Relevance & Research Question: Verbal and numeric labels in scalar questions have received a lot of scholarly attention. It is well documented that verbal orientation of scales and adding numbers to anchor the scales influence respondents’ answers. Little is known, however, about switching the scale directions within one questionnaire. This may affect the answers as some respondents miss the fact of the switch (Dillman et al. 2009).
Methods & Data: We test whether switching the verbal orientation of scales and whether the choice of a starting point of numbered scales affect the answers. For that purpose, two experiments will be implemented in the upcoming wave of the GESIS Panel, a probability-based offline recruited mixed-mode (online and mail) panel of the general population in Germany. The first experiment tests the effects of switching the direction of the scale (“strongly agree” to “strongly disagree” and vice versa). The second experiment tests different labeling options. It has three conditions: 1) verbal labeling for end points only and no numbers, 2) fully labeled, starting with 0 and 3) fully labeled, starting with 1.
Results: The field time of the questionnaire starts on December 12, 2013 and ends in the middle of January 2014. The results will be available at that point. The analysis plan for the first experiment is to examine measurement error by comparing means and response distributions. We expect these to differ due to switching as respondents overlook that the order has been reversed. Effects are expected to be more pronounced among respondents who tend to satisfice. Thus, we will compare respondent groups by motivation, cognitive ability, and panel tenure. For the second experiment we will examine the occurrence of a midpoint shift and which of the numbered scales is closer to the scale with polar labels only.
Added Value: This presentation will be of interest to survey researchers who work with online access panels as well as online access panel providers. Often the questionnaire is put together by multiple researchers or clients in which case using multiple scales and various numbering schemes within one questionnaire are very likely.