Use of General Linear Models to Improve Regression-Based Norming of Cognitive
Assessment Tests 
   Lead Investigator:    William Goette
   Institution      :    University of Texas Southwestern
   E-Mail           :    William.Goette@UTSouthwestern.edu
   Proposal ID      :    1433



   Proposal Description:
   Background: Regression-based norms are helpful for clinical applications  because they are flexible and easy to use, but when these are used for           predicting raw scores, ordinary least squares regression can be problematic.     Count data (i.e., the number of correct or incorrect responses) is discrete and  often skewed, making traditional linear regression problematic.                   Goal: Identify whether traditional ordinary least squares regressions are         appropriate for predicting raw neuropsychological test scores.                   Objectives:                                                                       1 - Examine fit statistics and assumption checks of ordinary least squares       regression in raw data for admissibility of these models.                         2 - Compute general linear model regressions (e.g., binomial, Poisson, negative  binomial) and re-examine predictive accuracy compared to OLS regressions.         3 - Obtain standardized discrepancy scores from obtained test data using the     extant NACC norms and the general linear regression models to examine             diagnostic accuracy of both in the NACC data and an external dataset.