APPLICATION OF LATENT STRUCTURE ANALYSIS TO THE REITAN-INDIANA APHASIA SCREENING TEST.

Abstract

The Reitan-Indiana Aphasia Screening Test was examined using a latent structure analysis which involved the assumptions that a latent variable X (brain function), could explain the relationships among the manifest variables A, B, C, D (categories of responses to the Aphasia Test). It was further assumed that within the latent variable a four-class model would be the preferred model. And, it was assumed that persons clinically assigned to the latent classes would be similarly assigned by their test responses. The model was represented by mathematical equations which express the probability of a given response pattern in a contingency table in terms of the joint probabilities of each latent class and the response pattern. Chi-square values were obtained for the model of independence, for the four-class model, for the three-class model and for two two-class models. Hierarchical subtractions were made which resulted in a preferred model, leaving a Chi-square value to be referred to the Chi-square distribution to determine whether or not the latent class model offered an improvement of fit to the data. Twelve comparisons were made using different item combinations. All 12 showed the four-class model offered a significant improvement over the model of independence and either of two two-class models tested. Eleven of the twelve comparisons showed that a four-class model provided an adequate fit to the data. In seven of those instances the four-class model was accepted as the preferred model. In the remaining four instances the heirarchical subtractions showed that a four-class model did not offer significant improvement over the three-class model, so the three-class model was accepted as the preferred model due to the criterion of parsimony. The study showed that there are four categories of brain function which can be identified by means of a behavioral test. It also gave the exact coefficient of agreement between assignment of individuals into those categories by clinical and test classification. It was concluded that latent structure analysis is an effective technique to describe brain-behavior relationships. It was further suggested that consideration be given to the use of latent trait statistics to continue to refine the Aphasia Test without compromising the extensive results already achieved.