This function provides a convenient interface to the pairs function to produce
enhanced scatterplot matrices, including univariate displays on the diagonal and a variety of fitted lines, smoothers, variance functions, and concentration ellipsoids.
spm is an abbreviation for scatterplotMatrix.
scatterplotMatrix(x, ...)# S3 method for formula
scatterplotMatrix(formula, data=NULL, subset, ...)
# S3 method for default
scatterplotMatrix(x, smooth = TRUE,
id = FALSE, legend = TRUE, regLine = TRUE,
ellipse = FALSE, var.labels = colnames(x), diagonal = TRUE,
plot.points = TRUE, groups = NULL, by.groups = TRUE,
use = c("complete.obs", "pairwise.complete.obs"), col =
carPalette()[-1], pch = 1:n.groups, cex = par("cex"),
cex.axis = par("cex.axis"), cex.labels = NULL,
cex.main = par("cex.main"), row1attop = TRUE, ...)
spm(x, ...)
a data matrix or a numeric data frame.
a one-sided “model” formula, of the form
~ x1 + x2 + ... + xk or ~ x1 + x2 + ... + xk | z where z
evaluates to a factor or other variable to divide the data into groups.
for scatterplotMatrix.formula,
a data frame within which to evaluate the formula.
expression defining a subset of observations.
specifies a nonparametric estimate of the mean or median
function of the vertical axis variable given the
horizontal axis variable and optionally a nonparametric estimate of the spread or variance function. If
smooth=FALSE neither function is drawn. If smooth=TRUE, then both the mean function
and variance funtions are drawn for ungrouped data, and the mean function only is drawn for grouped
data. The default smoother is loessLine, which uses the loess function from
the stats package. This smoother is fast and reliable. See the details below for changing
the smoother, line type, width and color, of the added lines, and adding arguments for the smoother.
controls point identification; if FALSE (the default), no points are identified;
can be a list of named arguments to the showLabels function;
TRUE is equivalent to list(method="mahal", n=2, cex=1, location="lr"),
which identifies the 2 points (in each group, if by.groups=TRUE) with the largest Mahalanobis distances from the center
of the data; list(method="identify") for interactive point identification is not allowed.
controls placement of a legend if the plot is drawn by groups; if FALSE, the legend
is suppressed. Can be a list with the named elementcoords specifying the position of the legend
in any form acceptable to the legend function;
TRUE (the default) is equivalent to list(coords=NULL), for which placement will vary
by the the value of the diagonal argument---e.g., "topright" for diagonal=TRUE.
controls adding a fitted regression line to each plot, or to each group of points
if by.groups=TRUE. If regLine=FALSE, no line
is drawn. This argument can also be a list with named list, with default regLine=TRUE equivalent
to regLine = list(method=lm, lty=1, lwd=2, col=col[1]) specifying the name of the function that
computes the line, with line type 1 (solid) of relative line width 2 and the color equal to the first
value in the argument col. Setting method=MASS::rlm would fit using a robust regression.
controls plotting data-concentration ellipses. If FALSE (the default), no
ellipses are plotted. Can be a list of named values giving levels, a vector of one or more
bivariate-normal probability-contour levels at which to
plot the ellipses; robust, a logical value determing whether to use
the cov.trob function in the MASS package
to calculate the center and covariance matrix for the data ellipses; and fill and fill.alpha,
which control whether the ellipse is filled and the transparency of the fill. TRUE is equivalent
to list(levels=c(.5, .95), robust=TRUE, fill=TRUE, fill.alpha=0.2).
variable labels (for the diagonal of the plot).
contents of the diagonal panels of the plot. If diagonal=TRUE adaptive kernel density
estimates are plotted, separately for each group if grouping is present. diagonal=FALSE suppresses
the diagonal entries. See details below for other choices for the diagonal.
if TRUE the points are plotted in each
off-diagonal panel.
a factor or other variable dividing the data into groups; groups are plotted with different colors and plotting characters.
if TRUE, the default, regression lines and smooths are fit by groups.
if "complete.obs" (the default), cases with missing data are omitted; if "pairwise.complete.obs"), all valid cases are used
in each panel of the plot.
plotting characters for points; default is the plotting characters in
order (see par).
colors for points; the default is carPalette starting at the second color. The color of
the regLine and smooth are the same as for points but can be changed using the the
regLine and smooth arguments.
relative size of plotted points
relative size of axis labels
relative size of labels on the diagonal
relative size of the main title, if any
If TRUE (the default) the first row is at the top, as in a matrix, as
opposed to at the bottom, as in graph (argument suggested by Richard Heiberger).
arguments to pass down.
NULL. This function is used for its side effect: producing
a plot. If point identification is used, a list of identified points is returned.
Many arguments to scatterplotMatrix were changed with verions 3 of car that we hope simplifies use of this function.
The smooth argument is usually either equal to TRUE or FALSE to draw, or omit,
the smoother. Alternatively smooth can equal a list of arguments. The default behavior of
smooth=TRUE is equivalent to smooth=list(smoother=loessLine, spread=TRUE, lty.smooth=1, lwd.smooth=1.5, lty.spread=3, lwd.spread=1), specifying the smoother to be used, including the spread or variance smooth,
and the line widths and types for the curves. You can also specify the colors you want to use for the mean and variance smooths with the arguments col.smooth and col.spread. Alternative smoothers are gamline that uses the
gam function from the mgcv package, and quantregLine that uses quantile regression to
estimate the median and quartile functiona using rqss frm the quantreg package. All of these
smoothers have one or more arguments described on their help pages, and these arguments can be added to the
smooth argument; for example, smooth = list(span=1/2) would use the default
loessLine smoother,
include the variance smooth, and change the value of the smoothing parameter to 1/2. For loessLine
and gamLine the variance smooth is estimated by separately
smoothing the squared positive and negative
residuals from the mean smooth, using the same type of smoother. The displayed curves are equal to
the mean smooth plus the square root of the fit to the positive squared residuals, and the mean fit minus
the square root of the smooth of the negative squared residuals. The lines therefore represent the
comnditional variabiliity at each value on the horizontal axis. Because smoothing is done separately for
positive and negative residuals, the variation shown will generally not be symmetric about the fitted mean
function. For the quantregLine method, the center estimates the median for each value on the
horizontal axis, and the spread estimates the lower and upper quartiles of the estimated conditional
distribution for each value of the horizontal axis.
By default the diagonal argument is used to draw kernel density estimates of the
variables by setting diagonal=TRUE, which is equivalent to setting diagonal =
list(method="adaptiveDensity", bw=bw.nrd0, adjust=1, kernel=dnorm, na.rm=TRUE). The additional argument
shown are descibed at adaptiveKernel. The other methods avaliable, with their default
arguments, are diagonal=list(method="denisty", bw="nrd0", adjust=1, kernel="gaussian", na.rm=TRUE)
that uses density for nonadaptive kernel density estimation; diagonal=list(method
="histogram", breaks="FD")
that uses hist for drawing a histogram that ignores grouping, if present;
diagonal=list(method="boplot") with no additional arguments that draws (parallel) boxplots;
diagonal=list(method="qqplot") with no additional arguments that draws a normal QQplot; and
diagonal=list(method="oned") with no additional arguments that draws a rug plot tilted to the
diagonal, as suggested by Richard Heiberger.
Earlier versions of scatterplotMatrix included arguments transform and famiy to estimate power transformations using the powerTransform function before drawing the plot. The sample functionality can be achieved by calling powerTransform directly to estimate a transformation, saving the transformed variables, and then plotting.
Fox, J. and Weisberg, S. (2019) An R Companion to Applied Regression, Third Edition, Sage.
pairs, scatterplot,
dataEllipse, powerTransform,
bcPower, yjPower, cov.trob,
showLabels, ScatterplotSmoothers.
# NOT RUN {
scatterplotMatrix(~ income + education + prestige | type, data=Duncan)
scatterplotMatrix(~ income + education + prestige | type, data=Duncan,
regLine=FALSE, smooth=list(spread=FALSE))
scatterplotMatrix(~ income + education + prestige,
data=Duncan, id=TRUE, smooth=list(method=gamLine))
# }
Run the code above in your browser using DataCamp Workspace