Improving Likert Scale Raw Scores Interpretability with K-means Clustering

Abstract

In this article, by applying k-means clustering, cut-off points are obtained for the recoding of raw scale scores into a fixed number of groupings that preserve the original scoring. The method is demonstrated on a Likert scale measuring xenophobia that was used in a large-scale sample survey conducted in Northern Greece by the National Centre for Social Research. Applying split-half samples and fuzzy c-means clustering, the stability of the proposed solution is validated empirically. Testing its performance against three single indicators of xenophobia shows that it differentiates well between non-xenophobic and xenophobic respondents. The proposed method may be easily applied to facilitate interpretation by providing a more concise and meaningful “profile” of Likert scale (or subscale) raw scores especially the negative and positive ends of the scale for evaluation and social policy purposes.