---
title: "Customizing ggplot2 Visualizations"
---
# Customizing ggplot2 Visualizations {#ggplot2customize}
```{r}
library(tidyverse)
library(plotly)
knitr::opts_chunk$set(echo = TRUE, warning = FALSE, message = FALSE)
```
The previous chapter covered the mechanics of building charts. This one covers customization — taking a chart from "communicates the data" to "publication-ready." The default `ggplot2` appearance is fine for exploratory work but rarely the right choice for slide decks, client reports, or anything that will be seen outside your console.
::: {.callout-warning}
## AI Pitfall: AI generates over-decorated charts that obscure the data
Ask an AI to "make this chart look professional" and you may get back code that adds drop shadows, gradients, custom backgrounds, three different fonts, and a logo. Each addition is technically possible. The cumulative effect is a chart that fights its own data for attention.
Edward Tufte's principle of the data-ink ratio still applies: maximize the ink that represents data, minimize the ink that does not. AI is enthusiastic about adding visual elements; it is not opinionated about removing them. When an AI-generated chart starts to feel busy, that is usually a sign to remove decoration, not add more.
Three concrete checks for any "polished" chart before it ships:
- Does every visual element (gridline, color, axis tick) help the reader interpret the data, or is it just decoration?
- Are the colors meaningful (encoding categories or values) or just decorative?
- If you removed everything that is not data, would the chart still tell its story? If yes, the rest is probably overdecoration.
:::
We will cover:
- **Themes** -- swap out the entire visual style in one line
- **Colors** -- use palettes that look polished and are accessible
- **Axis formatting** -- dollar signs, commas, percentages on your axes
- **Legends** -- move them, restyle them, or get rid of them
- **Annotations** -- draw attention to the things that matter
- **Combining plots** -- put multiple charts together for a complete story
We will use the built-in `mpg`, `iris`, and `diamonds` datasets throughout. This chapter also uses three additional packages --- **patchwork** (combining plots), **ggthemes** (publication-style themes), and **ggrepel** (non-overlapping labels). Install them once if you haven't already:
```{r}
#| eval: false
install.packages(c("patchwork", "ggthemes", "ggrepel"))
```
## Themes -- One-Line Style Upgrades
### Built-in Themes
Every ggplot2 plot uses a theme to control non-data elements like backgrounds, gridlines, and fonts. The default is `theme_gray()`. Here are the ones worth knowing:
```{r}
base_plot <- ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(aes(color = class)) +
labs(title = "Engine Displacement vs. Highway MPG")
library(patchwork)
p1 <- base_plot + theme_gray() + ggtitle("theme_gray() (default)")
p2 <- base_plot + theme_bw() + ggtitle("theme_bw()")
p3 <- base_plot + theme_minimal() + ggtitle("theme_minimal()")
p4 <- base_plot + theme_classic() + ggtitle("theme_classic()")
p5 <- base_plot + theme_light() + ggtitle("theme_light()")
(p1 | p2 | p3) / (p4 | p5)
```
Here is the quick rundown:
- **`theme_bw()`** -- Removes the gray background. Clean, great for printed reports.
- **`theme_minimal()`** -- Even cleaner. This is probably the most popular choice for business presentations.
- **`theme_classic()`** -- Looks like a textbook chart. Axis lines, no gridlines.
- **`theme_light()`** -- Light gray gridlines. A nice middle ground.
For most business work, `theme_minimal()` or `theme_bw()` will serve you well.
### Themes from ggthemes
The **ggthemes** package provides themes inspired by well-known publications. If you want your chart to look like it belongs in The Economist or the Wall Street Journal:
```{r}
library(ggthemes)
t1 <- base_plot + theme_economist() + ggtitle("theme_economist()")
t2 <- base_plot + theme_fivethirtyeight() + ggtitle("theme_fivethirtyeight()")
t3 <- base_plot + theme_tufte() + ggtitle("theme_tufte()")
t4 <- base_plot + theme_wsj() + ggtitle("theme_wsj()")
(t1 | t2) / (t3 | t4)
```
Fun to play with, and occasionally the right choice for a specific audience.
### Setting a Global Theme
Tired of adding `+ theme_minimal()` to every plot? Set it once at the top of your script:
```{r}
theme_set(theme_minimal())
# Now every plot uses theme_minimal() automatically
ggplot(mpg, aes(x = class)) +
geom_bar(fill = "steelblue") +
labs(title = "Vehicle Class Counts", x = "Class", y = "Count")
```
```{r}
# Reset to default for the rest of the chapter
theme_set(theme_gray())
```
## Fine-Tuning with theme()
Sometimes you want a built-in theme but with a few tweaks. The `theme()` function lets you modify individual elements using four helpers:
| Helper Function | Controls | Key Arguments |
|:---|:---|:---|
| `element_text()` | Text | `family`, `face`, `size`, `color`, `angle` |
| `element_line()` | Lines | `color`, `linewidth`, `linetype` |
| `element_rect()` | Rectangles/borders | `fill`, `color` |
| `element_blank()` | Removes the element | (none) |
### Example: Making Titles Stand Out
```{r}
ggplot(mpg, aes(x = displ, y = hwy, color = class)) +
geom_point(size = 2) +
labs(
title = "Fuel Efficiency by Engine Size",
subtitle = "Larger engines tend to have lower highway mileage",
caption = "Source: EPA fuel economy data (mpg dataset)",
x = "Engine Displacement (liters)",
y = "Highway MPG",
color = "Vehicle Class"
) +
theme(
plot.title = element_text(face = "bold", size = 16, color = "navy"),
plot.subtitle = element_text(face = "italic", size = 12, color = "gray40"),
plot.caption = element_text(size = 9, hjust = 0, color = "gray50"),
axis.title = element_text(face = "bold", size = 12)
)
```
### Example: Cleaning Up the Background
```{r}
ggplot(mpg, aes(x = cty, y = hwy)) +
geom_point(alpha = 0.5, color = "darkorange") +
labs(title = "City vs. Highway Fuel Economy") +
theme(
panel.background = element_rect(fill = "white"),
plot.background = element_rect(fill = "gray95"),
panel.grid.major = element_line(color = "gray80", linewidth = 0.5),
panel.grid.minor = element_blank()
)
```
### Example: Rotating Long Axis Labels
When your category labels overlap, angle them:
```{r}
ggplot(mpg, aes(x = manufacturer)) +
geom_bar(fill = "coral") +
labs(title = "Vehicles by Manufacturer", x = NULL, y = "Count") +
theme(
axis.text.x = element_text(angle = 45, hjust = 1, size = 9),
plot.title = element_text(face = "bold", size = 14)
)
```
## Colors That Look Good
Color is one of the most impactful things you can change. ggplot2 gives you several ways to control it.
### Manual Colors -- Pick Your Own
Use `scale_color_manual()` (for points and lines) or `scale_fill_manual()` (for bars and areas) to set exact colors:
```{r}
ggplot(mpg, aes(x = displ, y = hwy, color = drv)) +
geom_point(size = 2) +
scale_color_manual(
values = c("f" = "#E41A1C", "r" = "#377EB8", "4" = "#4DAF4A"),
labels = c("f" = "Front", "r" = "Rear", "4" = "Four-Wheel")
) +
labs(title = "MPG by Engine Size and Drive Type", color = "Drive Type")
```
This is great when your company has brand colors or when you want precise control.
### ColorBrewer -- Pre-Made Palettes That Work
ColorBrewer palettes were designed by a cartographer to be visually clear and print-friendly. They are excellent defaults:
```{r}
RColorBrewer::display.brewer.all()
```
Three categories:
- **Sequential** (Blues, Greens, OrRd) -- for ordered data, low to high
- **Diverging** (RdBu, RdYlGn) -- for data with a meaningful midpoint
- **Qualitative** (Set1, Set2, Dark2) -- for unordered categories
```{r}
ggplot(mpg, aes(x = class, fill = class)) +
geom_bar() +
scale_fill_brewer(palette = "Set2") +
labs(title = "Vehicle Counts by Class (Set2 palette)") +
theme(legend.position = "none")
```
```{r}
ggplot(iris, aes(x = Sepal.Length, y = Sepal.Width, color = Species)) +
geom_point(size = 2.5) +
scale_color_brewer(palette = "Dark2") +
labs(title = "Iris Measurements (Dark2 palette)")
```
### Viridis -- Colorblind-Friendly and Beautiful
The viridis palettes are perceptually uniform and work for people with color vision deficiencies. They are a great default choice, especially for continuous data:
```{r}
diamonds_sample <- diamonds[sample(nrow(diamonds), 2000), ]
v1 <- ggplot(diamonds_sample, aes(x = carat, y = price, color = depth)) +
geom_point(alpha = 0.6) +
scale_color_viridis_c(option = "viridis") +
labs(title = "viridis")
v2 <- ggplot(diamonds_sample, aes(x = carat, y = price, color = depth)) +
geom_point(alpha = 0.6) +
scale_color_viridis_c(option = "magma") +
labs(title = "magma")
v3 <- ggplot(diamonds_sample, aes(x = carat, y = price, color = depth)) +
geom_point(alpha = 0.6) +
scale_color_viridis_c(option = "plasma") +
labs(title = "plasma")
v4 <- ggplot(diamonds_sample, aes(x = carat, y = price, color = depth)) +
geom_point(alpha = 0.6) +
scale_color_viridis_c(option = "inferno") +
labs(title = "inferno")
(v1 | v2) / (v3 | v4)
```
All four palettes go from dark to light, but the color journeys differ. **Viridis** (the default) runs from purple through green to yellow --- the best all-rounder. **Magma** and **inferno** go through warm reds and oranges, making high values "pop" more dramatically. **Plasma** takes a purple-to-yellow path through pinks. All four are designed to be readable even when printed in black and white and are safe for people with color vision deficiencies. When in doubt, stick with the default viridis.
For discrete (categorical) data, use `scale_color_viridis_d()` or `scale_fill_viridis_d()`:
```{r}
ggplot(mpg, aes(x = class, fill = class)) +
geom_bar() +
scale_fill_viridis_d(option = "plasma") +
labs(title = "Vehicle Class with Viridis Discrete (Plasma)") +
theme(legend.position = "none")
```
## Axis Formatting -- Dollars, Commas, and Percentages
Nothing says "this chart is for business people" like properly formatted axis labels.
### The scales Package
The **scales** package gives you label formatters that handle the common cases:
```{r}
diamonds_summary <- diamonds %>%
group_by(cut) %>%
summarize(avg_price = mean(price), count = n())
ggplot(diamonds_summary, aes(x = cut, y = avg_price)) +
geom_col(fill = "darkgreen") +
scale_y_continuous(labels = scales::dollar) +
labs(title = "Average Diamond Price by Cut", x = "Cut Quality", y = "Average Price")
```
```{r}
ggplot(diamonds_summary, aes(x = cut, y = count)) +
geom_col(fill = "steelblue") +
scale_y_continuous(labels = scales::comma) +
labs(title = "Diamond Count by Cut", x = "Cut Quality", y = "Number of Diamonds")
```
```{r}
mpg_pct <- mpg %>%
count(class) %>%
mutate(pct = n / sum(n))
ggplot(mpg_pct, aes(x = reorder(class, pct), y = pct)) +
geom_col(fill = "tomato") +
scale_y_continuous(labels = scales::percent) +
coord_flip() +
labs(title = "Percentage of Vehicles by Class", x = NULL, y = "Percentage")
```
### Log Scales
When your data spans several orders of magnitude (like revenue data that goes from $100 to $100M), log scales help:
```{r}
p_log <- ggplot(diamonds, aes(x = carat, y = price)) +
geom_point(alpha = 0.05, color = "purple") +
scale_x_log10() +
scale_y_log10(labels = scales::dollar) +
labs(title = "Diamond Price vs. Carat (Log Scale)", x = "Carat (log)", y = "Price (log)")
ggplotly(p_log)
```
On the original linear axes, the diamond data was a curved blob. On log scales, the relationship between carat and price becomes nearly linear, which means the relationship is *multiplicative* --- doubling the carat weight roughly multiplies the price by a fixed factor rather than adding a fixed dollar amount. Log scales are your friend whenever data spans orders of magnitude or when exponential/power-law relationships are at play.
### Zooming vs. Filtering -- An Important Distinction
There are two ways to limit your axes, and they behave very differently:
- **`coord_cartesian()`** zooms in *without removing data*. Trend lines and statistics stay accurate.
- **`scale_*_continuous(limits = ...)`** *removes data* outside the range, which can change trend lines.
```{r}
p_base <- ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point() +
geom_smooth(method = "lm")
p_scale <- p_base +
scale_x_continuous(limits = c(3, 6)) +
ggtitle("scale limits (data removed)")
p_coord <- p_base +
coord_cartesian(xlim = c(3, 6)) +
ggtitle("coord_cartesian (zoom only)")
p_scale | p_coord
```
See how the regression lines differ? The left panel's line is steeper because it was fitted only to the visible data (3--6 liters), ignoring the smaller and larger engines. The right panel's line was fitted to *all* the data and then the view was zoomed in. This distinction matters when you are showing a zoomed-in portion of a chart in a presentation --- use `coord_cartesian()` so your trend lines and statistics reflect the full picture, not just the window.
## Legend Customization
### Moving the Legend Around
```{r}
base <- ggplot(mpg, aes(x = displ, y = hwy, color = class)) +
geom_point()
l1 <- base + theme(legend.position = "top") + ggtitle("Top")
l2 <- base + theme(legend.position = "bottom") + ggtitle("Bottom")
l3 <- base + theme(legend.position = "left") + ggtitle("Left")
l4 <- base + theme(legend.position = "right") + ggtitle("Right (default)")
(l1 | l2) / (l3 | l4)
```
### Removing the Legend
When the information is obvious from context (like a bar chart where color matches the x-axis labels), just remove it:
```{r}
ggplot(mpg, aes(x = class, fill = class)) +
geom_bar() +
theme(legend.position = "none") +
labs(title = "Bar Chart Without Legend")
```
### Customizing Legend Title and Labels
```{r}
p_custom_legend <- ggplot(iris, aes(x = Sepal.Length, y = Petal.Length, color = Species)) +
geom_point(size = 2) +
scale_color_manual(
values = c("setosa" = "#1b9e77", "versicolor" = "#d95f02", "virginica" = "#7570b3"),
labels = c("Setosa", "Versicolor", "Virginica")
) +
labs(color = "Iris Species", title = "Custom Legend Title and Labels")
ggplotly(p_custom_legend)
```
## Annotations -- Drawing Attention Where It Matters
### annotate()
Add text, shapes, or arrows to specific spots on your chart:
```{r}
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(color = "gray50") +
annotate("text", x = 6, y = 40, label = "Efficient outliers",
color = "red", fontface = "bold", size = 5) +
annotate("rect", xmin = 5.5, xmax = 7, ymin = 25, ymax = 45,
alpha = 0.1, fill = "red", color = "red", linetype = "dashed") +
labs(title = "Using annotate() for Text and Shapes")
```
### geom_text() and geom_label() for Data-Driven Labels
```{r}
top_classes <- mpg %>%
count(class) %>%
arrange(desc(n)) %>%
head(5)
ggplot(top_classes, aes(x = reorder(class, n), y = n)) +
geom_col(fill = "steelblue") +
geom_text(aes(label = n), hjust = -0.3, size = 5, fontface = "bold") +
coord_flip() +
scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
labs(title = "Top 5 Vehicle Classes", x = NULL, y = "Count")
```
### ggrepel for Non-Overlapping Labels
When labels pile on top of each other, the **ggrepel** package automatically spaces them out:
```{r}
library(ggrepel)
best_mpg <- mpg %>%
group_by(manufacturer) %>%
slice_max(hwy, n = 1) %>%
ungroup() %>%
slice_max(hwy, n = 8)
ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(color = "gray70") +
geom_point(data = best_mpg, color = "red", size = 3) +
geom_label_repel(
data = best_mpg,
aes(label = paste(manufacturer, model)),
size = 3,
max.overlaps = 20,
fill = "lightyellow"
) +
labs(
title = "Most Fuel-Efficient Vehicles by Manufacturer",
subtitle = "Top 8 models highlighted",
x = "Engine Displacement (L)",
y = "Highway MPG"
)
```
## Combining Plots with patchwork
In real life, you often need to present multiple charts together -- a scatter plot next to a bar chart, or a grid of related views. The **patchwork** package makes this easy with intuitive operators:
- `|` places plots side by side
- `/` stacks plots vertically
- `()` groups plots to control layout order
```{r}
p_scatter <- ggplot(mpg, aes(x = displ, y = hwy)) +
geom_point(color = "steelblue") +
labs(title = "Scatter Plot")
p_box <- ggplot(mpg, aes(x = drv, y = hwy, fill = drv)) +
geom_boxplot(show.legend = FALSE) +
labs(title = "Box Plot")
p_bar <- ggplot(mpg, aes(x = class)) +
geom_bar(fill = "coral") +
labs(title = "Bar Chart") +
theme(axis.text.x = element_text(angle = 45, hjust = 1))
p_scatter | p_box | p_bar
```
### Controlling Layout
```{r}
(p_scatter | p_box) / p_bar +
plot_layout(heights = c(2, 1))
```
### Adding a Shared Title and Tags
Tags automatically label your panels A, B, C -- perfect for reports and papers:
```{r}
(p_scatter | p_box) / p_bar +
plot_annotation(
title = "Overview of the mpg Dataset",
subtitle = "Three complementary views of the data",
caption = "Source: EPA fuel economy data",
tag_levels = "A"
) +
plot_layout(heights = c(2, 1))
```
### Sharing a Legend Across Plots
When multiple plots use the same color scheme, collect the legends into one:
```{r}
p_a <- ggplot(mpg, aes(x = displ, y = hwy, color = drv)) +
geom_point()
p_b <- ggplot(mpg, aes(x = cty, y = hwy, color = drv)) +
geom_point()
p_a + p_b +
plot_layout(guides = "collect") +
plot_annotation(title = "Shared Legend Between Plots")
```
## Summary
Here is your customization toolkit at a glance:
1. **Themes** -- Use `theme_minimal()` or `theme_bw()` for clean, professional charts. Try `ggthemes` for publication-specific styles. Set a global default with `theme_set()`.
2. **theme() tweaks** -- Fine-tune titles, axes, gridlines, and backgrounds with `element_text()`, `element_line()`, `element_rect()`, and `element_blank()`.
3. **Colors** -- Use `scale_color_brewer()` for categorical data, `scale_color_viridis_c()` for continuous data, and `scale_color_manual()` when you need exact colors.
4. **Axis formatting** -- Use the `scales` package for `dollar`, `comma`, and `percent` labels. Use `coord_cartesian()` to zoom without distorting statistics.
5. **Legends** -- Move with `theme(legend.position = ...)`. Remove with `"none"`. Customize labels in the scale function.
6. **Annotations** -- Use `annotate()` for fixed labels, `geom_text()` for data-driven labels, and `ggrepel` to prevent overlaps.
7. **Multi-panel figures** -- Use patchwork with `|`, `/`, `plot_layout()`, and `plot_annotation()` to combine charts into a cohesive story.
With these tools, you can take any default ggplot2 chart and turn it into something you would be proud to put in front of a client, a manager, or an audience of 500 people.
