John Hattie reviews methods for assessing unidimensionality of tests and items. He discusses indices based on answer patterns, reliability, components, factor analysis, and latent traits. Many indices lack a clear rationale and are adjustments of previous indices. He suggests that indices based on residuals after fitting a two- or three-parameter latent trait model may be most useful for detecting unidimensionality. Unidimensionality is defined as the existence of one latent trait underlying the data. It differs from reliability, internal consistency, and homogeneity. Reliability is the ratio of true score variance to observed score variance. Internal consistency estimates reliability based on item variances and covariances. Homogeneity refers to item correlations but is often used interchangeably with unidimensionality. Unidimensionality is crucial for measurement theory, as it assumes items measure one common thing. Despite its importance, there is no accepted index for unidimensionality. Hattie clarifies that unidimensionality is not the same as local independence, which is a key assumption in latent trait models. He reviews various indices, including those based on answer patterns, reliability, principal components, factor analysis, and latent traits. Indices based on latent traits, such as those using residuals from a three-parameter logistic model, are suggested as more useful. Hattie also discusses the limitations of indices like alpha, which can overestimate unidimensionality. He concludes that unidimensionality is best assessed using latent trait models, where local independence holds and a single latent trait explains item responses.John Hattie reviews methods for assessing unidimensionality of tests and items. He discusses indices based on answer patterns, reliability, components, factor analysis, and latent traits. Many indices lack a clear rationale and are adjustments of previous indices. He suggests that indices based on residuals after fitting a two- or three-parameter latent trait model may be most useful for detecting unidimensionality. Unidimensionality is defined as the existence of one latent trait underlying the data. It differs from reliability, internal consistency, and homogeneity. Reliability is the ratio of true score variance to observed score variance. Internal consistency estimates reliability based on item variances and covariances. Homogeneity refers to item correlations but is often used interchangeably with unidimensionality. Unidimensionality is crucial for measurement theory, as it assumes items measure one common thing. Despite its importance, there is no accepted index for unidimensionality. Hattie clarifies that unidimensionality is not the same as local independence, which is a key assumption in latent trait models. He reviews various indices, including those based on answer patterns, reliability, principal components, factor analysis, and latent traits. Indices based on latent traits, such as those using residuals from a three-parameter logistic model, are suggested as more useful. Hattie also discusses the limitations of indices like alpha, which can overestimate unidimensionality. He concludes that unidimensionality is best assessed using latent trait models, where local independence holds and a single latent trait explains item responses.