Is there a ggplot2 analogue to the avPlots function in R? - r

Question

When undertaking regression modelling it is useful to produce added variable plots for the explanatory variables in the model, to check whether the posited relationships to the response variable are appropriate to the data. The avPlots function in the car package in R takes a model input, and produces a grid of added-variable plots using the base graphics system. This function is extremely user-friendly, insofar as all you need to do is put in the model object as an argument, and it automatically produces all the added variable plots for each explanatory variable. This matrix of plots contains all the desired information, but unfortunately the plots look poor, owing to the fact that it uses the base graphics system rather than the ggplot2 package. For example, using data found here (downloaded as the file Trucking.csv) here is the output of the avPlots function.
#Load required libraries
library(car);
#Import data, fit model, and show AV plots
DATA <- read.csv('Trucking.csv');
MODEL <- lm(log(PRICPTM) ~ DISTANCE + PCTLOAD + ORIGIN + MARKET + DEREG + PRODUCT,
data = DATA);
avPlots(MODEL);
Question: Is there an equivalent function in ggplot2 that produces a matrix of each of the added-variable plots for a model, but with "prettier" plots? Is it possible to produce these plots, but then customise them using standard ggplot syntax?

Answer 1

I am not aware of any automated function that produces the added variable plots using ggplot. However, as well as giving a plot output as a side-effect of the function call, the avPlots function produces an object that is a list containing the data values used in each of the added variable plots. It is relatively simple to extract data frames of these variables and use these to generate added variable plots using ggplot. This can be done for a general model object using the following functions.
avPlots.invis <- function(MODEL, ...) {
ff <- tempfile()
png(filename = ff)
OUT <- car::avPlots(MODEL, ...)
dev.off()
unlink(ff)
OUT }
ggAVPLOTS <- function(MODEL, YLAB = NULL) {
#Extract the information for AV plots
AVPLOTS <- avPlots.invis(MODEL)
K <- length(AVPLOTS)
#Create the added variable plots using ggplot
GGPLOTS <- vector('list', K)
for (i in 1:K) {
DATA <- data.frame(AVPLOTS[[i]])
GGPLOTS[[i]] <- ggplot2::ggplot(aes_string(x = colnames(DATA)[1],
y = colnames(DATA)[2]),
data = DATA) +
geom_point(colour = 'blue') +
geom_smooth(method = 'lm', se = FALSE,
color = 'red', formula = y ~ x, linetype = 'dashed') +
xlab(paste0('Predictor Residual \n (',
names(DATA)[1], ' | others)')) +
ylab(paste0('Response Residual \n (',
ifelse(is.null(YLAB),
paste0(names(DATA)[2], ' | others'), YLAB), ')')) }
#Return output object
GGPLOTS }
The function ggAVPLOTS will take an input model and produce a list of ggplot objects for each of the added variable plots. These have been constructed to give "pretty" plots with blue points and a dashed red regression line through each plot. If you want all the added variable plots to show up in a single plot, it is relatively simple to do this using the grid.arrange function in the gridExtra package. Below we apply this to your model and show the resulting plot.
#Produce matrix of added variable plots
library(gridExtra)
PLOTS <- ggAVPLOTS(MODEL)
K <- length(PLOTS)
NCOL <- ceiling(sqrt(K))
AVPLOTS <- do.call("arrangeGrob", c(PLOTS, ncol = NCOL, top = 'Added Variable Plots'))
ggsave('AV Plots - Trucking.jpg', width = 10, height = 10)
It is possible to make whatever alterations you want to these plots in the ggplot code above, so if a user prefers to change the colours, font sizes, etc., this is done using standard syntax in ggplot. This method works by importing the data for the added variable plots from the avPlots function, but once you have done that, you can use this data to produce any kind of plot.