MESS (version 0.5.12)

wallyplot.default: Plots a Wally plot

Description

Produces a 3x3 grid of residual- or qq-plots plots from a lm object. One of the nine subfigures is the true residual plot/qqplot while the remaining are plots that fulfill the assumptions of the linear model

Usage

# S3 method for default
wallyplot(
  x,
  y = x,
  FUN = residualplot,
  hide = TRUE,
  simulateFunction = rnorm,
  model = NULL,
  ...
)
# S3 method for lm
wallyplot(
  x,
  y = x,
  FUN = residualplot,
  hide = TRUE,
  simulateFunction = lmsimresiduals,
  ...
)
wallyplot(
  x,
  y = x,
  FUN = residualplot,
  hide = TRUE,
  simulateFunction = rnorm,
  ...
)

Arguments

x

a numeric vector of x values, or an lm object.

y

a numeric vector of y values of the same length as x or a n * 9 matrix of y values - one column for each of the nine plots to make. The first column is the one corresponding to the results from the dataset

FUN

a function that accepts an x, y and ... argument and produces a graphical model validation plots from the x and y values.

hide

logical; if TRUE (the default) then the identity of the true residual plot is hidden until the user presses a key. If FALSE then the true residual plot is shown in the center.

simulateFunction

The function used to produce y values under the null hypothesis. Defaults to rnorm

model

Optional input to simulateFunction

...

Other arguments passed to the plot function FUN

Author

Claus Ekstrom

Details

Users who look at residual plots or qqnorm plots for the first time often feel they lack the experience to determine if the residual plot is okay or if the model assumptions are indeed violated. One way to convey "experience" is to plot a series of graphical model validation plots simulated under the model assumption together with the corresponding plot from the real data and see if the user can pinpoint one of them that looks like an odd-one-out. If the proper plot from the real data does not stand out then the assumptions are not likely to be violated.

The Wallyplot produces a 3x3 grid of plots from a lm object or from a set of pairs of x and y values. One of the nine subfigures is the true plot while the remaining are plots that fulfill the assumptions of the linear model. After the user interactively hits a key the correct residual plot (correponding to the provided data) is shown.

The plotting function can be set using the FUN argument which should be a function that accepts x, y and ... arguments and plots the desired figure. When y is a single vector the same length as x then the function simulateFunction is used to generate the remaining y values corresponding the situations under the null.

For a description of the features of the default residual plot see the help page for residualplot.

References

Ekstrom, CT (2014) Teaching 'Instant Experience' with Graphical Model Validation Techniques. Teaching Statistics (36), p 23-26

Examples

if (FALSE) {
data(trees)
res <- lm(Volume ~ Height + Girth, data=trees)
wallyplot(res)


# Create a grid of QQ-plot figures
# Define function to plot a qq plot with an identity line
qqnorm.wally <- function(x, y, ...) { qqnorm(y, ...) ; abline(a=0, b=1) }
wallyplot(res, FUN=qqnorm.wally, main="")

# Define function to simulate components+residuals for Girth
cprsimulate <- function(n) {rnorm(n)+trees$Girth}
# Create the cpr plotting function
cprplot <- function(x, y, ...) {plot(x, y, pch=20, ...) ;
                                 lines(lowess(x, y), lty=3)}
# Create the Wallyplot
wallyplot(trees$Girth, trees$Girth+rstudent(res), FUN=cprplot,
          simulateFunction=cprsimulate, xlab="Girth")
}