How to plot parallel coordinates with multiple categorical variables in R

Lexi picture Lexi · Jun 4, 2017 · Viewed 7.9k times · Source

I am facing a difficulty while plotting a parallel coordinates plot using the ggparcoord from the GGally package. As there are two categorical variables, what I want to show in the visualisation is like the image below. I've found that in ggparcoord, groupColumn is only allowed to a single variable to group (colour) by, and surely I can use showPoints to mark the values on the axes, but i also need to vary the shape of these markers according to the categorical variables. Is there other package that can help me to realise my idea?

Any response will be appreciated! Thanks!

university and country are two categories

Answer

eipi10 picture eipi10 · Jun 4, 2017

It's not that difficult to roll your own parallel coordinates plot in ggplot2, which will give you the flexibility to customize the aesthetics. Below is an illustration using the built-in diamonds data frame.

To get parallel coordinates, you need to add an ID column so you can identify each row of the data frame, which we'll use as a group aesthetic in ggplot. You also need to scale the numeric values so that they'll all be on the same vertical scale when we plot them. Then you need to take all the columns that you want on the x-axis and reshape them to "long" format. We do all that on the fly below with the tidyverse/dplyr pipe operator.

Even after limiting the number of category combinations, the lines are probably too intertwined for this plot to be easily interpretable, so consider this merely a "proof of concept". Hopefully, you can create something more useful with your data. I've used colour (for the lines) and fill (for the points) aesthetics below. You can use shape or linetype instead, depending on your needs.

library(tidyverse)
theme_set(theme_classic())

# Get 20 random rows from the diamonds data frame after limiting
#  to two levels each of cut and color
set.seed(2)
ds = diamonds %>% 
  filter(color %in% c("D","J"), cut %in% c("Good", "Premium")) %>%
  sample_n(20)

ggplot(ds %>% 
         mutate(ID = 1:n()) %>%             # Add ID for each row
         mutate_if(is.numeric, scale) %>%   # Scale numeric columns
         gather(key, value, c(1,5:10)),     # Reshape to "long" format
       aes(key, value, group=ID, colour=color, fill=cut)) +
  geom_line() +
  geom_point(size=2, shape=21, colour="grey50") +
  scale_fill_manual(values=c("black","white"))

enter image description here

I haven't used ggparcoords before, but the only option that seemed straightforward (at least on my first try with the function) was to paste together two columns of data. Below is an example. Even with just four category combinations, the plot is confusing, but maybe it will be interpretable if there are strong patterns in your data:

library(GGally)

ds$group = with(ds, paste(cut, color, sep="-"))

ggparcoord(ds, columns=c(1, 5:10), groupColumn=11) +
  theme(panel.grid.major.x=element_line(colour="grey70"))

enter image description here