Modelling picture naming in aphasia: The relationship between consistency and predictability of responses.

Objective: Two aspects of aphasic picture naming were examined: response consistency. that is. the extent to which the accuracy of the response to the same stimulus is replicated in a successive examination, and response predictability, that is, the extent to which accuracy depends on the characteristics of each stimulus. Methods: Thirty-eight aphasic participants were examined twice. The response pattern was the same across the 2 presentations (response stability) for 36 participants, who were classified into 3 groups according to the prevailing error-type (lexical-semantic, phonological, or a balance between the two error-types): Their item-consistency was quantified with Cohen's kappa. In each case the roles played by lexical frequency, precocity of acquisition and length of the target word, and visual complexity and image agreement of the stimulus picture were examined; the ability to predict response accuracy of a model simultaneously including these 5 variables was quantified by means of the McFadden index. Finally, the relationship between predictability (McFadden index) and consistency (Cohen's kappa) was analyzed. Results: For 34 of 36 participants, consistency was higher than chance. Consistency was directly correlated to the prevalence of lexical-semantic errors. On regression analysis. the relationship between consistency and predictability was significant. Conclusions: Response consistency reflects the existence of a clear difficulty gradient within the items of a battery. The significant relationship between consistency and error type suggests that, in principle, lexical-semantic errors might be more predictable than phonological errors based on the characteristics of each stimulus.