Data Visualization
data.frame (or tibble) and gets pulled into the ggplot() object.x, y: variablescolour: colors the lines of geometriesfill: fill color of geometriesgroup: groups based on the datashape: shape of point, an integer value 0 to 24, or NAlinetype: type of line, an integer value 0 to 6, or a stringsize: sizes of elements, a non-negative numeric valuealpha: changes the transparency, a numeric value 0 to 1
geom_bar(position = "<position>")dodge, fill, stack (default)jitter
plot.object <- ggplot()
plot.object <- plot.object + geom_*()
plot.object <- plot.object + coord_*() + theme()
plot.object or print(plot.object)
palmerpenguins data set: size measurements for three penguin species observed on three islands in the Palmer Archipelago, Antarctica.species, island, and sex are factor variables.
ggplot2 requires the data as an object of class data.frame or tibble (common in the tidyverse). More complex plots also require the long data frame format.Customizing plots means adjusting elements to improve readability, presentation, and informativeness. Title and axis components can change in size, color, and face:
labs() modifies plot labels, including the x-axis, y-axis, and plot title:
xlab() and ylab() set the x-axis and y-axis labels individually:
The name argument of scale functions also sets axis labels:
expression() allows mathematical expressions, Greek letters, superscripts, and subscripts in labels:
element_text() inside theme() sets text properties such as size, color, and font face:
The face argument can be bold, italic, or bold.italic:
vjust controls the vertical alignment, typically ranging between 0 and 1, but it can extend beyond this range:
Use margin() with parameters t (top) and r (right) to add distance between the axis and its title. For the y-axis, change the right margin, not the bottom margin:
axis.text (and axis.text.x / axis.text.y) modify the appearance of the axis numbers:
angle, hjust, and vjust rotate and position any text element (hjust: 0 = left, 1 = right; vjust: 0 = top, 1 = bottom):
element_blank() removes axis text and ticks entirely:
xlim() and ylim() limit the axis range:
scale_y_continuous(limits = c(0, 20)) or coord_cartesian(ylim = c(0, 20)). The former removes data points outside the range, while the latter zooms without removing data points.
library(showtext)
font_add_google("Playfair Display", "Playfair")
font_add_google("Bangers", "Bangers")
showtext_auto()
p + labs(title = "Relationship between bill length and depth") +
theme(plot.title = element_text(family = "Bangers", hjust = 0.5, size = 25),
plot.subtitle = element_text(family = "Playfair", hjust = 0.5, size = 15))
ggplot2 adds a legend by default when a variable is mapped to an aesthetic. The default legend title is the variable specified in the color argument:
Three equivalent ways to turn the legend off:
theme(legend.title = element_blank()) removes the legend title; setting the name to NULL via scale_color_discrete(name = NULL) or labs(color = NULL) achieves the same:
Three equivalent ways: labs(color = "new title"), scale_color_discrete(name = "new title"), or guides(color = guide_legend("new title")):
A coordinate pair places the legend inside the panel; a transparent background keeps it unobtrusive:
guides(color = guide_legend(direction = "horizontal")) switches the key layout from vertical to horizontal.
Reorder the underlying factor to reorder the legend:
scale_color_discrete(name, labels) renames the keys:
legend.key changes the background boxes; override.aes changes the symbol size:
Elements of a theme, by Isabella Benabaye:
The default theme is theme_gray(), with two arguments for the base font size (base_size, a number) and font family (base_family, a string such as “serif”, “sans”, “mono”):
plot + theme_gray()plot + theme_bw()plot + theme_linedraw()plot + theme_light()plot + theme_dark()plot + theme_minimal()plot + theme_classic()plot + theme_void()The ggthemes package offers additional predefined themes.
theme() has many arguments to modify individual components of a plot, including:
The panel background is the area where the data is plotted. panel.background adjusts its fill and outline:
The panel border is an overlay on top of panel.background that outlines the panel:
panel.grid.major and panel.grid.minor style the two sets of grid lines (panel.grid styles both at once):
panel.grid.major.x, panel.grid.major.y, and their minor counterparts style each axis separately:
p2 + theme(panel.grid.major = element_line(linewidth = .5, linetype = "dashed"),
panel.grid.minor = element_line(linewidth = .25, linetype = "dotted"),
panel.grid.major.x = element_line(color = "red1"),
panel.grid.major.y = element_line(color = "blue1"),
panel.grid.minor.x = element_line(color = "red4"),
panel.grid.minor.y = element_line(color = "blue4"))
element_blank() removes grid lines selectively (panel.grid.minor) or entirely (panel.grid):
color argument defines the outline color and fill the filling color of plot elements:
geom_point(color = "steelblue", size = 2)
scale_color_* and scale_fill_* functions modify colors mapped to variables; they differ for categorical (qualitative) and continuous (quantitative) variables.scale_color_manual() and scale_fill_manual() specify colors by hand:
scale_color_brewer() and scale_fill_brewer() use predefined ColorBrewer palettes (run RColorBrewer::display.brewer.all() to see them all):
The {ggthemes} package gives access to the well-known Tableau palette:
The {ggsci} package provides scientific-journal-themed palettes (such as Science via scale_color_aaas() or Nature via scale_color_npg()):
scale_color_gradient() and scale_fill_gradient() apply a sequential gradient:
scale_color_viridis_c() uses the Viridis palettes, which are perceptually uniform and suitable for colorblind viewers:
scale_color_carto_c() (from rcartocolor) uses CARTO palettes:
The {scico} package provides access to the perceptually uniform palettes by Fabio Crameri:
Have you ever considered how your figure might appear under various forms of colorblindness? The colorBlindness package simulates it:
Viridis works for both discrete and continuous mappings:
Adds horizontal lines at specified y-axis values (yintercept):
Adds vertical lines at specified x-axis values (xintercept), with color, linewidth, and linetype for styling:
Adds line segments that do not span the entire plot range, using xmin/xmax or ymin/ymax:
Adds line segments with arbitrary start and end points (x, xend, y, yend):
geom_encircle() from ggalt automatically encloses points in a polygon, useful for highlighting clusters:
library(ggalt)
p + geom_encircle(data = subset(penguins, species == "Adelie"),
colour = "blue", spread = 0.002) +
geom_encircle(data = subset(penguins, species == "Chinstrap"),
colour = "purple", spread = 0.002) +
geom_encircle(data = subset(penguins, species == "Gentoo"),
colour = "red", spread = 0.002) + ylim(10, 23)
Adds labels with a rectangle around the text; hjust and vjust control the justification:
Like geom_label(), but without the rectangle:
geom_text_repel() and geom_label_repel() from ggrepel repel overlapping labels:
Adds a single text or label annotation at specified coordinates:
Adds custom annotations using grid graphical objects:
library(grid)
my_grob <- grobTree(textGrob("This is species type!", x = .1, y = .9,
hjust = 0, gp = gpar(col = "black", fontsize = 15,
fontface = "bold")))
p + annotation_custom(my_grob) + facet_wrap(~species, scales = "free_x") +
scale_y_continuous(limits = c(NA, 20)) + theme(legend.position = "none")
ggtext renders text as markdown or HTML:
geom_textbox() provides dynamic wrapping for longer annotations:
lab_long <- "**Association**<br><i style='font-size:8pt;color:black;'>This graph is a scatter plot showing the association between bill length and bill depth for each species type, so we can see that there is a clear association.</i>"
p + geom_textbox(aes(x = 45, y = 20, label = lab_long),
width = unit(25, "lines"), stat = "unique")
Flips the x and y coordinates, making horizontal plots vertical and vice versa; particularly useful for bar charts and boxplots:
Fixes the aspect ratio between the units of the x and y axes:
Converts the plot to polar coordinates, often used for circular bar charts:
coord_polar(theta = "y") turns a single stacked bar into a pie chart:
scale_x_continuous() / scale_y_continuous() customize breaks, labels, and limits:
stat_smooth() adds a LOESS fit (method = "loess") for fewer than 1,000 points, or a GAM (method = "gam") otherwise:
It is just as easy to add a standard linear fit:
You can specify the model yourself, for example a polynomial regression:
ncol and nrow control the number of columns and rows in facet_wrap():
Use theme() to customize the appearance of the facet labels:
element_textbox_highlight() from ggtext can highlight specific labels:
library(ggtext)
library(purrr) # for %||%
p + facet_wrap(~ species, ncol = 3, scales = "free_x") +
theme(strip.background = element_blank(),
strip.text = element_textbox_highlight(
family = "Playfair", size = 12, face = "bold",
fill = "white", box.color = "chartreuse4", color = "chartreuse4",
halign = .5, linetype = 1, r = unit(5, "pt"), width = unit(1, "npc"),
padding = margin(5, 0, 3, 0), margin = margin(0, 1, 3, 1),
hi.labels = c("Adelie"), hi.fill = "chartreuse4",
hi.box.col = "black", hi.col = "white"))p1 + p2, p1 / p2, or (g + p2) / p1, and define complex layouts with a design matrix via plot_layout(design = layout).plot_grid(plot_grid(g, p1), p2, ncol = 1).{gridExtra}: arrange plots with grid.arrange(g, p1, p2, layout_matrix = rbind(c(1, 2), c(3, 3))).
p1 <- ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm, colour = species)) +
geom_point()
p2g <- ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm, colour = species)) +
geom_density2d()
p3 <- ggplot(penguins, aes(x = species, fill = island)) +
geom_bar()
p4 <- ggplot(penguins, aes(x = species, y = bill_depth_mm, fill = species)) +
geom_boxplot()
library(patchwork)
p1 + p2g + p3 + p4How many penguins of each species are in this data set?
Number of penguin species on each island:
Body mass by species and sex (stat = "identity" plots the values as given):
A histogram is an accurate graphical representation of the distribution of numeric data. There is only one aesthetic required: the x variable.
Body mass distribution of penguins by species:
Boxplot with annotations via geom_signif() from ggsignif:
A violin plot visualizes the distribution of a numeric variable for one or several groups. It is close to a boxplot but allows a deeper understanding of the distribution.
geom_line() displays values over time and requires a group = aesthetic: use group = 1 for a single line, or group = variable_name to split lines by a variable.
The lines are very close together, so instead of geom_label() we use ggrepel, which gives the labels space and connects them with their lines:
ggsave() saves the last displayed plot (or a named plot object) to disk; the file extension determines the format (.png, .pdf, .svg, .tiff).width, height, and dpi explicitly so figures are reproducible and meet journal or donor requirements: