variableOut: Excluding a variable from the model
In glmxdiag: A Collection of Graphic Tools for GLM Diagnostics and some Extensions

Description Usage Arguments Details Value Author(s) Examples

Graphically compares models in terms of Information Criterion, each one corresponds to a model where a specific variable is deleted from the linear predictor.

1 2	variableOut(model, k = 2, update.it = FALSE, xlab, ylab, pch, col, lty, ylim, ...)

`model`	a model supported by `glmxdiag`.
`k`	numeric, the penalty per parameter to be used; the default `k = 2` is the classical AIC.
`update.it`	logical; if TRUE, model without the variable corresponding to the lowest score is returned.
`xlab`	title for the x axis.
`ylab`	title for the y axis, by default is 'AIC'; should be changed if a different k is chosen.
`pch`	type of points.
`col`	color of points and segments.
`lty`	type of horizontal line.
`ylim`	y limits of the plot.
`...`	further arguments passed to `plot.`

Each plotted point corresponds to the score of the model where the variable indicated on x axis is excluded. A dashed line is drawn in correspondence of the full model score.

Points and segments corresponding to variables whose deletion lead to a increment of the scored are black, those who lead to a decrement are red.

The output plot can be seen as a graphic version of the first step of stepAIC function inside MASS package.

Theory about Information Criterion suggests that if the minimum score doesn't belong to the full model, then the linear predictor may not be appropriate.

Called for side effects, but if update.it is set to TRUE returns a model without the variable corresponding to the lowest score.

Giuseppe Reale

data("moons")
model <- glm(Moons ~ Diameter * Mass + Distance, family = poisson, data = moons)
variableOut(model)

n <- nobs(model)
new.model <- variableOut(model, k = log(n), ylab = 'BIC', update.it = TRUE)