Complex Behavioral Indices Weighted by Linear Discriminant Functions for the Prediction of Cerebral Damage

Abstract

Linear discriminant functions were applied to seven dichotomized measures from each of 224 patients who were divided into control, left cerebral damage, right cerebral damage, and diffuse or bilateral damage groups. The measures were: Wechsler Verbal Weighted Score, Wechsler Performance Weighted Score, Halstead Impairment Index, Trail Making Test A, Trail Making Test B, Aphasia 4-rule Prediction, and age of patient. Four comparisons were made, one for each criterion group vs all remaining groups. The discriminant function in each comparison produced a single weighted score per S, and an optimum, least-squares type of separation between the two sets of scores. The resulting distributions of summed, weighted scores in each comparison were inspected for the point of minimum overlap, and an individual's weighted score, falling above or below this point, categorized him as belonging either in the single criterion group or in any one of the remaining three groups. These assignments, when compared with the actual criterion classes of the patients, were expressed as percentages of correct prediction: control vs non-control, 83.0%; left damage vs non-left, 87.5%; right damage vs non-right, 85.7%; diffuse (or bilateral) damage vs non-diffuse, 84.4%. Each measure was examined individually for percentages of correct prediction, but the discriminant function was superior in all instances, being approached only by the Halstead Impairment Index (one comparison) and by the Aphasia 4-rule Prediction (two comparisons). The seven-variable discriminant function was approximately as efficient as either of two previous functions that included more than 20 variables each.