Roger K. Blashfield

The Methods and Problems of Cluster Analysis

Cluster analysis methods have a long history. The earliest known procedures were suggested by anthropologists (Czekanowski, 1911; Driver and Kroeber, 1932). Later, these ideas were picked up in psychology. For instance, Zubin (1938) proposed a rather simple method for sorting a correlation matrix which would yield clusters. About the same time, Stephenson (1936) suggested the use of inverted factor analysis to find clusters of people.

Neuropsychology and cluster analysis: Potentials and problems

This report presents a selective overview of the cluster analysis literature and its potential uses in neuropsychology. In addition, an actual problem involving data from the Florida Longitudinal Project is presented to provide a practical example of many of the processes and problems involved in cluster analytic techniques. It is hoped that the reader will gain a theoretical and practical understanding of such methods and their potential usefulness in neuropsychology and other related areas.

A Comparison of Cluster Analysis Techniques Withing a Sequential Validation Framework.

This study involved the comparison of 23 different methods of cluster analysis in a four-stage sequential validation design. These stages included derivation, replication, external validation, and cross-validation. Data were obtained from 750 alcohol abusers on a number of socio-behavioral variables. Results demonstrated that the solution given by Wards method of cluster analysis was particularly powerful in comparison to solutions yielded by other techniques.

Performance of a composite as a function of the number of judges

Abstract In decision situations where relevant variables cannot be easily measured, mathematical aggregation of individual judgments may prove to be a useful decision aid. A wide variety of studies indicate that composite judgments formed by equal-weighting aggregation models outperform the average individual judge making up the composite. However, the use of these composite judgments in actual decision making situations has not been suggested because of inefficiencies caused by the need to include the judgments of a large number of individuals in the formation of the composite. If composites formed by pooling judgments of fewer decision makers produce similar incremental performance, these inefficiencies would be substantially reduced, making use of such a technique more practical. This study empirically tested the effects of group size upon the incremental accuracy of an equal-weighted composite judge in three different judgment tasks. The results indicate that on average the majority of the increment gained by aggregating large numbers of judges can be obtained by aggregating three judges.

Suggested guidelines for including or excluding categories in the DSM-IV

A set of guidelines is proposed for the possible inclusion or exclusion of diagnostic categories in the DSM-IV. For possible inclusion, a new category should meet all of the following five guidelines: adequate literature, specified diagnostic criteria, acceptable interclinician reliability, evidence that the criteria forms a syndrome, and differentiation from other categories. For possible exclusion, a category should possess an inadequate literature, extremely low coverage, or evidence of diagnostic bias. None of the exclusionary guidelines would be invoked if the category refers to a demonstrable disease.

Metaphorical analysis of psychiatric classification as a psychological test.

In this taxonomic article we explore the metaphor of comparing a psychiatric classification to a psychological test. Structurally, diagnostic criteria are like test items; diagnostic categories are like scales; and classification are like tests. Analytically, the ideas of reliability and validity are the primary concepts invoked in the empirical evaluation of both classifications and tests. However, when the metaphor is explored in more detail, the differences between classifications and tests become clear. These differences are discussed in terms of the structural and analytical relations between tests and classifications. This metaphorical analysis of classifications as tests suggests that certain issues that have been discussed in regard to psychological tests, particularly reliability and validity, may require modification when applied to psychiatric classification.

MMPI scales for DSM‐III personality disorders: A preliminary validation study

A preliminary investigation of the external validity of MMPI scales designed to measure certain DSM-III Personality disorders (N = 108) is reported. Results indicate that many of these scales are effective for discriminating clinically diagnosed personality disorders from control subjects (N = 640).

Exemplar prototypes of personality disorder diagnoses

Abstract The applicability of the prototype model to the DSM-III personality disorders was examined by attempting to isolate prototype cases (both high interclinician agreement and fast diagnostic reaction time). Twenty clinicians assigned diagnoses to 30 cases selected to represent the 11 personality disorders. Prototype cases were located for seven of the personality disorders. Also examined were differences among the clinicians according to profession (psychology/psychiatry) and level of experience (student/faculty). No difference was found according to profession using reaction time or agreement data, but interprofessional diagnostic differences were noted on individual cases. Experience had a significant effect on diagnostic speed, but not on agreement. A multidimensional scaling solution yielded a circumplex ordering of the personality disorders.

Developmental classification of reading-disabled children

The present study developed and used longitudinal cluster analysis, a multivariate classification technique, to classify a sample of 200 nonclinical normal and reading-disabled males based on their performances on a neuropsychological battery at kindergarten, second, and fifth grades. The resulting classification was examined against various internal and external validation criteria. Using a validation framework, five developmental subtypes of children, two normal and three deficit reader groups, were found. Three of these groups could best be described as partitions of a multivariate normal distribution; the other two, both containing deficit readers, showed different covariance structures, suggesting differing developmental patterns. These groups were shown to differ with respect to the domains of academic achievement, parental achievement, neurological status, birth histories, school behaviors, and neuropsychological-cognitive development. The results suggest that developmental classifications can be formed by using multivariate classification methods and the subtypes support findings from cross-sectional classification research on similar populations.

The classification of depression through cluster analysis

Summary Cluster analysis research has the potential of clarifying the confusion currently extant in psychiatric classification. The paper reviews 11 studies that have used cluster analysis to create new classifications of depression. These studies consistently found a category of retarded/psychotic/endogenous depression. The validation studies by Raskin, Overall, Paykel, and their associates demonstrate the predictive validity of cluster generated classifications. Despite these favorable results, the cluster-analysis studies have had little impact on the literature about depression. The reasons for this lack of impact seem to be: (1) the unsolved statistical problems associated with cluster analysis, (2) the naive empiricism inherent in most of this research, and (3) the failure to be aware of the political nature of classificatory change.