A LATENT-TRAIT INVESTIGATION OF THE LURIA-NEBRASKA NEUROPSYCHOLOGICAL BATTERY (ITEM RESPONSE THEORY, BRAIN DAMAGE, REHABILITATION).

This project represents a descriptive analysis of Forms I and II of the Luria-Nebraska Neuropsychological Battery and an investigation of the applicability of Item Response Theory in neuropsychological assessment. Test protocols of 1280 Form I examinees and 405 Form II examinees were analyzed by item and scale using Item Response Theory. The analysis consisted of investigation of the fit of LNNB data to the one and two-parameter IRT models, analysis of item and scale residuals and information values, comparison of traditional and IRT approaches in derivation of the Pathognomonic, Right Hemisphere and Left Hemisphere scales and comparison of two IRT approaches to the identification of biased items. In general, the one-parameter model did not fit the LNNB data. The two-parameter model, however provided a generally good fit to the data. Scale residuals and information functions indicate that the LNNB scales are unidimensional and accurately measure their underlying pyschological constructs. Item analysis identified several items on each scale that do not contribute to the measurement of the scalar trait. Suggestions were made for deletion, relocation or alteration of these items to improve their measurement properties. Substantial differences were found between Form I and Form II based on item characteristic curves and b-value differences. The nature of these differences suggested that the size and ability distribution of the Form II sample may have prevented accurate parameter estimation, obscuring the comparison of the two forms of the battery. A number of items on the Pathognomonic, right Hemisphere and Left Hemisphere scales were identified that contribute little to the measurement properties of these scales. Additional items not on these scales, were identified that are candidates for inclusion on these scales. Comparison of the statistical tests of b-value differences with ICC differences, for identification of potentially biased items, suggests that the latter method may be more efficacious in the neuropsychological domain. It is concluded that the LNNB is an accurate and content valid test of neuropsychological abilities; that IRT methods could improve the measurement properties of the scales and that ICC differences are an effective approach to item bias detection