Dot-dot-dot: Three dotplots to stake your life on

Celebrating the beauty of dots and points in plotting data.
statistical analysis
Author

Paw Hansen

Published

June 9, 2023

Here is a quick post on a few basic plots you can make with geom_point. The goal is simply to celebrate the dot as a way of graphing data. Who knows? Maybe someday there will be a post on the line too! For all plots, we will be working with the gapminder data.

Packages used in this post 
library(tidyverse)
library(broom)
library(modelsummary)
library(ggsci)
library(gapminder)

theme_set(theme_minimal())
data("gapminder")

Plot 1: The Cleveland dotplot

To get started, let’s make a classic Cleveland dot plot. These are good for providing at quick glance of summary statistics, here the average life expectancy for each continent:

Code 
gapminder %>%
  group_by(continent) %>%
  summarise(mean = mean(lifeExp, na.rm = T)) %>%
  mutate(continent = fct_reorder(continent, mean)) %>%
  ggplot(aes(mean, continent)) + 
  geom_point(size = 4) +
  labs(x = "Years",
       y = NULL,
       caption = "Source: gapminder")

Figure 1: Mean life expectancy across continents

One small tweak could be to highlight one of the continents. For instance, suppose your analysis was about explaining what’s going on with one particular observation. In particular, when looking at the above plot, Africa, seems to be falling behind the other continents. An easy way to highlight that would be using color:

Code 
gapminder %>%
  group_by(continent) %>%
  summarise(mean = mean(lifeExp, na.rm = T)) %>%
  mutate(continent = fct_reorder(continent, mean)) %>%
  ggplot(aes(mean, continent)) + 
  geom_point(aes(color = I(ifelse(continent == "Africa", 
                                  "firebrick", 
                                  "black"))), 
             size = 4) +
  labs(x = "Years",
       y = NULL, 
       caption = "Source: gapminder")

Figure 2: What’s the matter with Africa? Mean life expectancy across continents

This kind of graph is en easy but powerfull way to spark initial curiosity with readers, which is a key part of storytelling with data

Plot 2: The dot-and-whisker

Next up, we have the dot-and-whisker plot. Often, researchers and data analyst will report model estimates using tables. Some scholars, teasingly, call these ‘BUTONs’: Big Ugly Table of Numbers.

A simple way to improve presentation is to make a dot-and-whisker plot of the model estimates and corresponding uncertainties. Let’s fit a model and make a plot:

# Rescale population and GDP per capita to more meaningfull scale. 
gapminder <- 
  gapminder %>% 
  mutate(population_scaled = pop/100000000,
         gdpPercap_scaled = gdpPercap/10000)

# The fit model
gap_model <- 
  lm(lifeExp ~ continent + year +  population_scaled + gdpPercap_scaled, 
     data = gapminder)
Code 
gap_model %>%
  tidy(conf.int = T) %>%
  filter(term != "(Intercept)") %>%
  mutate(term = str_replace_all(term, 
                                "continent", 
                                "Continent: "),
         term = str_to_title(term),
         term = fct_reorder(term, 
                            estimate)) %>%
  ggplot(aes(estimate, 
             term, 
             xmin = conf.low, 
             xmax = conf.high)) + 
  geom_vline(xintercept = 0, 
             linetype = "dashed",
             color = "grey") + 
  geom_pointrange() + 
  labs(y = NULL, 
       x = "Difference in years")

Figure 3: Explaining life expectancy across the world

For reference, this is what the same results would have looked like if we had used a standard regression table:

Code 
modelsummary(models = list("Life expectancy" = gap_model), 
             fmt = fmt_significant(2),
             stars = T,
             gof_map = c("nobs", "rmse","r.squared"))
Table 1: Explaining life expectancy
Life expectancy
(Intercept) −518***
(20)
continentAmericas 14.29***
(0.49)
continentAsia 9.38***
(0.47)
continentEurope 19.36***
(0.52)
continentOceania 20.6***
(1.5)
year 0.29***
(0.01)
population_scaled 0.18
(0.16)
gdpPercap_scaled 3.0***
(0.2)
Num.Obs. 1704
RMSE 6.87
R2 0.717
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001

The dot-and-whisker plot helps readers quickly compare the model estimates and their corresponding uncertainty.

Plot 3: The dumbell chart

Finally, the dumbell chart can be useful when the goal is to highlight differences, especially across groups. Let’s make one to compare the change in mean life expectancy from the 1950s to the 2000s across all five continents:

Code 
gapminder |> 
  mutate(decade = factor(year %/% 10 * 10)) |> 
  filter(decade %in% c(1950, 2000)) |> 
  group_by(continent, decade) |> 
  summarize(lifeexp = mean(lifeExp, na.rm = T)) |> 
  mutate(continent = fct_reorder(continent, lifeexp)) |> 
  ggplot(aes(lifeexp, continent, 
             group = decade)) + 
  geom_line(aes(group = continent), 
            color = "grey80", 
            linewidth = 1) + 
  geom_point(aes(color = continent), 
             size = 5) + 
  scale_color_brewer(type = "qual", palette = "Set1") + 
  labs(x = "Life expectancy (years)",
       y = NULL,
       caption ="Source: gapminder") + 
  theme(legend.position = "none")

Figure 4: Mean life expectancy in 1950s and 2000s

The graph makes it easy to see that Oceania has the highest average but Asia has seen the highest growth. This might be a useful “Figure 1”; illustrating a puzzle which your paper then tries to explain.

Final thoughts: Three dotplots to stake your life on

Points or dots can be a useful way of plotting data. In this post, I have shown you three different and quite versatile plots you can use in your next analysis.

Cool! Where can I learn more?

  • Healy, Kieran. Data visualization: a practical introduction. Princeton University Press, 2018.