The unidimensionality of a psychiatric scale: A statistical point of view

Abstract

In quantitative psychopathology, one of the most crucial questions is whether a set of items measure just one thing in common. This property may be defined as unidimensionality. After a formal definition ofunidimensionality, this paper discusses the use of the different tools traditionally related to this area. For quantitative measurements, factor analysis remains a good approach, goodness'of-fit tests, however, are of questionable value. Cronbach's alpha coefficient is more related to reliability than unidimensionality. The scree plot and the proportion of variance accounted for by the first principal component are, in practice, interesting tools. Item response theory leads to models that require unidimensionality to obtain efficient estimates of latent attributes; these methods, however, arenotreally adapted for assessing the unidimensionality of a set of items. Finally, even if unidimensionality is a fundamental psychometric property, there is a need for general multidimensional instruments that reflect the heterogeneity of psychiatric disorders.