A Comparison Between the Rating Scale Model and Dual Scaling for Likert Scales

Abstract

The fundamental requirement of using Likert
\nscales to measure affective behaviors-that all
\nrespondents must use the ordered response
\ncategories of all scale statements in the same
\nway-is examined. Scaling problems arise when
\nattitude statements within a Likert scale are
\nunfolding preference data, and when the position
\nand spacing of the ordered response categories are
\ninterpreted differently across scale statements and
\namong respondents. The assignment of a neutral
\nresponse category in the center of the response
\nscale is questionable. These problems pertain to
\nissues of the existence of an interval scale, the
\ndimensionality of the trait, and patterns of item
\nresponse functions. An attempt is made to resolve
\nthese problems using two contrasting scaling
\nmethods-item response theory modeling and dual
\nscaling. The primary result is that conformity of
\nresponses to the item response theory model
\nrequirements and targeting of scale statements and
\ntheir ordered response categories to the respondents
\nin dual scaling are vital for a resolution of
\nthese problems. This study establishes the similarity
\nof the two scaling methods. Index terms:
\ndual scaling, items response theory, jackknife technique,
\nLikert scaling, Rasch model, rating scale model.