R Graphics Using the ggplot2 Package

We'll begin by plotting the pairwise association between two continuous variables in our data set: Points For (For) and Wins (W).

ggplot(afl_data, aes(x = For, y = W)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins")

The default is to use either a generalized additive model (GAM) or loess for smoothing (the decision is automatically made by ggplot depending on the sample size)

geom_smooth()

ggplot(afl_data, aes(x = For, y = W)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth()

However, for many purposes, we may want to use a linear trend line (e.g., if we are planning to use linear regression).

geom_smooth(method = "lm")

ggplot(afl_data, aes(x = For, y = W)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm")

Notice that ggplot automatically added a +/- 1SE confidence band around the trend line.

To remove this:

geom_smooth(method = "lm", se = FALSE)

ggplot(afl_data, aes(x = For, y = W)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm", se = FALSE)

One way to group by a categorical variable is to use different shapes for the points.

shape = Team_f

ggplot(afl_data, aes(x = For, y = W, shape = Team_f)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins")

Another way to group by a categorical variable is to use different colors for the points.

colour = Team_f

ggplot(afl_data, aes(x = For, y = W, colour = Team_f)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins")

What if we wanted to include both the linear trend line and the grouping by team?

colour = Team_f

geom_smooth(method = "lm")

ggplot(afl_data, aes(x = For, y = W, colour = Team_f)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm")

Was that plot too messy? One way to clean it up is to make a single trend line for all of the data, ignoring Team.

ggplot(afl_data, aes(x = For, y = W)) +
  geom_point(aes(colour = Team_f)) +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm")

We moved colour = Team_f out of the main aesthetics and into the aesthetics for geom_point(). This means that the geom_smooth() function is ignorant of the colour aesthetics passed to geom_point(); as such, it doesn't create a separate trend line for each team.

Another way we could make our visualizations cleaner would be to focus on WA teams.

afl_data <- afl_data %>%
  mutate(Team_wa = case_when(!Team %in% c("West Coast", "Fremantle") ~ "Other",
                             TRUE ~ Team) %>% factor())

colour = Team_wa

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm")

https://colorbrewer2.org/

https://colorbrewer2.org/

?scale_colour_brewer

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_brewer(palette = "Dark2")

What if we wanted the confidence band to be color-coded as well?

fill = Team_wa

scale_fill_brewer(palette = "Dark2")

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_brewer(palette = "Dark2") +
  scale_fill_brewer(palette = "Dark2")

If you want to use your own custom colors, you can set them manually.

Check the Addins dropdown menu above for a comprehensive list of all colors in R.

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_point(colour = "burlywood4") +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm", colour = "peachpuff") 

Note that colour = is NOT included inside aes() This makes customization a bit more challenging.

scale_colour_manual(values = afl_primary_colours)

scale_fill_manual(values = afl_secondary_colours)

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours) +
  scale_fill_manual(values = afl_secondary_colours)

You can also color the points according to a continuous variable if desired.

shape = Team_wa

scale_colour_distiller(palette = "Reds") Note: distiller instead of brewer!

ggplot(afl_data, aes(x = For, y = W, fill = Team_wa)) +
  geom_point(aes(shape = Team_wa, colour = Pt_diff), size = 3) +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_distiller(palette = "Reds", direction = 1) +
  scale_fill_manual(values = afl_secondary_colours) +
  scale_shape_discrete()

If you're picky, you may not like the label that's been applied to the legend. We can fix that.

name = "Team"

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team")

As we get closer to perfecting our plot, let's add a title.

ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams, 2000-2021", subtitle = "2000-2021")

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_point() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021")

We're getting even more pedantic now. Wins is technically not a continuous variable (it's a count variable). As such, some of the points on the y-axis obscure other points. If we want to improve our ability to see each individual point, we can apply jitter by replacing

geom_point()

with

geom_jitter()

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021")

Ok, this looks pretty good. But are you going to be submitting it for publication? There is some “chart junk” here. If you have ever heard of Edward Tufte, you probably know what that is. If we wanted to please Tufte and remove all “chart junk”, we could use the tufte theme from the ggthemes package.

theme_tufte()

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021") +
  theme_tufte()

That may be a bit extreme! My favorite theme for publication-ready figures is the cowplot theme from the cowplot package.

theme_cowplot()

theme_cowplot()

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021") +
  theme_cowplot()

theme_few()

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021") +
  theme_few()

theme_minimal()

ggplot(afl_data, aes(x = For, y = W, colour = Team_wa, fill = Team_wa)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  scale_colour_manual(values = afl_primary_colours, name = "Team") +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  ggtitle("Association Between Points and Wins in WA Teams and Other AFL Teams",
  subtitle = "2000-2021") +
  theme_minimal()

Want to separate your data into multiple smaller plots? Use a facet plot.

facet_wrap(~ Year)

ggplot(afl_data, aes(x = For, y = W)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  ggtitle("Year-to-Year Association Between Points and Wins",
  subtitle = "2000-2021") +
  theme_cowplot() +
  facet_wrap(~ Year)

Use facet_grid if you have multiple dimensions of faceting.

facet_grid(WA_Team ~ Year)

ggplot(afl_data, aes(x = For, y = W)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  ggtitle("Year-to-Year Association Between Points and Wins",
  subtitle = "2000-2021") +
  theme_cowplot() +
  facet_grid(WA_Team ~ Year)

The ggthemes package has themes that attempt to reproduce the styles used in professional publications, like The Economist

theme_economist()

ggplot(afl_data, aes(x = For, y = W)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  ggtitle("Year-to-Year Association Between Points and Wins",
  subtitle = "2000-2021") +
  theme_economist() +
  facet_wrap(~ Year)

And The Wall Street Journal

theme_wsj()

ggplot(afl_data, aes(x = For, y = W)) +
  geom_jitter() +
  xlab("Points For") +
  ylab("Wins") +
  geom_smooth(method = "lm") +
  ggtitle("Year-to-Year Association Between Points and Wins",
  subtitle = "2000-2021") +
  theme_wsj() +
  facet_wrap(~ Year)

ggpairs() comes from the GGally package

ggpairs(afl_data, columns = c(3, 5, 7, 9, 11), ggplot2::aes(colour = Team_wa, fill = Team_wa)) +
  scale_colour_manual(values = afl_secondary_colours) +
  scale_fill_manual(values = afl_secondary_colours) +
  theme_cowplot()

ggplot(afl_data, aes(x = Pt_diff)) + 
  geom_histogram(fill = "peachpuff", colour = "burlywood4") +
  theme_cowplot()

ggplot(afl_data, aes(x = Pt_diff, fill = Team_wa)) + 
  geom_histogram() +
  scale_fill_manual(values = afl_secondary_colours) +
  theme_cowplot()

ggplot(afl_data, aes(x = Pt_diff)) + 
  geom_density() +
  theme_cowplot()

ggplot(afl_data, aes(x = Pt_diff, fill = Team_wa)) + 
  geom_density() +
  scale_fill_manual(values = afl_secondary_colours) +
  theme_cowplot()

ggplot(afl_data, aes(x = Pt_diff, fill = Team_wa)) + 
  geom_density(alpha = .5) +
  scale_fill_manual(values = afl_secondary_colours, name = "Team") +
  theme_cowplot()

afl_data$Year_f <- factor(afl_data$Year)

ggplot(afl_data, aes(x = Year_f, y = For)) + 
  geom_boxplot() +
  xlab("Year") +
  ylab("Points For") +
  theme_cowplot()

Axis labels too scrunched together?

scale_x_discrete(guide = guide_axis(n.dodge = 2))

ggplot(afl_data, aes(x = Year_f, y = For)) + 
  geom_boxplot() +
  xlab("Year") +
  ylab("Points For") +
  theme_cowplot() +
  scale_x_discrete(guide = guide_axis(n.dodge = 2))

ggplot(afl_data, aes(x = Year_f, y = For)) + 
  geom_violin() +
  xlab("Year") +
  ylab("Points For") +
  theme_cowplot() +
  scale_x_discrete(guide = guide_axis(n.dodge = 2))

Spaghetti plots are useful for plotting longitudinal data, grouped by individual cases (in this instance, Teams)

ggplot(afl_data, aes(x = Year, y = W, group = Team_f, colour = Team_wa)) +
  geom_line(size = 2) +
  scale_colour_manual(values = afl_secondary_colours, name = "Team") +
  ylab("Wins") +
  theme_cowplot()

For some data, like bar charts, you may want to plot summaries of data, rather than individual data points.

stat_summary(fun = mean, geom = "bar")

ggplot(afl_data, aes(x = Team_f, y = W)) +
  stat_summary(fun = mean, geom = "bar") +
  theme_cowplot() +
  theme(axis.text.x = element_text(angle = 75, hjust = 1, vjust = 1))

With summaries of data, we often want confidence intervals as well.

stat_summary(fun.data = mean_se, geom = "errorbar")

ggplot(afl_data, aes(x = Team_f, y = W)) +
  stat_summary(fun = mean, geom = "bar") +
  stat_summary(fun.data = mean_se, geom = "errorbar") +
  theme_cowplot() +
  theme(axis.text.x = element_text(angle = 75, hjust = 1, vjust = 1))

Sometimes you may want to flip the orientation of your plots.

coord_flip() (works on all plots, not just bar charts)

ggplot(afl_data, aes(x = Team_f, y = W)) +
  stat_summary(fun = mean, geom = "bar") +
  stat_summary(fun.data = mean_se, geom = "errorbar") +
  theme_cowplot() +
  coord_flip()

Want the y-axis in alphabetical order?

x = forcats::fct_rev(Team_f)

ggplot(afl_data, aes(x = forcats::fct_rev(Team_f), y = W)) +
  stat_summary(fun = mean, geom = "bar") +
  stat_summary(fun.data = mean_se, geom = "errorbar") +
  theme_cowplot() +
  coord_flip()

Want the y-axis in descending order?

x = forcats::fct_reorder(.f = Team_f, .x = W, .fun = mean)

ggplot(afl_data, aes(x = forcats::fct_reorder(.f = Team_f, .x = W, .fun = mean), y = W)) +
  stat_summary(fun = mean, geom = "bar") +
  stat_summary(fun.data = mean_se, geom = "errorbar") +
  theme_cowplot() +
  coord_flip() +
  xlab("Team")

Or maybe we'd prefer to stratify by another variable (e.g., Year), rather than aggregate.

position = "dodge"

ggplot(afl_data %>% filter(Team %in% c("West Coast", "Fremantle")), aes(x = Team_f, y = W, fill = Year_f)) +
  geom_col(position = "dodge", colour = "black") +
  theme_cowplot() +
  theme(axis.text.x = element_text(angle = 75, hjust = 1, vjust = 1)) +
  scale_fill_ordinal()

It is possible to add text as a new plot layer.

annotate("text", x = 1, y = 2, label = "Fremantle"

ggplot(afl_data %>% filter(Team %in% c("West Coast", "Fremantle")), aes(x = Team_f, y = W, fill = Year_f)) +
  geom_col(position = "dodge", colour = "black") +
  theme_cowplot() +
  theme(axis.title.x = element_blank(), axis.text.x = element_blank()) +
  scale_fill_ordinal(name = "Year") +
  annotate("text", x = 1, y = 2, label = "Fremantle", colour = "#2a0d54", size = 12.5) +
  annotate("text", x = 2, y = 2, label = "West Coast", colour = "#f2a900", size = 12.5)

Put the four plots together in a 2x2 grid

(plot1 | plot2) / (plot3 | plot4)