When 20 or more measurements and (or) counts are made on 100 or more samples (thin sections, bottom samples, pollen or foram concentrates, heavy minerals, etc.), the resulting table of data is so large that interpretation by eye becomes difficult. In some geologic studies it is desirable to group together similar samples and to measure the degree of similarity between different groups of samples. Several measures of similarity are available: the product-moment correlation coefficient, cosine-theta (Imbrie, 1962), the matching coefficient, and the distance function (Sokal, 1961). The resulting matrix of intercorrelations is still too large for direct interpretation.

Cluster analysis, a technique developed by psychologists, is a method of searching for structure, or relationships, in a matrix of correlation coefficients. Although not so sophisticated as factor analysis, cluster analysis is a useful tool. The results of a cluster analysis can be presented in an hierarchical diagram in two dimensions that will show where the natural breaks occur between groups. A computer program has been written for the IBM 704 that will handle up to 150 measurements on as many as 200 samples. Non-overlapping clusters are used; that is, a sample can appear in only one cluster.

A 12 variable 40-sample problem based on constituent particle composition of Bahamian sediment samples (Imbrie and Purdy, 1962) is used to demonstrate the options of the program. The clear-cut groups in the cluster analysis solution are similar to the facies described by Imbrie and Purdy (1962) based on factor analysis. The cluster can be used as a basis for facies maps.

End_of_Article - Last_Page 540------------