
Accession Number : AD0717333
Title : Cluster Analysis.
Descriptive Note : Technical rept.,
Corporate Author : YALE UNIV NEW HAVEN CONN DEPT OF STATISTICS
Personal Author(s) : Ling,R. F.
Report Date : JAN 1971
Pagination or Media Count : 201
Abstract : Current clustering techniques possess several common features which seem undesirable. For example, a 'cluster' remains an undefined concept; each clustering technique tends to work properly only under unstated, but often restrictive, implied assumptions; and the nonexistence of clustering statistics or the lack of theory about the sampling distributions of the statistics (when they do exist) makes the assessment of the statistical significance of a cluster quite impossible. In this paper after a brief review and critique of the clustering methods that are most widely used, definitions of a cluster and its related concepts are proposed. The clusters so defined and their associated statistics will remain invariant under any monotonic transformation of the elements of the data matrix on which they depend. Their sampling distributions are investigated by analytic and Monte Carlo methods. Both aritificial and real data are employed to illustrate the methodology and probability theory of the proposed clustering method. (Author)
Descriptors : (*STATISTICAL DATA, CORRELATION TECHNIQUES), PSYCHOLOGICAL TESTS, FACTOR ANALYSIS, MONTE CARLO METHOD, PATTERN RECOGNITION, STATISTICAL DISTRIBUTIONS, PROBABILITY, COMPUTER PROGRAMS, ALGORITHMS, TABLES(DATA)
Subject Categories : Statistics and Probability
Distribution Statement : APPROVED FOR PUBLIC RELEASE