Paul R. Yarnold
Optimal Data Analysis, LLC
An inherent limitation of ordinary least-squares regression models is that dependent measure values near the mean for the sample are predicted well, while values greater or less than the mean are predicted poorly. A UniODA-based methodology for circumventing this limitation and explicitly maximizing classification accuracy (ESS) has been demonstrated for integer dependent variables having ten or fewer levels. This note describes three methods for extending this methodology for problems involving dependent measures having a larger domain of values.