ISSN:
1573-8868
Keywords:
classification
;
cluster analysis
;
numerical taxonomy
Source:
Springer Online Journal Archives 1860-2000
Topics:
Geosciences
,
Mathematics
Notes:
Abstract Coefficients of association have been widely employed in cluster analysis. However, their use has been, for the most part, restricted to binary data. This limitation can be overcome by redefining positive and negative matches and mismatches in terms of minimum and maximum values of paired elements of parallel vector arrays. Rewriting the algorithms of coefficients of association with these new components gives the new “quantified” coefficients general utility for binary, ordered multistate, and quantitative data, while retaining their original analytic properties. Quantified coefficients of association avoid several problems of shape and size that are associated with correlation coefficients and measures of Euclidean distance. However, when measuring similarity, quantified coefficients weight each attribute of an object by that attribute's magnitude. A related set of similarity indices termed “mean ratios” is introduced; these indices give each attribute equal weight in all situations. Both quantified coefficients of association and mean ratios are related to a number of measures of similarity introduced to various fields of scientific research during the past 50 years. A review of this literature is included in an attempt to consolidate methodology and simplify nomenclature.
Type of Medium:
Electronic Resource
URL:
http://dx.doi.org/10.1007/BF02080152