ecdf {Hmisc} | R Documentation |
Computes coordinates of cumulative distribution function of x, and by defaults
plots it as a step function. A grouping variable may be specified so that
stratified estimates are computed and (by default) plotted. If there is
more than one group, the labcurve
function is used (by default) to label
the multiple step functions or to draw a legend defining line types, colors,
or symbols by linking them with group labels. A weights
vector may
be specified to get weighted estimates. Specify normwt
to make
weights
sum to the length of x
(after removing NAs). Other wise
the total sample size is taken to be the sum of the weights.
ecdf
is actually a method, and ecdf.default
is what's
called for a vector argument. ecdf.data.frame
is called when the
first argument is a data frame. This function can automatically set up
a matrix of ECDFs and wait for a mouse click if the matrix requires more
than one page. Categorical variables, character variables, and
variables having fewer than a set number of unique values are ignored.
If par(mfrow=..)
is not set up before ecdf.data.frame
is
called, the function will try to figure the best layout depending on the
number of variables in the data frame. Upon return the original
mfrow
is left intact.
When the first argument to ecdf
is a formula, a Trellis/Lattice function
ecdf.formula
is called. This allows for multi-panel
conditioning, superposition using a groups
variable, and other
Trellis features, along with the ability to easily plot transformed
ECDFs using the fun
argument. For example, if fun=qnorm
,
the inverse normal transformation will be used for the y-axis. If the
transformed curves are linear this indicates normality. Like the
xYplot
function, ecdf
will create a function Key
if
the groups
variable is used. This function can be invoked by the
user to define the keys for the groups.
ecdf(x, ...) ## Default S3 method: ecdf(x, what=c('F','1-F','f'), weights, normwt=FALSE, xlab, ylab, q, pl=TRUE, add=FALSE, lty=1, col=1, group=rep(1,length(x)), label.curves=TRUE, xlim, subtitles=TRUE, datadensity=c('none','rug','hist','density'), side=1, frac=switch(datadensity,none=NA,rug=.03,hist=.1,density=.1), dens.opts=NULL, lwd, ...) ## S3 method for class 'data.frame': ecdf(x, group=rep(1,nrows), weights, normwt, label.curves=TRUE, n.unique=10, na.big=FALSE, subtitles=TRUE, vnames=c('labels','names'),...) ## S3 method for class 'formula': ecdf(x, data, groups, prepanel=prepanel.ecdf, panel=panel.ecdf, ..., xlab, ylab, fun=function(x)x, subset=TRUE)
x |
a numeric vector, data frame, or Trellis/Lattice formula |
what |
The default is "F" which results in plotting the fraction of values
<= x. Set to "1-F" to plot the fraction > x or "f" to plot the
cumulative frequency of values <= x.
|
weights |
numeric vector of weights. Omit or specify a zero-length vector or NULL to get unweighted estimates. |
normwt |
see above |
xlab |
x-axis label. Default is label(x) or name of calling argument. For
ecdf.formula , xlab defaults to the label attribute
of the x-axis variable.
|
ylab |
y-axis label. Default is "Proportion <= x" , "Proportion > x" ,
or "Frequency <= x" depending on value of what .
|
q |
a vector for quantiles for which to draw reference lines on the plot. Default is not to draw any. |
pl |
set to F to omit the plot, to just return estimates. |
add |
set toTRUE to add the cdf to an existing plot. |
lty |
integer line type for plot. If group is specified, this can be a vector.
|
lwd |
line width for plot. Can be a vector corresponding to group s.
|
col |
color for step function. Can be a vector. |
group |
a numeric, character, or factor categorical variable used for stratifying
estimates. If group is present, as many ECDFs are drawn as there are
non–missing group levels.
|
label.curves |
applies if more than one group exists.
Default is TRUE to use labcurve to label curves where they are farthest
apart. Set label.curves to a list to specify options to
labcurve , e.g., label.curves=list(method="arrow", cex=.8) .
These option names may be abbreviated in the usual way arguments
are abbreviated. Use for example label.curves=list(keys=1:5)
to draw symbols periodically (as in pch=1:5 - see points )
on the curves and automatically position a legend
in the most empty part of the plot. Set label.curves=FALSE to
suppress drawing curve labels. The col , lty , and type
parameters are automatically passed to labcurve , although you
can override them here. You can set label.curves=list(keys="lines") to
have different line types defined in an automatically positioned key.
|
xlim |
x-axis limits. Default is entire range of x .
|
subtitles |
set to FALSE to suppress putting a subtitle at the bottom left of each
plot. The subtitle indicates the numbers of
non-missing and missing observations, which are labeled n , m .
|
datadensity |
If datadensity is not "none" , either scat1d or histSpike is called to
add a rug plot (datadensity="rug" ), spike histogram
(datadensity="hist" ), or smooth density estimate ("density" ) to
the bottom or top of the ECDF.
|
side |
If datadensity is not "none" , the default is to place the additional
information on top of the x-axis (side=1 ). Use side=3 to place at
the top of the graph.
|
frac |
passed to histSpike
|
dens.opts |
a list of optional arguments for histSpike
|
... |
other parameters passed to plot if add=F. For data frames, other
parameters to pass to ecdf.default .
For ecdf.formula , if groups is not used, you can also add
data density information to each panel's ECDF by specifying the
datadensity and optional frac , side ,
dens.opts arguments.
|
n.unique |
minimum number of unique values before an ECDF is drawn for a variable in a data frame. Default is 10. |
na.big |
set to TRUE to draw the number of NAs in larger letters in the middle of
the plot for ecdf.data.frame
|
vnames |
By default, variable labels are used to label x-axes. Set vnames="names"
to instead use variable names.
|
method |
method for computing the empirical cumulative distribution. See
wtd.ecdf . The default is to use the standard "i/n" method as is
used by the non-Trellis versions of ecdf .
|
fun |
a function to transform the cumulative proportions, for the
Trellis-type usage of ecdf
|
data |
|
groups |
|
subset |
|
prepanel |
|
panel |
the usual Trellis/Lattice parameters, with groups
causing ecdf.formula to overlay multiple ECDFs on one panel. |
for ecdf.default
an invisible list with elements x and y giving the
coordinates of the cdf. If there is more than one group
, a list of
such lists is returned. An attribute, N
, is in the returned
object. It contains the elements n
and m
, the number of
non-missing and missing observations, respectively.
plots
Frank Harrell
Department of Biostatistics, Vanderbilt University
f.harrell@vanderbilt.edu
wtd.ecdf
, label
, table
, cumsum
, labcurve
, xYplot
, histSpike
set.seed(1) ch <- rnorm(1000, 200, 40) ecdf(ch, xlab="Serum Cholesterol") scat1d(ch) # add rug plot histSpike(ch, add=TRUE, frac=.15) # add spike histogram # Better: add a data density display automatically: ecdf(ch, datadensity='density') label(ch) <- "Serum Cholesterol" ecdf(ch) other.ch <- rnorm(500, 220, 20) ecdf(other.ch,add=TRUE,lty=2) sex <- factor(sample(c('female','male'), 1000, TRUE)) ecdf(ch, q=c(.25,.5,.75)) # show quartiles ecdf(ch, group=sex, label.curves=list(method='arrow')) # Example showing how to draw multiple ECDFs from paired data pre.test <- rnorm(100,50,10) post.test <- rnorm(100,55,10) x <- c(pre.test, post.test) g <- c(rep('Pre',length(pre.test)),rep('Post',length(post.test))) ecdf(x, group=g, xlab='Test Results', label.curves=list(keys=1:2)) # keys=1:2 causes symbols to be drawn periodically on top of curves # Draw a matrix of ECDFs for a data frame m <- data.frame(pre.test, post.test, sex=sample(c('male','female'),100,TRUE)) ecdf(m, group=m$sex, datadensity='rug') freqs <- sample(1:10, 1000, TRUE) ecdf(ch, weights=freqs) # weighted estimates # Trellis/Lattice examples: region <- factor(sample(c('Europe','USA','Australia'),100,TRUE)) year <- factor(sample(2001:2002,1000,TRUE)) ecdf(~ch | region*year, groups=sex) Key() # draw a key for sex at the default location # Key(locator(1)) # user-specified positioning of key age <- rnorm(1000, 50, 10) ecdf(~ch | equal.count(age), groups=sex) # use overlapping shingles ecdf(~ch | sex, datadensity='hist', side=3) # add spike histogram at top