Correspondence analysis in the study of lithogeochemical data: General strategy and the usefulness of various data-coding schemes

Abstract

Descriptive factor analysis methods are useful tools for the study of lithogeochemical data, because they are designed to extract non-redundant variation patterns from large data tables. Correspondence analysis is used here because it treats samples (rows) and variables (columns) in a symmetrical manner, and offers good flexibility in data coding. The strategy used in the study of lithogeochemical data is based on the recognition that the behavior of major elements (“framework” elements) and trace elements (“substitution” elements) is determined by different parameters. Major-element behavior is directly related to mineral stability, whereas trace-element behavior is dependent upon the interaction between the trace elements and existing host-minerals. The lithogeochemical data are studied in several stages. Firstly, the factors obtained for major elements are interpreted in terms of the mineralogy and petrography of the samples. Secondly, the factors obtained for trace elements are correlated to the major element factor space and the trace element behavior is interpreted in this framework. Using this approach, the correlations and discrepancies (non-correlations) between major and trace elements are evaluated more objectively than when all elements are treated together; discrepancies in particular, provide information that is unique to trace elements. Thirdly, several coding schemes are applied to the data in order to enhance the underlying variation patterns in different ways. The use of rank coding (ordinal, reduced coding) follows the study of the raw analytical data; it increases the robustness of the factors and the homogeneity of the variation patterns, particularly for trace elements; it is also an intermediate step before the use of disjunctive coding. Disjunctive coding (binary coding) gives access to another level of information by permitting the study of non-linear patterns of element behavior. In this paper, the usefulness of such an approach is illustrated by using lithogeochemical data from the Helikian Athabasca Basin of Saskatchewan, which is the host to unconformity-type U deposits.