Pleiad {shipunov}

Correlation circles (correlation pleiads)

Description

Plot correlation circles (correlation pleiads, correlograms)

Usage

Pleiad(tbl, abs=FALSE, corr=FALSE, dist=FALSE, treshold=FALSE,
 circ=list(1, 1, 1), breaks=5, auto=TRUE, gr=6, lwd=NULL, lty=NULL,
 lcol=NULL,  abbr=-1, lbltext="internal", lblcex=1, off=1.09, hofft=0.07,
 hoff=1.02, legend=TRUE,  legtext=1, legpos="topright", leghoriz=FALSE,
 show.int=FALSE, dig.lab=1, neg.col=NULL, ...)

Arguments

tbl

Data: square, numeric, symmetric matrix with same row and column names

abs

If TRUE, uses absolute values instead of real

corr

If TRUE, uses absolute values instead of real and cuts from 0 to 1, this is good for correlation matrices

dist

If TRUE, converts distance matrix to the data frame – good for "dist" objects

treshold

If this is (saying) =.5, selects for plotting (with lty=1) only those values which are >.5

circ

Line type, width and color for the cirle; if first or third =0, no cicrle

breaks

How to cut() values, if "cramer", then =c(0, .1, .3, .5, 1)

auto

If FALSE, specify 'lwd', 'lty' and 'lcol'

gr

Grayscale scheme starts from 6 breaks

lwd

If auto=FALSE, specify here the vector concerted with breaks

lty

If auto=FALSE, specify here the vector concerted with breaks

lcol

If auto=FALSE, specify here the vector concerted with breaks; if length(lcol) == 1, all lines are of particular color

abbr

If =-1, no abbreviation; if =0, no labels; other values run abbreviate(..., abbr)

lbltext

If this is a vector starting from something else, will replace dimnames

lblcex

Magnification of labels

off

Radial offset of labels, be careful!

hofft

Treshold determining which labels are rigtmost/leftmost, 'hofft=0' put all labels into this group

hoff

Horizontal offset for rightmost/leftmost labels; 'hoff=1' removes offset

legend

If FALSE, no legend

legtext

If =1 then "weaker ... stronger"; if =2, shows cutting intervals; if =3, then 1:5; if >3, issues error

legpos

Position of the legend, see help(legend)

leghoriz

Make the legend horizontal?

show.int

Show intervals in (...] form

dig.lab

dig.lab for cut(), use to change notation

neg.col

If not NULL and 'abs' or 'corr' are TRUE, colorize negative correlations using specified color

...

Further arguments to points()

Details

Correlation circles (correlation pleiads, correlograms) based on the works of Petr Terentjev's (Saint-Petersburg) school.

Please be sure to use 'corr=TRUE' or 'dist=TRUE' when working with correlation matrices or 'dist' objects, respectively.

It is probably a good idea to order data entries with hierarchical clustering to optimize the resulted graph (see examples).

Note that: (1) 'lty', 'lwd' and 'lcol' are not recycling; (2) plot has no margins so consider second 'legend()' to add any text to the plot (see examples); (3) it works best when number of items is relatively low (< 30); (4) it can visualize (colorize) negative correlations (see examples) but this works bes with default (black) color; (5) 'dist=TRUE' will convert dissimilarities into similarities by substracting from maximum.

Alternatives: those graph plotting packages which are able to plot "correlogram".

Value

Data frame (invisibly) with position of points, might help in plot enchancing.

Author(s)

Alexey Shipunov

Examples

l.c <- cor(datasets::longley, method="spearman", use="pairwise")
Pleiad(l.c, corr=TRUE, legtext=2, pch=21, cex=2, bg="white", breaks=3,
 gr=3, hoff=1, show.int=TRUE)
legend("topleft", legend="Macroeconomic correlations", text.font=2,
bty="n")

## colorize negative correlations and use hclust() to re-order
reorder <- hclust(dist(t(longley)))$order
Pleiad(l.c[reorder, reorder], corr=TRUE, neg.col="red")

dr.c <- cor(drosera[, -1], method="spearman", use="pairwise")
## simple example with most defaults
Pleiad(dr.c, corr=TRUE)
## complex example with user-specified colors etc.
Pleiad(dr.c, corr=TRUE, legtext=2, pch=19, cex=1.2, hoff=1.06,
 auto=FALSE, lwd=c(1, 1, 1, 2.5, 4), lty=rep(1, 5),
 lcol=colorRampPalette(c("grey", "#D8284F"))(5),
 circ=c(2, 1, 1))

## visualize distances
Pleiad(dist(t(atmospheres)), dist=TRUE, breaks=3, legtext=2, dig.lab=3)