Attitude Surveys

History

Throughout the early and mid-twentieth century, the Thurstone scaling methods dominated the field of attitude measurement, including the method of equal-appearing intervals, the method of successive intervals, and the general paired-comparison methodology that was at the time known under the name “law of comparative judgment.” From the perspective of modern psychometric theory, the Thurstone scaling methods can be understood as early probabilistic item response theory models for the scaling of attitudinally relevant stimuli. The first methods that simultaneously scaled stimuli and people emerged in the mid-twentieth century, including the Guttman scalogram analysis and the Coombs unfolding model. Although their initial formulations were deterministic, probabilistic versions such as the Rasch two-parameter logistic model were soon developed. A common characteristic of all these early methods is that they were based on stringent psychometric models that allowed for goodness-of-fit tests and checks of the resulting level of measurement. However, they often involved complex computations and could become quite onerous at a time when computers were still rare and costly. The Likert method of summated ratings was originally developed as a quick-and-easy alternative to Thurstone methods. The Likert method omitted all stimulus scaling and only scaled people, treating the items in an attitude scale as equivalent and assuming a priori that the resulting person score would have interval-scale level.

Current State of the Art

Likert scales are still the dominant method of attitude measurement. Their construction is simple and involves four steps. In the first step, the researcher generates a pool of items that indicate either a positive or a negative evaluation of the attitude object. In the second step, responses to these items are collected in an attitude survey, using a rating-scale response format (between five and eleven response categories, with the extreme categories labeled “strongly disagree” and “strongly agree”). In the third step, the resulting data are subjected to item and scale analysis in the tradition of classical test theory. This involves assessment of scale dimensionality by means of factor analysis, selection of items with high discrimination and medium difficulty, elimination of ambiguous and nondiscriminating items, and assessment of the [Page 70]reliability and validity of the resulting scale. All items are assumed to have a linear relationship with the underlying attitude dimension. In the fourth step, person scores are computed for each participant. A participant's score is the sum of her or his responses to the items in the final scale. The person score is assumed to be continuous and to have an interval-scale level of measurement. An important special case of the Likert method is the semantic differential. In addition to a vignette that describes the attitude object, a semantic differential consists of bipolar Likert items in which the extreme categories are labeled with evaluative opposites. Very often, the labels “bad-good,” “unpleasant-pleasant,” “unfavorable-favorable,” and “negative-positive” are used. The semantic differential is the most popular way of measuring attitudes.

...