The Adequacy of Non-Metric Data in Geology: Tests Using a Divisive-Omnithetic Clustering Technique

Abstract

In many geological investigations precise metric data may be unnecessary because of the relatively imprecise ways in which they are analyzed and interpreted. The hypothesis is tested that many results obtained by numerical analysis of metric data might equally well have come from non-metric (binary) versions of the same data sets. Six multivariate metric data sets representing a diverse suite of potential applications were selected from the literature, and each reduced to binary form by using the mean value of each attribute as its presence threshold. A divisive-omnithetic clustering technique (dissimilarity analysis) was applied to each set to give hierarchical classifications having many theoretically optimal properties. The agreement between the results of this approach and classifications based on the original metric data was reasonable in all cases, in a qualitative and qualitative sense. For most similar studies it is probably entirely adequate to rely on non-metric data, implying a considerable reduction of effort during data collection. [The recent Krithe producta and Pleistocene Poecilozonites cupula were analyzed.].