Transformations to Improve Reliability and/or Validity for Affective Scales

TLDR: The authors showed that the use of normalized ranks or, in the case of large numbers of categories, normal deviates results in an increasing, monotonic relationship between reliability and number of categories.

Abstract: THE psychological literature is rich with studies demonstrating that collecting responses to affective stimuli using a large number of ordered categories is not effective when compared with using some smaller number (e.g., Matell and Jacoby, 1971; Komorita and Graham, 1965). Thus, a 20-point scale is no better, and perhaps worse, in terms of reliability and validity than, say, a 5-point scale. However, these studies uniformly rank the categories and analyze these ranks. There is a small, circumscribed body of literature (Liu, 1971 ; Warren, Klonglan, and Sabri, 1969) which indicates that the use of normalized ranks or, in the case of large numbers of categories, normal deviates results in an increasing, monotonic relationship between reliability and number of categories. Such transformations weight highly response differences in the ends of the scale