Item- and scale-level analysis of clinical and non-clinical sample responses to the MMPI-2 depression scales employing item response theory

The MMPI-2 was constructed by selecting items that maximally discriminate a target group (e.g., a clinically depressed group) from a control group (e.g., a sample of individuals from a non-clinical population). However, researchers have not had the opportunity to use recent psychometric techniques to examine a large set of responses to the inventory from a population with psychological disorders. In this study, data from populations with different clinical backgrounds is employed to examine how responses to the MMPI-2 differ from responses given by the "normative" sample used to restandardize the MMPI. To evaluate the MMPI scales, item response theory (IRT) is utilized to provide both item analysis and scale analysis. Part of item analysis includes investigations of differential item functioning (DIF) between clinical and non-clinical responses and between female and male responses to items in the MMPI-2. To illustrate the valuable analytical tools of IRT, this dissertation focuses on responses to the MMPI-2 clinical Depression scale and the Harris-Lingoes Depression subscales. For each depression scale, this study employs a methodology that: considers if the IRT assumptions of unidimensionality and local independence of the item responses hold for the measure; fits the IRT two-parameter logistic model to the data, investigates item and scale functioning; explores the possibility of differential item functioning among groups; and makes recommendations for item and scale revisions based on study findings. Study results suggest many of the depression scales to be relatively unidimensional, and the two-parameter logistic IRT model to be a reasonable model for the item responses in the scales. However, multidimensionality exists within a few of the scales (e.g., Harris-Lingoes Psychomotor Retardation subscale), and differential item functioning among study groups exists for a few items within each of the scales. Also, results find several items to be scored in a direction inconsistent with the description of the scale, and recommend reverse scoring or removal of the items from the scale. This dissertation suggests a full-scale IRT analysis and reconstruction of the scales will make the MMPI-2 shorter, more reliable, and provide accurate estimates of respondent trait levels.