Why R plots Residuals vs Leverage instead of Residuals vs Factor Levels (ANOVA test & model with &#0-CodePudding

I'm analysing a data set with the weight of newborn babies and some info about their mothers, including a categorical variable 'smoke' - whether a mother is a smoker, or not.

I did an aov test and wanted to plot diagnostic plots of an ANOVA model with its help. I expected to get four plots, including a 'Residuals vs Factor Levels' plot. Instead, I got a 'Residuals vs Leverage' plot, as if my categorical variable was a numeric one.

You can find the dataset here:

I tried to check whether 'smoke' is a factor as I wanted or not, like that:

> head(babies$smoke)
[1] non smoker non smoker smoker     non smoker smoker     non smoker
Levels: non smoker smoker

Since 'smoke' is a factor (as I understand) and a categorical variable, why is there leverage as per numeric variable? How to fix this and get the proper plot?

Thanks for the help in advance!

CodePudding user response：

There's no plot.aov method, so you got the plot.lm method (since the class of aov(babies$bwt ~ babies$smoke) would be c("aov", "lm")).

If you want to plot something else for "aov" objects, write a function plot.aov to do it. For example,

plot.aov <- function(x, which = 1:3, ...) {
  NextMethod(x, which = which, ...)
  for (f in names(x$xlevels)) {
    xvar <- x$model[, f]
    plot(xvar, residuals(x), xlab = f, ylab = "Residuals")
  }
}

Then

plot(aov(babies$bwt ~ babies$smoke))

will plot the first 3 plot.lm plots, followed by residual plots against the factor in your model. (It will do each of the factors if there are others.) I get

Alternatively, you can use the termplot function, e.g.

termplot(aov(babies$bwt ~ babies$smoke), partial.resid=TRUE)

which gives this plot: