Data Visualization with ggplot2

some morals

“The simple graph has brought more information to the data analyst’s mind than any other device.” —John Tukey

“It is true that data visualization is part data science and part art. That being said, even the most creative art is supported by theories that explain why it works.”
—Michiko Wolcott

what is tidy data

what is tidy data

what is tidy data

in essence, tidy data is data that can be put through standardized tools

the grammar of graphics

the “grammar of graphics” is the answer to the question “what is a statistical graphic?”

graphic from https://r.qcbs.ca/workshop03/book-en/grammar-of-graphics-gg-basics.html

basic structure

the basic structure of a ggplot looks something like:

library(ggplot2)
ggplot(dataset, aes(x = some_column, y = another_column)) + 
  geom_point() + 
  <more options as desired>

let’s do an example

palmerpenguins

here’s the palmerpenguins dataset

library(palmerpenguins)
head(penguins)

# A tibble: 6 × 8
  species island    bill_length_mm bill_depth_mm flipper_length_mm body_mass_g
  <fct>   <fct>              <dbl>         <dbl>             <int>       <int>
1 Adelie  Torgersen           39.1          18.7               181        3750
2 Adelie  Torgersen           39.5          17.4               186        3800
3 Adelie  Torgersen           40.3          18                 195        3250
4 Adelie  Torgersen           NA            NA                  NA          NA
5 Adelie  Torgersen           36.7          19.3               193        3450
6 Adelie  Torgersen           39.3          20.6               190        3650
# ℹ 2 more variables: sex <fct>, year <int>

our first ggplot

ggplot(penguins, aes(x = bill_length_mm, bill_depth_mm)) + 
  geom_point()

our first ggplot

ggplot(penguins, aes(x = bill_length_mm, bill_depth_mm, color = species)) + 
  geom_point()

our first ggplot

ggplot(penguins, aes(x = bill_length_mm, bill_depth_mm, color = species)) + 
  geom_point() + 
  xlab("Bill Length [mm]")

our first ggplot

ggplot(penguins,
       aes(
         x = bill_length_mm,
         y = bill_depth_mm,
         color = species,
         shape = species
       )) + 
  geom_point() + 
  xlab("Bill Length [mm]") +
  ylab("Bill Depth [mm]")

our first ggplot

• code
• plot

ggplot(penguins,
       aes(
         x = bill_length_mm,
         y = bill_depth_mm,
         color = species,
         shape = species,
         label = species,
         group = species
       )) + 
  stat_ellipse() + 
  geom_point() + 
  geom_label(
    data = penguins |> group_by(species) |> summarize(across(c(bill_length_mm, bill_depth_mm), mean, na.rm=T)),
    alpha = 0.8
  ) + 
  xlab("Bill Length [mm]") +
  ylab("Bill Depth [mm]") + 
  ggtitle("Relationship of Species, Bill Length, and Bill Depth",
          "Penguins observed near Palmer Station, Antarctica, 2007-2009") +
  theme_bw() + 
  theme(legend.position = 'none')

visual channels

visual channels

univariate geoms

geom_bar

# if you want ggplot to tally up your data, use geom_bar
ggplot(penguins, aes(x = species)) + 
  geom_bar()

geom_col

# if your data are already tallied up, use geom_col
counts_df <- count(penguins, species)
counts_df

# A tibble: 3 × 2
  species       n
  <fct>     <int>
1 Adelie      152
2 Chinstrap    68
3 Gentoo      124

ggplot(counts_df, aes(x = species, y = n)) + 
  geom_col()

geom_histogram

ggplot(penguins, aes(x = flipper_length_mm)) + 
  geom_histogram()

geom_histogram

ggplot(penguins, aes(x = flipper_length_mm)) + 
  geom_histogram(bins = 50)

geom_histogram

ggplot(penguins, aes(x = flipper_length_mm, fill = species)) + 
  geom_histogram()

geom_density

ggplot(penguins, aes(x = flipper_length_mm)) + 
  geom_density()

geom_density

ggplot(penguins, aes(x = flipper_length_mm, fill = species)) + 
  geom_density()

geom_density

ggplot(penguins, aes(x = flipper_length_mm, fill = species)) + 
  geom_density(alpha = 0.7)

bivariate geoms

geom_point

ggplot(penguins, 
       aes(
         x = flipper_length_mm, 
         y = bill_length_mm, 
         color = species)) +
  geom_point()

geom_point

ggplot(penguins, 
       aes(
         x = flipper_length_mm, 
         y = bill_length_mm, 
         color = species)) +
  geom_point(size = 3)

geom_point

ggplot(penguins, 
       aes(
         x = flipper_length_mm, 
         y = bill_length_mm, 
         color = species,
         size = body_mass_g)) +
  geom_point()

geom_point

ggplot(penguins, 
       aes(
         x = flipper_length_mm, 
         y = bill_length_mm, 
         color = species,
         size = body_mass_g)) +
  geom_point(alpha = 0.6)

geom_line

lines are great to use to depict observations that are conceptually connected — either across time, or thematically.

for example, maybe we would want to know if the average flipper length in the population of penguins observed is changing over time.

flipper_length_over_time <- 
  data.frame(
  year = c(2007, 2007, 2007, 
           2008, 2008, 2008, 
           2009, 2009, 2009),
  flipper_length_mm = 
    c(186.5, 192.4, 215.1,
      191.0, 197.7, 217.5,
      192.0, 198.0, 218.4), 
  species = as.factor(
    c(
      "Adelie", "Chinstrap", "Gentoo",
      "Adelie", "Chinstrap", "Gentoo",
      "Adelie", "Chinstrap", "Gentoo"
    )))

ggplot(flipper_length_over_time,
       aes(x = year,
           y = flipper_length_mm,
           color = species)) + 
  geom_line()

geom_line

a lot of the time when using geom_line, I like to pair that with a geom_point layer on top of it, so where the observations are is more clear.

ggplot(flipper_length_over_time,
       aes(x = year,
           y = flipper_length_mm,
           color = species)) + 
  geom_line() + 
  geom_point()

geom_area

geom_area when used with only one group of data is similar to geom_line, except it fills in the area beneath. this can be useful for depicting how a population breaks down into strata over time.

• code
• data
• plot

ggplot(population_over_time,
       aes(x = year,
           y = population_size,
           fill = group)) +
  geom_area()

year	group	population_size
1947	Employed	60.323
1947	Unemployed	235.600
1947	Armed.Forces	159.000
1948	Employed	61.122
1948	Unemployed	232.500
1948	Armed.Forces	145.600
1949	Employed	60.171
1949	Unemployed	368.200
1949	Armed.Forces	161.600
1950	Employed	61.187
1950	Unemployed	335.100
1950	Armed.Forces	165.000
1951	Employed	63.221
1951	Unemployed	209.900
1951	Armed.Forces	309.900
1952	Employed	63.639
1952	Unemployed	193.200
1952	Armed.Forces	359.400
1953	Employed	64.989
1953	Unemployed	187.000
1953	Armed.Forces	354.700
1954	Employed	63.761
1954	Unemployed	357.800
1954	Armed.Forces	335.000
1955	Employed	66.019
1955	Unemployed	290.400
1955	Armed.Forces	304.800
1956	Employed	67.857
1956	Unemployed	282.200
1956	Armed.Forces	285.700
1957	Employed	68.169
1957	Unemployed	293.600
1957	Armed.Forces	279.800
1958	Employed	66.513
1958	Unemployed	468.100
1958	Armed.Forces	263.700
1959	Employed	68.655
1959	Unemployed	381.300
1959	Armed.Forces	255.200
1960	Employed	69.564
1960	Unemployed	393.100
1960	Armed.Forces	251.400
1961	Employed	69.331
1961	Unemployed	480.600
1961	Armed.Forces	257.200
1962	Employed	70.551
1962	Unemployed	400.700
1962	Armed.Forces	282.700

geom_text

we can use geom_text similarly to geom_point, but instead of plotting a small mark, geom_text places text on the graph.

ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm, label = species)) + 
  geom_text()

geom_text

one use of geom_text that I particularly enjoy is to make what usually has to be guesstimated precise:

# remember counts_df from earlier? 
counts_df

# A tibble: 3 × 2
  species       n
  <fct>     <int>
1 Adelie      152
2 Chinstrap    68
3 Gentoo      124

ggplot(counts_df, aes(x = species, y = n)) + 
  geom_col() + 
  geom_text(
    mapping = aes(y = n + 10, label = n),
    size = 8
  )

geom_boxplot

ggplot(penguins, aes(x = species, y = flipper_length_mm)) + 
  geom_boxplot()

geom_boxplot

if i am going to use boxplots, something i often like to do is to plot the data with jitter behind the boxplots and give the boxplots some transparency.

also, i’m a sucker for colorful figures, so i’ll almost always add color

ggplot(penguins, aes(x = species, y = flipper_length_mm,
                     color = species, fill = species)) + 
  geom_jitter() + 
  geom_boxplot(alpha = 0.6, color = 'black', outlier.color = NA)

geom_violin

ggplot(penguins, aes(x = species, y = flipper_length_mm)) + 
  geom_violin()

geom_violin

a similar layout can be done with geom_violin

ggplot(penguins, aes(x = species, y = flipper_length_mm,
                     color = species, fill = species)) + 
  geom_jitter() + 
  geom_violin(color = 'black', alpha = 0.5)

geom_tile

heatmaps can be created with geom_tile

ggplot(df,
       aes(x = week,
           y = weekday,
           fill = productivity)) + 
  geom_tile()

geom_tile

ggplot(df, aes(x = week, y = weekday, fill = productivity)) + 
  geom_tile() + 
  scale_fill_viridis_c()

facets

facet_wrap

faceting generates multiple panels within a visualization, each showing a different subset of the data.

ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm)) + 
  geom_point() + 
  facet_wrap(~species)

facet_grid

facet_grid gives you more precision in how the faceting panels are laid out by using the left-hand-side and right-hand-side to indicate to the rows and columns.

ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm)) + 
  geom_point() + 
  facet_grid(sex~species)

the easiest way for me to remember which variable corresponds to the columns vs. rows in the facet_grid formula is to remember how formulas usually look, as in y ~ x or y = m*x + b.

y is the variable that corresponds to the vertical dimension and is on the left-hand-side – so y or the left-hand-side corresponds to the rows, while x informs us about the horizontal dimension, so x or the right-hand-side corresponds to the columns.

statistics

let’s say you have a ton of data such that creating a scatter plot isn’t all that useful.

df <- data.frame(
  x = c(rnorm(n = 10000, sd = .5), rnorm(n = 20000, mean = 1.5), sd = .35),
  y = c(rnorm(n = 10000, sd = .5), rnorm(n = 20000, mean = 1.5), sd = .35)
)

ggplot(df, aes(x = x, y = y)) + 
  geom_point()

we want to know if this is one cluster or two… we might look at histograms in x and y.

ggplot(df, aes(x = x)) + 
  geom_histogram()

ggplot(df, aes(x = y)) + 
  geom_histogram()

geom_density2d

but when that still doesn’t work, we can turn to calculating some summary statistics in 2d with geom_density2d

ggplot(df, aes(x = x, y = y)) + 
  geom_point() + 
  geom_density_2d()

another handy way to deal with overplotting is to increase the transparency of the points plotted

ggplot(df, aes(x = x, y = y)) + 
  geom_point(alpha = 0.05)

stat_summary

stat_summary is a very flexible layer that can perform computations for you and depict the results with a geom of your choice

ggplot(penguins, aes(x = species, y = body_mass_g)) + 
  stat_summary(fun = mean, geom = 'point', size = 5)

stat_summary

we can use stat_summary to show us the mean, min and max using a pointrange geom

ggplot(penguins, aes(x = species, y = body_mass_g)) + 
  stat_summary(fun = mean, geom = 'pointrange',
               fun.max = max, fun.min = min,
               size = 5)

stat_summary

we can use stat_summary to show us the mean, min and max using a pointrange geom

ggplot(penguins, aes(x = species, y = body_mass_g)) + 
  stat_summary(fun.data = mean_cl_normal, geom = 'pointrange',
               size = 5)

coordinates

scale_x_* and scale_y_*

occasionally we have data that is best represented on a non-linear scale, like the log-scale. ggplot makes it very easy to do this. compare:

ggplot(df, aes(x = x)) + 
  geom_histogram()

ggplot(df, aes(x = x)) + 
  geom_histogram() + 
  scale_x_log10()

scale_color_* and scale_fill_*

often you will want to customize the color and fill palettes in your plots. there are a handful of ways to do this, and they typically fall into 3 categories:

• categorical
  • scale_*_discrete()
  • scale_*_brewer()
• quantitative
  • scale_*_continuous()
  • scale_*_distiller()
  • scale_*_gradient()
• manual
  • scale_*_manual()

scale_color_* and scale_fill_*

RColorBrewer: https://www.datanovia.com/en/blog/the-a-z-of-rcolorbrewer-palette/

ggplot(df, aes(x = x, y = y, fill = z)) + 
  geom_tile() +
  scale_fill_distiller()

ggplot(df, aes(x = x, y = y, fill = z)) + 
  geom_tile() +
  scale_fill_distiller(palette = 'RdYlBu')

viridis: https://cran.r-project.org/web/packages/viridis/vignettes/intro-to-viridis.html

ggplot(df, aes(x = x, y = y, fill = z)) + 
  geom_tile() +
  scale_fill_viridis_c()

ggplot(df, aes(x = x, y = y, fill = z)) + 
  geom_tile() +
  scale_fill_viridis_c(option = 'magma')

Having trouble picking a color palette for your #Rstats visualization? Well here's a MEGA thread about all the ways you can choose a palette! 🧵[1/22]

— Moriah Taylor (she/her) (@moriah_taylor58) May 20, 2021

https://twitter.com/moriah_taylor58/status/1395431000977649665

coordinate labelling

sometimes you want to format the way the axis numbers appear in a particular way, like in scientific format or in dollars.

ggplot(df, aes(x = household_income)) + 
  geom_histogram() + 
  scale_x_log10(labels = scales::dollar_format())

ggplot(df, aes(x = program, y = admit_rate)) + 
  geom_point(size = 5) + 
  scale_y_continuous(labels = scales::percent_format(), limits = c(0, NA))

themes

themes

you can basically customize every aspect of the theme in ggplot.

plt <- ggplot(df, aes(x = program, y = admit_rate)) + 
  geom_point(size = 5) + 
  scale_y_continuous(labels = scales::percent_format(), limits = c(0, NA))

plt + theme_classic()

plt + theme_linedraw()

plt + theme_dark()

plt + theme_void()

labels

use the labs() function to set labels for any aesthetics.

ggplot(penguins,
       aes(x = bill_length_mm, y = bill_depth_mm, color = species)) + 
  geom_point() + 
  labs(
    x = "Bill Length [mm]",
    y = "Bill Depth [mm]",
    color = "Species of Penguin",
    title = "Relationship between Bill Length and Bill Depth"
  )

legend position

the legend position can be moved using the legend.position argument to theme()

ggplot(penguins,
       aes(x = bill_length_mm, y = bill_depth_mm, color = species)) + 
  geom_point() + 
  labs(
    x = "Bill Length [mm]",
    y = "Bill Depth [mm]",
    color = "Species of Penguin",
    title = "Relationship between Bill Length and Bill Depth"
  ) + 
  theme(legend.position = 'bottom')

saving plots

use ggsave() to save your plots. keep in mind that the filename saved will be relative to your working directly unless you use an absolute path.

# one option: 
# render your ggplot as normal
ggplot(df, aes(...)) + ...
# then run 
ggsave(filename = "your_filename.png", width = 5, 
  height = 7) # width and height default to inches
  
# another option:
# assign your plot to an object
plt <- ggplot(df, aes(...)) + ...
ggsave(filename = "your_filename.png", 
  plot = plt, # give the plt object to ggsave explicitly
  width = 5, 
  height = 7)

saving plots with ggsave() is one of the most important skills in using ggplot2.

it’s great if you can make your graphics in R, but you need to render your graphics to image files in order to be able to integrate them into manuscripts, websites, etc. and ggsave() gives you lots of great options for the filetype, dimensions, resolution, etc.

extensions

ggdist

learn more: https://mjskay.github.io/ggdist/

ggrepel

learn more: https://ggrepel.slowkow.com/

patchwork

learn more: https://patchwork.data-imaginist.com/

key takeaways

• ggplot2 is based on the grammar of graphics, meaning every plot is made up of:
  • data, aesthetics, geometries, facets, statistics, coordinates, and a theme.
• ggplot2 is “plug and play” (like legos) in the sense that there is a ton of varied geoms, stats, and customizations you can make with interchangeable layers
• it’s easiest to build up your plots incrementally, starting from an overly simple, crude version and working your way up to something more refined; don’t try to make a masterpiece all in one go because it will make debugging your code harder
• if the kind of data visualization you want to make isn’t supported out of the box in ggplot2, likely you can either use or create an extension to ggplot2 that will do what you want.