Graphic Visualization of Correlation and Similarity

摘要

Evaluation of the similarity between samples, locations, or variables isfrequently an important part of data analysis, from initial data exploration through multivariatedata analyses. These evaluations include, for example, covariance among concentrationsof sediment chemicals at a site, correlations between chemical and biologicalmeasures, and (multivariate) similarity measures between sampling locations. The mostreadily constructed means of displaying the results of these evaluations is in tabularformat, as exemplified by a correlation matrix. However, graphic representations of thesame information can facilitate better understanding and communication of the relationshipsinherent in a data set. By linking together several special-purpose software tools,graphic visualizations of correlation and similarity can be produced easily enough thatthey can be routinely incorporated into initial data explorations. In addition to illustratingpurely statistical relationships, graphic representations of similarity can also incorporateexplicit spatial relationships. The resulting superposition of spatial and statistical relationshipswithin a data set creates an easily understood visual display of potentially complexor otherwise imperceptible relationships. Several approaches to automating theproduction of correlation and similarity graphics are illustrated here, using readily availableopen-source and custom software.