plotblock: Factor Variable and Random Effects Plots

Description

Function to plot effects for model terms including factor, or group variables for random effects, typically used for objects created within bayesx or read.bayesx.output.

Usage

plotblock(x, residuals = FALSE, range = c(0.3, 0.3), 
  col.residuals = "black", col.lines = "black", c.select = NULL, 
  fill.select = NULL , col.polygons = NULL, data = NULL,
  shift = NULL, trans = NULL, ...)

Arguments

x: either a list of length of the unique factors, where each list element contains the estimated effects for one factor as a matrix, see fitted.bayesx, or one data matrix with first column as the group or factor variable. Also formulas are accepted, e.g it is possible to specify the plot with f ~ x or f1 + f2 ~ x. By convention, the covariate for which effects should be plotted, is always in the first column in the resulting data matrix, that is used for plotting, i.e. in the second formula example, the data matrix is cbind(x, f1, f2), also see argument c.select and fill.select.
residuals: if set to TRUE, partial residuals will be plotted if available. Partial residuals may be set as an attribute of x named "partial.resids", where the partial residuals must be a matrix with first column specifying the covariate, and second column the partial residuals that should be plotted.
range: numeric vector, specifying the left and right bound of the block.
col.residuals: the color of the partial residuals.
col.lines: vector of maximum length of columns of x minus 1, specifying the color of the lines.
c.select: integer vector of maximum length of columns of x, selects the columns of the resulting data matrix that should be used for plotting. E.g. if x has 5 columns, then c.select = c(1, 2, 5) will select column 1, 2 and 5 for plotting. Note that first element of c.select should always be 1, since this is the column of the covariate the effect is plotted for.
fill.select: integer vector, select pairwise the columns of the resulting data matrix that should form one polygon with a certain background color specified in argument col. E.g. x has three columns, or is specified with formula f1 + f2 ~ x, then setting fill.select = c(0, 1, 1) will draw a polygon with f1 and f2 as boundaries. If x has five columns or the formula is e.g. f1 + f2 + f3 + f4 ~ x, then setting fill.select = c(0, 1, 1, 2, 2), the pairs f1, f2 and f3, f4 are selected to form two polygons.
col.polygons: specify the background color for the upper and lower confidence bands, e.g. col = c("green", "red").
data: if x is a formula, a data.frame or list. By default the variables are taken from environment(x): typically the environment from which plotblock is called.
shift: numeric. Constant to be added to the smooth before plotting.
trans: function to be applied to the smooth before plotting, e.g., to transform the plot to the response scale.
...: graphical parameters, please see the details.

Author

Nikolaus Umlauf, Thomas Kneib, Stefan Lang, Achim Zeileis.

Details

Function plotblock draws for every factor or group the effect as a "block" in one graphic, i.e. similar to boxplots, estimated fitted effects, e.g. containing quantiles for MCMC estimated models, are drawn as one block, where the upper lines represent upper quantiles, the middle line the mean or median, and lower lines lower quantiles, also see the examples. The following graphical parameters may be supplied additionally:

cex: specify the size of partial residuals,
lty: the line type for each column that is plotted, e.g. lty = c(1, 2),
lwd: the line width for each column that is plotted, e.g. lwd = c(1, 2),
poly.lty: the line type to be used for the polygons,
poly.lwd: the line width to be used for the polygons,
density angle, border: see polygon,
...: other graphical parameters, see function plot.

Examples

Run this code

## generate some data
set.seed(111)
n <- 500

## regressors
dat <- data.frame(fac = factor(rep(1:10, n/10)))

## response
dat$y <- with(dat, 1.5 + c(2.67, 5, 6, 3, 4, 2, 6, 7, 9, 7.5)[fac] + 
  rnorm(n, sd = 0.6))

if (FALSE) {
## estimate model
b <- bayesx(y ~ fac, data = dat)
summary(b)

## plot factor term effects
plot(b, term = "fac")

## extract fitted effects
f <- fitted(b, term = "fac")

## now use plotblock
plotblock(f)

## some variations
plotblock(f, col.poly = c(2, 3))
plotblock(f, col.poly = NA, lwd = c(2, 1, 1, 1, 1))
plotblock(f, col.poly = NA, lwd = 3, range = c(0.5,0.5))
plotblock(f, col.poly = NA, lwd = 3, col.lines = 1:5, lty = 1)
plotblock(f, col.poly = NA, lwd = 3, col.lines = 1:5, 
  lty = c(3, 1, 2, 2, 1))
plotblock(f, resid = TRUE)
plotblock(f, resid = TRUE, cex = 0.1)
plotblock(f, resid = TRUE, cex = 0.1, col.resid = 2)
plotblock(f, resid = TRUE, cex = 2, col.resid = 3, pch = 3)
plotblock(f, lty = 0, poly.lty = 1, density = c(5, 20))
plotblock(f, lty = 0, poly.lty = 1, density = c(5, 20), 
  poly.lwd = c(1, 2))
plotblock(f, lty = 0, poly.lty = c(1, 2), density = c(5, 20))
plotblock(f, lty = 0, poly.lty = c(1, 2), density = c(5, 20), 
  border = c("red", "green3"))
plotblock(f, lty = 0, poly.lty = c(1, 2), density = c(5, 20), 
  border = c("red", "green3"), col.poly = c("blue", "yellow"))
plotblock(f, lty = c(1,0,0,0,0), poly.lty = c(1, 2), 
  density = c(5, 20), border = c("red", "green3"), 
  col.poly = c("blue", "yellow"))
plotblock(f, lty = c(1,0,0,0,0), poly.lty = c(1, 2), 
  density = c(20, 20), border = c("red", "green3"), 
  col.poly = c("blue", "yellow"), angle = c(10, 75))
}

## another example
plotblock(y ~ fac, data = dat, range = c(0.45, 0.45))

dat <- data.frame(fac = factor(rep(1:10, n/10)))
dat$y <- with(dat, c(2.67, 5, 6, 3, 4, 2, 6, 7, 9, 7.5)[fac])
plotblock(y ~ fac, data = dat)
plotblock(cbind(y - 0.1, y + 0.1) ~ fac, data = dat)
plotblock(cbind(y - 0.1, y + 0.1) ~ fac, data = dat, 
  fill.select = c(0, 1, 1))
plotblock(cbind(y - 0.1, y + 0.1) ~ fac, data = dat, 
  fill.select = c(0, 1, 1), poly.lty = 2, lty = 1, 
  border = "grey5")