Design aspects of rating scales in questionnaires

Abstract

Since their introduction by Thurstone (1929) and Likert (1932), rating scales have been determinant in questionnaires. A rating scale usually defines the graduations out of a continuum such as agreement, intensity, frequency, or satisfaction. Respondents evaluate questions and items by marking the appropriate category, which usually concerns personal characteristics, opinions, and behavior Parducci (1983) defines responses as functions of the continuum of a rating scale. They range between the end poles and depend on the graduation of the scale; however, their quality should not be influenced by the characteristics of the rating scale. Menold and Bogner (2016) review characteristics of rating scales as follows: total number of categories, usage of middle and “do not know” options, category labeling, scale orientation (starting with a negative or a positive value, or a lower or a higher value), scale polarity (usage of verbal opposites), and visual presentation. The best design of rating scales remains controversial. Moreover, characteristics of rating scales can affect the quality of measurement (Krosnick and Fabrigar, 1997). Menold and Tausch (2016) demonstrate that different total numbers of categories have different psychometric properties and that verbalisation affects measurement. Data can no longer be compared if they have been produced from different rating scales. Graduations used in rating scales involve metric properties, because they are supposed to correspond to equal differences between categories. Orth (1982) and Westermann (1985) criticized this assumption of equidistance. The socalled “visual design” (Christian and Dillman, 2004; Tourangeau et al., 2007) was introduced to make out the influence of the graphical presentation of rating scales on response. Responses could, conciously or not, be biased by the graphical features. According to Schaefer and Dykema (2011: 912), “although past research often allows us to predict how a marginal distribution will be affected ... we are too often unable to say which version of a question is more reliable or valid.” That is why the reliability and validity of rating scales constitute an issue. This special