AN ITEM SELECTION PROCEDURE TO MAXIMISE SCALE RELIABILITY AND VALIDITY

In industrial- and social-psychological research, the variables of interest are often measured by means of Likert-type questionnaires or scales. The internal validity of the conclusions reached in such research directly depends on the reliability (e.g., internal consistency) and the validity of the questionnaires or scales used. Typically, item-analytic procedures have been used to arrive at sets of internally consistent items and subsequently the validity of these scales was investigated by correlating the total scores on them with criteria of interest. In the case of multi-dimensional scales, both the convergent validity of the respective subscales (i.e., the degree to which the items within a particular subscale measure the same uni-dimensional construct), and their discriminant validity (i.e., the degree to which the items in different subscales measure different rather than the same construct) need to be considered. The analyses to investigate internal consistency, convergent validity and discriminant validity cannot proceed in just any order as the analysis of some of these properties presupposes that particular results have been obtained for the other. For example, in the case of multi-dimensional scales it makes little sense to investigate the internal consistency of the total scale by means of Cronbach’s alpha, as this index assumes uni-dimensionality. Green, Lizzits and Mulaik (1977) have shown that items showing