WAIS-III index score profiles in the Canadian standardization sample

Abstract

Representative index score profiles were examined in the Canadian standardization sample of the Wechsler Adult Intelligence Scale-Third Edition (WAIS-III). The identification of profile patterns was based on the methodology proposed by Lange, Iverson, Senior, and Chelune (2002) that aims to maximize the influence of profile shape and minimize the influence of profile magnitude on the cluster solution. A two-step cluster analysis procedure was used (i.e., hierarchical and k-means analyses). Cluster analysis of the four index scores (i.e., Verbal Comprehension [VCI], Perceptual Organization [POI], Working Memory [WMI], Processing Speed [PSI]) identified six profiles in this sample. Profiles were differentiated by pattern of performance and were primarily characterized as (a) high VCI/POI, low WMI/PSI, (b) low VCI/POI, high WMI/PSI, (c) high PSI, (d) low PSI, (e) high VCI/WMI, low POI/PSI, and (f) low VCI, high POI. These profiles are potentially useful for determining whether a patient's WAIS-III performance is unusual in a normal population.

Notes

Special thanks to The Psychological Corporation, a Harcourt Assessment company, for permission to use the Canadian WAIS-III standardization data; and to Dr. Gordon Chelune for his helpful comments on a previous version of this manuscript.

¹Many statistical stopping rules have been proposed which purport to determine the number of clusters in a data set based on jumps in the fusion coefficients themselves. Milligan and Cooper (1985) examined the validity of 30 statistical stopping rules to extract a predetermined cluster solution in a number of simulation studies. While some stopping rules tended to perform better than others (e.g., Gamma index [Baker & Hubert, 1975], C-index [Hubert & Levin, 1976], Point-Biserial [Milligan, 1980, 1981], Tau index [Rohlf, 1974], G(+) index [Milligan 1981], and the Rand index [Dreger, 1986]), their effectiveness varied depending on the structure of the data set and were not recommended for use (Milligan & Cooper, 1985).

²This study developed a single case methodology for allocating individual cases to clusters. The methodology makes use of (a) Fisher's classification function coefficients, (b) Mahalanobis distance statistics, (c) equations derived from discriminant function and regression analyses, and (d) matrix algebra. This work is currently in preparation for publication.