plsmo: Plot smoothed estimates

Description

Plot smoothed estimates of x vs. y, handling missing data for lowess or supsmu, and adding axis labels. Optionally suppresses plotting extrapolated estimates. An optional group variable can be specified to compute and plot the smooth curves by levels of group. When group is present, the datadensity option will draw tick marks showing the location of the raw x-values, separately for each curve. plsmo has an option to plot connected points for raw data, with no smoothing. The non-panel version of plsmo allows y to be a matrix, for which smoothing is done separately over its columns. If both group and multi-column y are used, the number of curves plotted is the product of the number of groups and the number of y columns.

method='intervals' is often used when y is binary, as it may be tricky to specify a reasonable smoothing parameter to lowess or supsmu in this case. The 'intervals' method uses the cut2 function to form intervals of x containing a target of mobs observations. For each interval the ifun function summarizes y, with the default being the mean (proportions for binary y). The results are plotted as step functions, with vertical discontinuities drawn with a saturation of 0.15 of the original color. A plus sign is drawn at the mean x within each interval. For this approach, the default x-range is the entire raw data range, and trim and evaluate are ignored. For panel.plsmo it is best to specify type='l' when using 'intervals'.

panel.plsmo is a panel function for trellis for the xyplot function that uses plsmo and its options to draw one or more nonparametric function estimates on each panel. This has advantages over using xyplot with panel.xyplot and panel.loess: (1) by default it will invoke labcurve to label the curves where they are most separated, (2) the datadensity option will put rug plots on each curve (instead of a single rug plot at the bottom of the graph), and (3) when panel.plsmo invokes plsmo it can use the "super smoother" (supsmu function) instead of lowess, or pass method='intervals'. panel.plsmo senses when a group variable is specified to xyplot so that it can invoke panel.superpose instead of panel.xyplot. Using panel.plsmo through trellis has some advantages over calling plsmo directly in that conditioning variables are allowed and trellis uses nicer fonts etc.

When a group variable was used, panel.plsmo creates a function Key in the session frame that the user can invoke to draw a key for individual data point symbols used for the groups. By default, the key is positioned at the upper right corner of the graph. If Key(locator(1)) is specified, the key will appear so that its upper left corner is at the coordinates of the mouse click.

For ggplot2 graphics the counterparts are stat_plsmo and histSpikeg.

Usage

plsmo(x, y, method=c("lowess","supsmu","raw","intervals"), xlab, ylab, 
      add=FALSE, lty=1 : lc, col=par("col"), lwd=par("lwd"),
      iter=if(length(unique(y))>2) 3 else 0, bass=0, f=2/3, mobs=30, trim, 
      fun, ifun=mean, group, prefix, xlim, ylim, 
      label.curves=TRUE, datadensity=FALSE, scat1d.opts=NULL,
      lines.=TRUE, subset=TRUE,
      grid=FALSE, evaluate=NULL, …)

#To use panel function:
#xyplot(formula=y ~ x | conditioningvars, groups,
#       panel=panel.plsmo, type='b', 
#       label.curves=TRUE,
#       lwd = superpose.line$lwd, 
#       lty = superpose.line$lty, 
#       pch = superpose.symbol$pch, 
#       cex = superpose.symbol$cex, 
#       font = superpose.symbol$font, 
#       col = NULL, scat1d.opts=NULL, \dots)

Arguments

vector of x-values, NAs allowed

vector or matrix of y-values, NAs allowed

method

"lowess" (the default), "supsmu", "raw" to not smooth at all, or "intervals" to use intervals (see above)

xlab

x-axis label iff add=F. Defaults of label(x) or argument name.

ylab

y-axis label, like xlab.

add

Set to T to call lines instead of plot. Assumes axes already labeled.

lty

line type, default=1,2,3,…, corresponding to columns of y and group combinations

col

color for each curve, corresponding to group. Default is current par("col").

lwd

vector of line widths for the curves, corresponding to group. Default is current par("lwd"). lwd can also be specified as an element of label.curves if label.curves is a list.

iter

iter parameter if method="lowess", default=0 if y is binary, and 3 otherwise.

bass

bass parameter if method="supsmu", default=0.

passed to the lowess function, for method="lowess"

mobs

for method='intervals', the target number of observations per interval

trim

only plots smoothed estimates between trim and 1-trim quantiles of x. Default is to use 10th smallest to 10th largest x in the group if the number of observations in the group exceeds 200 (0 otherwise). Specify trim=0 to plot over entire range.

fun

after computing the smoothed estimates, if fun is given the y-values are transformed by fun()

ifun

a summary statistic function to apply to the y-variable for method='intervals'. Default is mean.

group

a variable, either a factor vector or one that will be converted to factor by plsmo, that is used to stratify the data so that separate smooths may be computed

prefix

a character string to appear in group of group labels. The presence of prefix ensures that labcurve will be called even when add=TRUE.

xlim

a vector of 2 x-axis limits. Default is observed range.

ylim

a vector of 2 y-axis limits. Default is observed range.

label.curves

set to FALSE to prevent labcurve from being called to label multiple curves corresponding to groups. Set to a list to pass options to labcurve. lty and col are passed to labcurve automatically.

datadensity

set to TRUE to draw tick marks on each curve, using x-coordinates of the raw data x values. This is done using scat1d.

scat1d.opts

a list of options to hand to scat1d

lines.

set to FALSE to suppress smoothed curves from being drawn. This can make sense if datadensity=TRUE.

subset

a logical or integer vector specifying a subset to use for processing, with respect too all variables being analyzed

grid

set to TRUE if the R grid package drew the current plot

evaluate

number of points to keep from smoother. If specified, an equally-spaced grid of evaluate x values will be obtained from the smoother using linear interpolation. This will keep from plotting an enormous number of points if the dataset contains a very large number of unique x values.

…

optional arguments that are passed to scat1d, or optional parameters to pass to plsmo from panel.plsmo. See optional arguments for plsmo above.

type

set to p to have panel.plsmo plot points (and not call plsmo), l to call plsmo and not plot points, or use the default b to plot both.

pch,cex,font

vectors of graphical parameters corresponding to the groups (scalars if group is absent). By default, the parameters set up by trellis will be used.

Value

plsmo returns a list of curves (x and y coordinates) that was passed to labcurve

Side Effects

plots, and panel.plsmo creates the Key function in the session frame.

Examples

Run this code

# NOT RUN {
set.seed(1)
x <- 1:100
y <- x + runif(100, -10, 10)
plsmo(x, y, "supsmu", xlab="Time of Entry") 
#Use label(y) or "y" for ylab


plsmo(x, y, add=TRUE, lty=2)
#Add lowess smooth to existing plot, with different line type


age <- rnorm(500, 50, 15)
survival.time <- rexp(500)
sex <- sample(c('female','male'), 500, TRUE)
race <- sample(c('black','non-black'), 500, TRUE)
plsmo(age, survival.time < 1, fun=qlogis, group=sex) # plot logit by sex

#Bivariate Y
sbp <- 120 + (age - 50)/10 + rnorm(500, 0, 8) + 5 * (sex == 'male')
dbp <-  80 + (age - 50)/10 + rnorm(500, 0, 8) - 5 * (sex == 'male')
Y <- cbind(sbp, dbp)
plsmo(age, Y)
plsmo(age, Y, group=sex)


#Plot points and smooth trend line using trellis 
# (add type='l' to suppress points or type='p' to suppress trend lines)
require(lattice)
xyplot(survival.time ~ age, panel=panel.plsmo)


#Do this for multiple panels
xyplot(survival.time ~ age | sex, panel=panel.plsmo)

#Repeat this using equal sample size intervals (n=25 each) summarized by
#the median, then a proportion (mean of binary y)
xyplot(survival.time ~ age | sex, panel=panel.plsmo, type='l',
       method='intervals', mobs=25, ifun=median)
ybinary <- ifelse(runif(length(sex)) < 0.5, 1, 0)
xyplot(ybinary ~ age, groups=sex, panel=panel.plsmo, type='l',
       method='intervals', mobs=75, ifun=mean, xlim=c(0, 120))


#Do this for subgroups of points on each panel, show the data
#density on each curve, and draw a key at the default location
xyplot(survival.time ~ age | sex, groups=race, panel=panel.plsmo,
       datadensity=TRUE)
Key()


#Use wloess.noiter to do a fast weighted smooth
plot(x, y)
lines(wtd.loess.noiter(x, y))
lines(wtd.loess.noiter(x, y, weights=c(rep(1,50), 100, rep(1,49))), col=2)
points(51, y[51], pch=18)   # show overly weighted point
#Try to duplicate this smooth by replicating 51st observation 100 times
lines(wtd.loess.noiter(c(x,rep(x[51],99)),c(y,rep(y[51],99)),
      type='ordered all'), col=3)
#Note: These two don't agree exactly
# }

Run the code above in your browser using DataLab