DescTools (version 0.99.19)

PlotFdist: Frequency Distribution Plot

Description

This function is designed to give a univariate graphic representation of a numeric vectors frequency distribution. It combines a histogram, a density curve, a boxplot and the empirical cumulative distribution function (ecdf) in one single plot. A rug as well as a model distribution curve (e.g. a normal curve) can optionally be superposed. This results in a dense and informative picture of the facts. Still the function remains flexible as all possible arguments can be passed to the single components (hist, boxplot etc.) as a list (see examples).

Usage

PlotFdist(x, main = deparse(substitute(x)), xlab = "", xlim = NULL, do.hist = NULL, args.hist = NULL, args.rug = NA, args.dens = NULL, args.curve = NA, args.boxplot = NULL, args.ecdf = NULL, args.curve.ecdf = NA, heights = NULL, pdist = NULL, na.rm = FALSE, cex.axis = NULL, cex.main = NULL, mar = NULL, las = 1)

Arguments

x

the numerical variable, whose distribution is to be plotted.

main

main title of the plot.

xlab

label of the x-axis, defaults to "". (The name of the variable is typically placed in the main title and would be redundant here.)

xlim

range of the x-axis, defaults to a pretty range(x, na.rm = TRUE).

do.hist

defines, whether a histogram or a plot with type = "h" should be used. Set this to NULL (default) will use a histogram, unless x is an integer with up to 12 unique values!

args.hist

list of additional arguments to be passed to the histogram hist(), ignored if do.hist = FALSE. The defaults chosen when setting args.hist = NULL are more or less the same as in hist.

args.rug

list of additional arguments to be passed to the function rug(). Use args.rug = NA if no rug should be added. This is the default. Use args.rug = NULL to add rug with reasonable default values.

args.dens

list of additional arguments to be passed to density. Use args.dens = NA if no density curve should be drawn. The defaults are taken from density.

args.curve

list of additional arguments to be passed to curve. This argument allows to add a fitted distribution curve to the histogram. By default no curve will be added (args.curve = NA). If the argument is set to NULL, a normal curve with mean(x) and sd(x) will be drawn. See examples for more details.

args.boxplot

list of additional arguments to be passed to the boxplot boxplot(). The defaults are pretty much the same as in boxplot. The two additional arguments pch.mean (default 23) and col.meanci (default "grey80") control, if the mean is displayed within the boxplot. Setting those arguments to NA will prevent them from being displayed.

args.ecdf

list of additional arguments to be passed to ecdf(). Use args.ecdf = NA if no empirical cumulation function should be included in the plot. The defaults are taken from plot.ecdf.

args.curve.ecdf

list of additional arguments to be passed to curve. This argument allows to add a fitted distribution curve to the cumulative distribution function. By default no curve will be added (args.curve.ecdf = NA). If the argument is set to NULL, a normal curve with mean(x) and sd(x) will be drawn. See examples for more details.

heights

heights of the plotparts, defaults to c(2,0.5,1.4) for the histogram, the boxplot and the empirical cumulative distribution function, resp. to c(2,1.5) for a histogram and a boxplot only.

pdist

distances of the plotparts, defaults to c(0, 0), say there will be no distance between the histogram, the boxplot and the ecdf-plot. This can be useful for instance in case that the x-axis has to be added to the histogram.

na.rm

logical, should NAs be omitted? Histogram and boxplot could do without this option, but the density-function refuses to plot with missings. Defaults to FALSE.

cex.axis

character extension factor for the axes.

cex.main

character extension factor for the main title. Must be set in dependence of the plot parts in order to get a harmonic view.

mar

A numerical vector of the form c(bottom, left, top, right) which gives the number of lines of outer margin to be specified on the four sides of the plot. The default is c(0, 0, 3, 0).

las

numeric in 0,1,2,3; the style of axis labels. See par.

Details

Performance has been significantly improved, but if x is growing large (n > 1e7) the function will take its time to complete. Especially the density curve and the ecdf, but as well as the boxplot (due to the chosen alpha channel) will take their time to calculate and plot. In such cases consider taking a sample, i.e. PlotFdist(x[sample(length(x), size=5000)]), the big picture of the distribution won't usually change much. .

See Also

hist, boxplot, ecdf, density, rug, layout

Examples

PlotFdist(x=d.pizza$delivery_min, na.rm=TRUE)

# define additional arguments for hist, dens and boxplot
# do not display the mean and its CI on the boxplot
PlotFdist(d.pizza$delivery_min, args.hist=list(breaks=50),
  args.dens=list(col="olivedrab4"), na.rm=TRUE,
  args.boxplot=list(col="olivedrab2", pch.mean=NA, col.meanci=NA))


# do a "h"-plot instead of a histogram for integers
x <- sample(runif(10), 100, replace = TRUE)
PlotFdist(x, do.hist=FALSE)

# special arguments for hist, density and ecdf
PlotFdist(x=faithful$eruptions,
          args.hist=list(breaks=20), args.dens=list(bw=.1),
          args.ecdf=list(cex=1.2, pch=16, lwd=1), args.rug=TRUE)

# no density curve, no ecdf but add rug instead, make boxplot a bit higher
PlotFdist(x=d.pizza$delivery_min, na.rm=TRUE, args.dens=NA, args.ecdf=NA,
  args.hist=list(xaxt="s"),  # display x-axis on the histogram
  args.rug=TRUE, heights=c(3, 2.5), pdist=2.5, main="Delivery time")

# alpha channel on rug is cool, but takes its time for being drawn...
PlotFdist(x=d.pizza$temperature, args.rug=list(col=SetAlpha("black", 0.1)), na.rm=TRUE)

# plot a normal density curve, but no boxplot nor ecdf
x <- rnorm(1000)
PlotFdist(x, args.curve = NULL, args.boxplot=NA, args.ecdf=NA)

# compare with a t-distribution
PlotFdist(x, args.curve = list(expr="dt(x, df=2)", col="darkgreen"),
          args.boxplot=NA, args.ecdf=NA)
legend(x="topright", legend=c("kernel density", "t-distribution (df=2)"),
       fill=c(getOption("col1", hred), "darkgreen"))

# add a gamma distribution curve to both, histogram and ecdf
ozone <- airquality$Ozone; m <- mean(ozone, na.rm = TRUE); v <- var(ozone, na.rm = TRUE)
PlotFdist(ozone, args.hist = list(breaks=15),
  args.curve = list(expr="dgamma(x, shape = m^2/v, scale = v/m)", col=hecru),
  args.curve.ecdf = list(expr="pgamma(x, shape = m^2/v, scale = v/m)", col=hecru),
  na.rm = TRUE, main = "Airquality - Ozone")

legend(x="topright",
       legend=c(expression(plain("gamma:  ") * Gamma * " " * bgroup("(", k * " = " *
           over(bar(x)^2, s^2) * " , " * theta * plain(" = ") * over(s^2, bar(x)), ")") ),
                "kernel density"),
       fill=c(hecru, getOption("col1", hred)), text.width = 0.25)