Background: In a previous study, discriminant function analyses (DFA) were employed to determine the accuracy of various story narrative and conversational discourse measures in classifying non-brain-injured (NBI) and closed-head-injured (CHI) adults (Coelho, Youse, Le, & Feinn, 2003). The DFAs correctly predicted group membership with 70–81% accuracy.Aims: The present study re-examined the performance of the CHI and NBI participants who were incorrectly classified in an effort to determine what aspects of their discourse performance contributed to the misclassifications. It was hypothesised that the misclassifications were due to the relatively broad range in performance on the discourse measures, resulting in considerable overlap between the NBI and CHI participants.Methods & Procedures: Scores for the story narrative and conversational discourse measures that made the largest contribution to the correct classification of the two participant groups were re-examined for the CHI and NBI participants who were misclassified by the DFA in the previous study (Coelho et al., 2003).Outcomes & Results: Results indicated that there was considerable overlap in the discourse performance of the two participant groups for several of the story narrative and conversational discourse measures.Conclusions: The performance overlaps occurred on many of the same discourse measures that were noted to be fairly good discriminators of CHI versus NBI discourse performance in the original study. Consequently, recommendations regarding elimination of certain measures to streamline the discourse analysis procedure could not be made. Other factors such as sampling discourse acontextually and specific participant characteristics undoubtedly influenced these findings as well. In addition, the DFA procedure utilised in the original study did not take into account the heterogeneity of discourse data. Nonparametric procedures such as classification and regression trees (CART) (Breiman, Friedman, Olshen, & Stone, 1984; Johnson & Wichern, 2002) may be better suited for the classification of non-homogeneous populations such as individuals with CHI.
Read full abstract