Statistical constraints on the use of single cases in neuropsychological research

In a recent study by Miceli, Silveri, Romani, and Caramazza (Brain and Language, 1989, 36, 447-492), free speech records for 20 unselected Italian-speaking agrammatic patients were analyzed along a variety of linguistic parameters, with particular emphasis on substitution and omission errors within traditional "part of speech" categories. The authors draw two strong conclusions: (1) the variability observed in this sample is too great to justify grouping these patients (or any coherent subset of the patients) into the "same" clinical category; (2) the extensive variability displayed by these patients reflects meaningful and very specific patterns of dissociation that would be lost if patients were grouped together in a common category called "aggrammatism." In a reanalysis of the Miceli et al. data, we conclude that the evidence does not justify either of these conclusions. First, we provide a set of mathematical simulations demonstrating that the dissociations claimed by Miceli et al. could have occurred entirely by chance, if samples of speech were drawn randomly from a "universe of discourse" characterized by error rates similar to those displayed by patients in this study. This randomization method has considerable generality, and could prove useful to other investigators interested in determining the number and type of dissociations that are likely to occur by chance in individual and/or group studies. Second, we demonstrate that there are many confounding factors that could account for the variability observed by Miceli et al., including differences in neurological status, education, and the strategies that patients elect to deal with their limitations.