rgr: tbplots – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

tbplots

Plot Vertical Tukey Boxplots

Description

Plots a series of vertical Tukey boxplots where the individual boxplots represent the data subdivided by the value of some factor. Optionally the y-axis may be scaled logarithmically (base 10) and the values of the Tukey fences used to identify near and far outliers may also be optionally based on the logarithmically transformed data. A variety of other plot options are available, see Details and Note below.

Usage

tbplots(x, by, log = FALSE, logx = FALSE, notch = TRUE, xlab = "", 
	ylab = deparse(substitute(x)), ylim = NULL, main = "", 
	label = NULL, plot.order = NULL, xpos = NA, width, space = 0.25, 
	las = 1, cex = 1, adj = 0.5, add = FALSE, ssll = 1, colr = 8, 
	...)

Arguments

`x`	name of the variable to be plotted.
`by`	the name of the factor variable to be used to subdivide the data. See Details below for when `by` is undefined.
`log`	to display the data with logarithmic (y-axis) scaling, set `log = TRUE`.
`logx`	if the position of the Tukey boxplot fences are to be computed on the basis of the log transformed data set `logx = TRUE`. When `logx = TRUE` it is ensured that `log = TRUE`.
`notch`	determines if the boxplots are to be “notched” such that the notches indicate the 95% confidence intervals for the medians. The default is to notch the boxplets, to suppress the notches set `notch = FALSE`. See Details below.
`xlab`	a title for the x-axis, by default none is provided.
`ylab`	by default the character string for `x` is used for the y-axis title. An alternate title can be displayed with `xlab = "text string"`, see Examples.
`ylim`	only for `log = FALSE`, defines the limits of the y-axis if the default limits based on the range of the data are unsatisfactory. It can be used to ensure the y-axis scaling in multiple sets of boxplots are the same to facilitate visual comparison.
`main`	a main title may be added optionally above the display by setting `main`, e.g., `main = "Kola Project, 1995"`.
`label`	by default the character strings defining the factors are used to label the boxplots along the x-axis. Alternate labels can be provided with `label = c("Alt1", "Alt2", "Alt3")`, see Examples.
`plot.order`	provides an alternate order for the boxplots. Thus, `plot.order = c(2, 1, 3)` will plot the 2nd ordered factor in the 1st position, the 1st in the 2nd, and the 3rd in its 3rd ordered postion, see Details and Examples below.
`xpos`	the locations along the x-axis for the individual vertical boxplots to be plotted. By default this is set to `NA`, which causes default equally spaced positions to be used, i.e. boxplot 1 plots at value 1 on the x-axis, boxplot 2 at value 2, etc., up to boxplot “n” at value “n”. See Details below for defining `xpos`.
`width`	the width of the boxes, by default this is set to the minimum distance between all adjacent boxplots times the value of `space`. With the default values of `xpos` this results in a minimum difference of `1`, and with the default of `space = 0.25` the width is computed as 0.25. To specify different widths for all boxplots use, for example, `width = c(0.3)`. See Details below for changing individual boxplot widths.
`space`	the space between the individual boxplots, by default this is 0.25 x-axis units.
`las`	controls whether the x-axis labels are written parallel to the x-axis, the default `las = 1`, or are written down from the x-axis by setting `las = 2`. See also, Details below.
`cex`	controls the size of the font used for the factor labels plotted along the x-axis. By default this is `1`, however, if the labels are long it is sometimes necessary to use a smaller font, for example `cex = 0.8` results in a font 80% of normal size.
`adj`	controls the justification of the x-axis labels. By default they are centred, `adj = 0.5`, to left justify them if the labels are written downwards set `adj = 0`.
`add`	permits the user to plot additional boxplots into an existing display. It is recommended that this option is left as `add = FALSE`.
`ssll`	determines the minimum data subset size for which a subset will be plotted. By default this is set to `1`, which leads to only a circle with a median bar being plotted, as the subset size increases additional features of the boxplot are displayed. If `ssll` results in subset boxplots not being plotted, a gap is left and the factor label is still plotted on the x-axis.
`colr`	by default the boxes are infilled in grey, `colr = 8` . If no infill is required, set `colr = 0`. See `display.lty` for the range of available colours.
`...`	further arguments to be passed to methods. For example, the size of the axis titles by setting `cex.lab`, and the size of the plot title by setting `cex.main`. For example, if it is required to make the plot title smaller, add `cex.main = 0.9` to reduce the font size by 10%.

Details

There are two ways to execute this function. Firstly by defining x and by, and secondly by combining the two variables with the split function. See the first two examples below. The split function can be useful if the factors to use in the boxplot are to be generated at run-time, see the last example below. Note that when the split construct is used instead of by the whole split statement will be displayed as the default y-axis title. Also note that when using by the subsets are listed in the order that the factors are encountered in the data, but when using split the subsets are listed alphabetically. In either case they can be re-ordered using plot.order, see Examples.

The width option can be used to define different widths for the individual boxplots. For example, the widths could be scaled to be proportional to the subset population sizes as some function of the square root (const * sqrt(n)) or logarithm (const * log10(n)) of those sizes (n). The constant, const, would need to be chosen so that on average the width of the individual boxes would be approximately 0.25, see Example below. It may be desirable for cosmetic purposes to adjust the positions of the boxes along the x-axis, this can be achieved by specifying xpos.

Long subset (factor) names can lead to display problems, changing the las parameter from its default of las = 1 which plots subset labels parallel to the axis to las = 2, to plot perpendicular to the axis, can help. It may also help to use label and split the character string into two lines, e.g., by changing the string "Granodiorite" that was supplied to replace the coded factor variable GRDR to "Grano-\ndiorite". If this, or setting las = 2, causes a conflict with the x-axis title, if one is needed, the title can be moved down a line by using xlab = "\nLithological Units". In both cases the \n forces the following text to be placed on the next lower line.

If there are more than 7 labels (subsets) and no alternate labels are provided las is set to 2, otherwise some labels may fail to be displayed.

The notches in the boxplots indicate the 95% confidence intervals for the medians and can extend beyond the upper and lower limits of the boxes indicating the middle 50% of the data when subset population sizes are small. The confidence intervals are estimated using the binomial theorem. It can be argued that for small populations a normal approximation would be better. However, it was decided to remain with a non-parametric estimate despite the fact that the calculation of the Tukey fence values involves normality assumptions.

Note

This function is based on a script shared by Doug Nychka on S-News, April 28, 1992.

Any less than detection limit values represented by negative values, or zeros or other numeric codes representing blanks in the data, must be removed prior to executing this function, see ltdl.fix.df.

Any NAs in the data vector are removed prior to preparing the boxplots.

For summary statistics displays to complement the graphics see gx.summary.groups or framework.summary.

Author(s)

Douglas W. Nychka and Robert G. Garrett

Examples

## Make test data kola.c available
data(kola.c)
attach(kola.c)

## Display a simple Tukey boxplot
tbplots(Cu, by = COUNTRY)
tbplots(split(Cu,COUNTRY))

## Display a more appropriately labelled and scaled boxplot
tbplots(Cu, by = COUNTRY, log =  TRUE, logx = TRUE, xlab = "Country", 
	ylab = "Ni (mg/kg) in <2 mm Kola C-horizon soil")

## Display a west-to-east re-ordered plot using the full country names
tbplots(split(Cu, COUNTRY), log = TRUE, logx = TRUE, 
	ylab = "Ni (mg/kg) in <2 mm Kola C-horizon soil",
	label = c("Finland", "Norway", "Russia"), 
	plot.order = c(2, 1, 3))

## Detach test data kola.c
detach(kola.c)

## Make test data kola.o available, setting a -9999, indicating a
## missing pH measurement, to NA
data(kola.o)
kola.o.fixed <- ltdl.fix.df(kola.o, coded = -9999)
attach(kola.o.fixed)

## Display relationship between pH in one pH unit intervals and Cu in 
## O-horizon (humus) soil
tbplots(split(Cu,trunc(pH+0.5)), log=TRUE, logx = TRUE, 
	xlab = "O-horizon soil pH to the nearest pH unit",
	ylab = "Cu (mg/kg) in <2 mm Kola O-horizon soil")

## As above, but demonstrating the use of variable box widths and the
## suppression of 95% confidence interval notches.  The box widths are
## computed as (Log10(n)+0.1)/5, the 0.1 is added as one subset has a
## population of 1.  Note: paste is used in constructing xlab, below,
## as the label is long and overflows the text line length
table(trunc(pH+0.5))
tbplots(split(Cu,trunc(pH+0.5)), log=TRUE, logx = TRUE, notch = FALSE,
	xlab = paste("O-horizon soil pH to the nearest pH unit,",
	"\nbox widths proportional to Log(subset_size)"),
	ylab = "Cu (mg/kg) in <2 mm Kola O-horizon soil",
	width = c(0.26, 0.58, 0.24, 0.02))

## Detach test data kola.o.fixed
detach(kola.o.fixed)

rgr

Applied Geochemistry EDA

v1.1.15

GPL-2

Authors

Robert G. Garrett

Initial release

2018-03-05