A closer look at the ggpubr package for R

Prerequisites

Load required packages:

Code

library(pacman)
p_load(tidyverse, rio, ggpubr, rstatix, scales, flextable)

Introduction

If you are using R there is a very good chance you are creating plots using the ggplot2 package.

What you might not know is that there are now over 100 registered extensions available which support and/or extend ggplot2¹.

One of my favourites is ggpubr - a package that provides some easy-to-use functions for creating clean and publication ready plots.

Let’s see what ggpubr has to offer using the sports car price data set.

Are faster sports cars more expensive?

The question we want to answer is will spending more money on a sports car gets you a faster car?

To answer this question we will create a plot and leverage some of the functions provided by ggpubr.

Start by downloading and importing the data set:

Code

data <- rio::import("sport_car_price.csv") %>% 
  as_tibble()
flextable::flextable(data %>% head)

Car Make	Car Model	Year	Engine Size (L)	Horsepower	Torque (lb-ft)	0-60 MPH Time (seconds)	Price (in USD)
Porsche	911	2,022	3	379	331	4	101,200
Lamborghini	Huracan	2,021	5.2	630	443	2.8	274,390
Ferrari	488 GTB	2,022	3.9	661	561	3	333,750
Audi	R8	2,022	5.2	562	406	3.2	142,700
McLaren	720S	2,021	4	710	568	2.7	298,000
BMW	M8	2,022	4.4	617	553	3.1	130,000

Then clean up the data and convert time taken to reach 60 mph (in seconds) from a continuous to a categorical variable:

Code

data <-
  data %>% filter(!grepl("Electric", `Engine Size (L)`)) %>% 
  filter(`Engine Size (L)` != "N/A") %>% 
  filter(`Engine Size (L)` != "0") %>% 
  filter(`Engine Size (L)` != "-") %>%
   filter(!grepl("Hybrid", `Engine Size (L)`)) %>%
  distinct(`Car Make`,`Car Model`,Year,`Engine Size (L)`,Horsepower, `Torque (lb-ft)`, .keep_all = TRUE)

data <- data %>% 
  mutate(across(everything(), ~str_replace_all(., ",","")))

data <- data %>% 
  mutate(across(c(`Price (in USD)`,`Horsepower`, `Torque (lb-ft)`, `Engine Size (L)`, `0-60 MPH Time (seconds)`),
         as.numeric))

data <- data %>% 
  mutate(`time(0-60mph)`=  cut(`0-60 MPH Time (seconds)`, breaks = c(2, 2.9, 4 , Inf),
     labels = c("< 3", "3-4", "> 4"),
     include.lowest = TRUE))

data <- data %>% 
  rename(price = `Price (in USD)`)

flextable::flextable(data %>% 
                       select(-`Engine Size (L)`, -Horsepower, -`Torque (lb-ft)`, -`0-60 MPH Time (seconds)`) %>% 
                       head())%>%
  align_text_col(align = "right") %>%
  flextable::set_table_properties(layout = "autofit",  width = 1)

Car Make	Car Model	Year	price	time(0-60mph)
Porsche	911	2022	101,200	3-4
Lamborghini	Huracan	2021	274,390	< 3
Ferrari	488 GTB	2022	333,750	3-4
Audi	R8	2022	142,700	3-4
McLaren	720S	2021	298,000	< 3
BMW	M8	2022	130,000	3-4

Then create boxplots comparing the price of the 0-to-60 mph groups using ggplot2 basic settings:

Code

plot <- 
  ggplot(data, aes(x=`time(0-60mph)`, y=price)) + 
  geom_boxplot(aes( fill = `time(0-60mph)`)) +
  ylab("Price (USD)") +
  xlab("Time 0-60 mph (sec)")

plot

The default ggplot2 look and feel is OK, but could definitely be improved.

This is where ggpubr can help.

Theme

First add the theme_pubr() theme to the plot:

Code

plot <- plot + ggpubr::theme_pubr()
plot

Then add the colour palette taken from the New England Journal of Medicine with the set_palette function:

Code

plot <- plot %>% 
  ggpubr::set_palette(palette = "nejm") +
  theme(legend.position = 'none')

plot

This is a much better looking plot, but the price is in scientific notation and the price range across groups is so large that it is making the boxplots appear compressed.

Let’s further improve the plot.

Scales

Using the scales package, set the y-axis to a logarithmic scale and convert it to dollar labels:

Code

plot <- plot +
  scale_y_continuous(
    trans = 'log10',
    labels = scales::label_dollar(scale_cut = cut_short_scale()
                                  ),
    breaks = scales::breaks_log(n = 5),
    expand = expansion(mult = c(0, 0.1))
    )

plot

Statistics

We can also improve the plot by adding statistics.

In combination with the rstatix package, ggpubr can make adding statistical labels to a plot relatively easy.

To add statistics, use the t_test function from the rstatix package to create a data frame with the comparisons required²:

Code

stat_test <- data %>% 
  rename(time = `time(0-60mph)`) %>% 
  rstatix::t_test(price ~ time, detailed = TRUE) %>%
  rstatix::adjust_pvalue(method = "bonferroni") %>%
  rstatix::add_significance("p.adj")

flextable::flextable(stat_test)

estimate	estimate1	estimate2	.y.	group1	group2	n1	n2	statistic	p	df	conf.low	conf.high	method	alternative	p.adj	p.adj.signif
1,116,868.26	1,305,436.8	188,568.5	price	< 3	3-4	76	173	6.979559	0.0000000008180	78.99424	798,356.27	1,435,380.3	T-test	two.sided	0.0000000024540	****
1,192,797.13	1,305,436.8	112,639.6	price	< 3	> 4	76	66	7.506716	0.0000000000912	76.81129	876,379.71	1,509,214.5	T-test	two.sided	0.0000000002736	****
75,928.87	188,568.5	112,639.6	price	3-4	> 4	173	66	2.449132	0.0150000000000	235.15664	14,851.02	137,006.7	T-test	two.sided	0.0450000000000	*

We can now add the significance values to our plot:

Code

# Add p-values onto the box plots
stat_test <- 
  stat_test %>%
  add_y_position(y.trans = log10,
                 step.increase = 0.5)

plot <- plot + 
  stat_pvalue_manual(
  stat_test,  
  label = "{p.adj.signif}",
  tip.length = 0.005,
  size = 6
  )

plot

Now make some final adjustments to the plot³:

Code

f1 <- function(x) {
  log10(mean(10 ^ x)) 
}

plot <- plot +
  stat_summary(fun.y = f1, colour = 'grey', size = 1.5) +
  theme(axis.text = element_text(size = 20),
        axis.title = element_text(size = 24))
plot

And there we have a really clean looking plot that is easy to read and shows statistical significance.

Conclusion

ggpubr is a fantastic extension to ggplot2 and makes it easy to turn good plots into great plots.

So does spending more money on a sports car get you a faster car?

Thanks to ggpubr, we can see that on average the > 4 second cars are the least expensive, the 3-4 seconds cars are in the middle, and the < 3 seconds cars are the most expensive.

Thus, spending more money on a sports car does appear to buy you a faster car.

Footnotes

ggplot2 extensions gallery ↩︎
I adjusted the alpha (α) level so that the probability of committing a type I error is controlled for. I also had to rename the time(0-60mph) variable to time. This is because the t_test function could not find the variable when it was input with back ticks (`) which is the only way I know how when there is a numeric value within the name.↩︎
Increase the axis text and title size and add the average price to the boxplot represented as a grey dot.↩︎