r - How do I interpret the output of corrplot? -
r - How do I interpret the output of corrplot? -
the corrplot
packages provides neat plots , documents how create lots of examples.
but don't understand output. can see if have matrix a_ij
, can plot arrangement of n
n
square tiles, color of tile ij
corresponds value of a_ij
. examples appear have more dimensions:
here can guess color shows correlation coefficient, , orientation of ellipse negative/positive correlation. eccentricity?
the documentation method
says:
the visualization method of correlation matrix used. currently, supports 7 methods, named "circle" (default), "square", "ellipse", "number", "pie", "shade" , "color". see examples details.
the areas of circles or squares show absolute value of corresponding correlation coefficients. method "pie" , "shade" came michael friendly’s job (with adjustment shade added on), , "ellipse" came d.j. murdoch , e.d. chow’s job, see in section references.
so know area, circles , squares, should show coefficient. other dimensions, , other methods?
there 1 dimension shown plot.
michael friendly, in corrgrams: exploratory displays correlation matrices (the corrplot
documentation confusingly refers "job"), says:
in shaded row, each cell shaded bluish or reddish depending on sign of correlation, , intensity of color scaled 0–100% in proportion magnitude of correlation. (such scaled colors computed using rgb coding red, (1, 0, 0), through white (1, 1, 1), bluish (0, 0, 1). simplicity, ignore non-linearities of color reproduction , perception, note these accommodated in color mapping function.) white diagonal lines added direction of correlation may still discerned in black , white. bipolar scale of color chosen leave correlations near 0 empty (white), , create positive , negative values of equal magnitude approximately as intensely shaded. grayness scale , other color schemes implemented in our software (section 6), not illustrated here.
the bar , circular symbols utilize same scaled colors, fill area proportional absolute value of correlation. bars, negative values filled bottom, positive values top. circles filled clockwise positive values, anti-clockwise negative values. the ellipses have eccentricity parametrically scaled correlation value (murdoch , chow, 1996). perceptually, have property of becoming visually less prominent magnitude of correlation increases, in contrast other glyphs.
(emphasis mine)
"murdoch , chow, 1996" publication describing equation drawing ellipses (a graphical display of big correlation matrices). ellipses apparently meant caricatures of bivariate normal distributions:
so in conclusion, only dimension shown correlation coefficient (or value of a_ij
, utilize question's terminology) itself. multiple apparent dimensions redundant.
r plot statistics data-visualization
Comments
Post a Comment