Methodology review: Assessing unidimensionality of tests and items

Various methods for determining unidimensionalityare reviewed and the rationale of these methods is assessed.Indices based on answer patterns, reliability,components and factor analysis, and latent traits arereviewed. It is shown that many of the indices lack arationale, and that many are adjustments of a previousindex to take into account some criticisms of it. Afterreviewing many indices, it is suggested that thosebased on the size of residuals after fitting a two- orthree-parameter latent trait model may be the mostuseful to detect unidimensionality. An attempt is madeto clarify the term unidimensional, and it is shownhow it differs from other terms often used interchangeablysuch as reliability, internal consistency,and homogeneity. Reliability is defined as the ratio oftrue score variance to observed score variance. Internalconsistency denotes a group of methods that areintended to estimate reliability, are based on the variancesand the covariances of test items, and depend ononly one administration of a test. Homogeneity seemsto refer more specifically to the similarity of the itemcorrelations, but the term is often used as a synonymfor unidimensionality. The usefulness of the terms internalconsistency and homogeneity is questioned. Unidimensionalityis defined as the existence of one latenttrait underlying the data.