Iconography of correlations

In exploratory data analysis, the iconography of correlations,^[1][2] or representation of correlations, is a data visualization technique which replaces a numeric correlation matrix by its graphical projection onto a diagram, on which the “remarkable” correlations are plotted as solid lines (positive correlations) or dotted lines (negative correlations); either shorter lengths, or thicker lines, or both, represent greater correlation projection components.

Representation of the proximity of food profiles in Europe.

History

This idea is used in Gaussian graphic models for genome mapping, particularly. But the iconography of correlations is more general, since it does not assume that the data is Gaussian; it only relies on representing the correlation coefficients geometrically.

The iconography of correlations first dates to 1975, applied to marine geochemistry in a 1981 thesis,^{[citation needed]} and later in a 1982 data analysis article.^[3] Afterward, the method was applied widely in the aerospace industry^[4][5] but for about fifteen years manufacturers kept it fairly confidential; generally, they preferred to not broadcast useful techniques to their competitors.

In 1997 the first company was incorporated to distribute iconography of correlations software.^[6] Since then the topic of iconography of correlations has been incorporated into some university courses, and typical topical articles' citation lists have rapidly and greatly expanded, particularly in the fields of medicine^[7] and mass spectrometry.^[8][9][10]

See also

• Bayesian network — a graph in which cause and effect are measured probabilistically instead of geometrically as in correlation iconography.