Friday, 1 February 2008
Graphical representation of large correlation matrices
I have a large multivariate dataset and need to display the correlation matrix of all variables in a publication. The traditional way to do this is to simply print the naked numbers, often just the upper or lower triangular matrix. D. J. Murdoch and E. D. Chow presented a much nicer way of displaying large correlation matrices by means of ellipses. The ellipse shape is given by the corresponding correlation coefficient and a bivariate normal distribution with unit variance. This results in a very intuitive graph that allows to easily spot extreme values and the direction of the correlation. By looking at such a graph, the size and direction of correlations can easily be spotted.
Duncon Murdoch maintains an R package called ellipse that includes the function "plotcorr" to produce such graphs. The example figure below was produced with a version of "plotcorr" that has been adapted to lattice.
Posted at 09:47 by Thomas in R | Comments[0]