In Guerry: Maps, Data and Methods Related to Guerry (1833) "Moral Statistics of France"

data(Guerry, package = "Guerry")
library(dplyr)
library(ggplot2)

Plots of means

In his discussion, Guerry often referred to differences among the Regions of France. One useful display to address this is a profile plot of the means of the moral variables for each Region, typically shown as a line plot of the mean value over the variables. Another term used for this is a parallel coordinates plot.

First, find the means of each variable.

guerry_sumry <- Guerry |>
  select(Region, Crime_pers:Suicides) |>
  filter(!is.na(Region)) |>
  mutate(Region = factor(Region, 
                         labels =c("Central", "East", "North", "South", "West"))) |>
  group_by(Region) |>
  summarise(Crime_pers=  mean(Crime_pers),
            Crime_prop = mean(Crime_prop),
            Literacy =   mean(Literacy),
            Donations =  mean(Donations),
            Infants =    mean(Infants),
            Suicides =   mean(Suicides),
            ) |>
  ungroup() |>
  mutate_at(vars(-Region), scales::rescale)

guerry_sumry |> knitr::kable(digits = 2)

Even in this simple table, the regions that stand out as best (1) and worst (0) can be seen. In particular, the Central, North and West are highest on personal crime, while personal crime is highest in the Central region and lowest in the South.

A plot of these means by region can be produced with the ggpcp package.

#| fig.width = 8
library(ggpcp)
guerry_sumry |>
  pcp_select(Region, Crime_pers:Suicides) |>
  pcp_scale(method="uniminmax") |>
  pcp_arrange() |>
  ggplot(aes_pcp()) + 
  geom_pcp_axes() + 
  geom_pcp(aes(colour = Region), linewidth = 2) +
  geom_pcp_labels() +
  xlab("Variable") +
  ylab("Percent of maximum") +
  theme_bw(base_size = 14) +
  theme(legend.position = 'bottom') 

Radar chart

A more attractive form for this display shows the axes for each variable in polar coordinates, known as a radar chart. Because the variables are on different scales, a first step is to normalize them, by scaling each to a range of (0,1). The scales::rescale() function handles this.

Here, I use the ggradar package by Ricardo Bion. It is not on CRAN, so you may need to install it to reproduce the figure here.

#| fig.width = 8
if(!require(ggradar)) remotes::install_github("ricardo-bion/ggradar")
library(ggradar)   # Create radar charts using ggplot2

guerry_sumry |>
  ggradar(
    grid.label.size = 4, 
    axis.label.size = 3,
    axis.label.offset = 1.2,
    group.point.size = 4,
    plot.extent.x.sf = 1,
    plot.extent.y.sf = 1.2,
    fill = TRUE,
    fill.alpha = 0.2,
    legend.position = "bottom",
    legend.title = "Region",
    legend.text.size = 12,
    plot.title = "Guerry data: Means by Region")

Guerry documentation built on Oct. 25, 2023, 1:07 a.m.