A test of the predictive validity and a cross-validation of the neurosensory center comprehensive examination for aphasia
Date
1976
Authors
Black, Iryna Lawriw
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The present investigation was primarily concerned with testing the predictive validity of the Neurosensory Center Comprehensive Examination for Aphasia (NCCEA). A total of 206 brain damaged subjects, diagnosed as either aphasic or non-aphasic were employed. 80 subjects were from Victoria, B. C., 38 from New York City, N. Y., and 88 from Iowa City, Iowa. Within the groups of aphasic and non-aphasic, Victoria and Iowa subjects were randomly assigned to Sample 1 or Sample 2, and Sample 5 or Sample 6, respectively. Multivariate analyses of variance indicated that there was a significant difference between the mean level of performance of aphasic and brain damaged non-aphasic subjects on the NCCEA for the following samples: Victoria Sample 1 + Sample 2, Victoria Sample 1, Victoria Sample 2, Iowa Sample 5 + Sample 6, and Iowa Sample 6. In addition, univariate analyses of variance indicated that in every sample, significant differences were observed for almost every subtest of the NCCEA. Data for the New York sample were not subjected to analysis because only two of the subjects in this sample were non-aphasic.
Stepwise discriminant analysis revealed that successful discrimination between aphasics and brain damaged non-aphasics, in terms of the percentage of total cases correctly classified (76.14 % - 90.00%) could be attained by the use of a maximum of five and a minimum of three tests of the NCCEA for both Victoria and Iowa samples. The majority of misclassifications which occurred for these samples were subjects diagnosed as either minimally aphasic, or subjects for whom there was no indication of the severity of language deficits.
The second objective of the present investigation was cross-validation of the predictive equations derived by stepwise discriminant analysis on samples drawn from the same geographical and clinical setting, and samples drawn from different geographical and clinical settings. Cross-validation analyses were performed for every possible combination of samples. The loss in the predictive accuracy of the equations did not exceed the predicted maximum of 20% in any of the cross-validation analyses. However, the percentage of total cases correctly classified was higher for cross-validations within the same geographical and clinical setting, and for cross validations on the New York sample, than for cross-validations between Iowa and Victoria samples.