.
In the second step, we use data from GLES long-term panel. As the second panel wave was collected in a mixed-mode design combining CATI with mail interviews, we compare changes for the sub-samples relative to the first (face-to-face) panel wave.
In terms of methods, we primarily rely on univariate statistics (frequencies, distributions, mean values, variance). We focus on well-established items measuring electoral choice, political attitudes, and psychological predispositions. Beyond, we estimate multivariate models, including interaction terms with survey mode.
(c) Results
Preliminary results uncover a number of clear mode effects between personal, phone, and web surveys. These differences are anything but uniform, though. Rather, we observe differences for specific variables and under specific conditions only.
(d) Added Value
Never before in German electoral science has a similarly coherent questionnaire been implemented in three different survey modes during the same field period. GLES data therefore allow for detailed and unprecedented analyses of mode effects, including mixed-mode surveys. Our results add to the understanding of differences between personal, phone, and web surveys, and suggest what can be done to prevent those differences.