Abstract

Approaching the measurement of psychological constructs from an item analysis tradition (e.g., item response theory (IRT)) often reveals the inadequacy of single-factor or simple structure models in describing complex psychological phenomena. When test analysts closely evaluate the interrelationships among a collection of item responses it is not uncommon to find that a more complex measurement model is needed. For example, consider the seemingly well-known mental health construct of depression. Though responses to scales measuring depression are routinely treated as though there is only a single latent trait accounting for their covariance, item analysis often uncovers the presence of a single general dimension common to all the items, but additional specific dimensions that account for the uniqueness of content clusters (e.g., somatic symptoms; Irwin et al., 2010). In these modeling scenarios, some type of general hierarchical IRT model (e.g., bifactor or two-tier) may be appropriate.