Item analysis of single-peaked response data : the psychometric evaluation of bipolar measurement scales

Abstract

The thesis explains the fundamental difference between unipolar and

bipolar measurement scales for psychological characteristics. We explore

the use of correspondence analysis (CA), a technique that is similar to

principal component analysis and is available in SAS and SPSS, to select items that together form a bipolar scale. CA estimates both item locations and subject locations and is a useful tool in the psychometric evaluation of bipolar measurement scales. Additionally we present a new methodology (OCM) to estimate item response functions of single-peaked items with a model-free approach.