Data Visualisation Reference
A practical guide to chart types, their purpose, and how to build them in R
2. How to choose the right visualisation
The first question is not “which chart”, it is “what is the message?”
Before picking a chart, write down the one sentence the chart should make true in the reader’s mind. The chart that supports that sentence the most directly is the right chart.
- “Region X grew the fastest” (slope, dumbbell, sorted bar of growth)
- “Most respondents agree” (diverging bar, stacked bar)
- “Y is correlated with X” (scatter with regression line)
- “This site is an outlier” (box plot, beeswarm, control chart)
- “Half the people are stuck at stage 2” (funnel)
- “These five regions account for 80% of cases” (Pareto, ordered bar)
The encoding hierarchy
Cleveland and McGill (1984) ranked how accurately the human eye can decode different visual encodings. Position beats length, length beats angle and area, and colour intensity is the least accurate. Use the most accurate encoding for the most important comparison, this is why bars beat pies and why heatmaps should carry numbers when precision matters.
Decision tree: chart by analytical question
| The reader needs to see, | Reach for, |
|---|---|
| Change over time | Line, area, sparkline |
| Composition over time | Stacked area, streamgraph, 100% stacked bar |
| Compare categories | Bar, lollipop, Cleveland dot plot |
| Compare two points in time | Dumbbell, slope |
| Distribution of one variable | Histogram, density, ECDF |
| Distribution by group | Box plot, violin, ridgeline, beeswarm |
| Relationship between two variables | Scatter, hexbin, regression plot |
| Three variables together | Bubble, coloured scatter |
| Composition of a whole | Bar > treemap > waffle > pie (5 slices max) |
| Hierarchy of composition | Sunburst, treemap |
| Geographic value | Choropleth, proportional symbol map |
| Movement / flow | Sankey, alluvial, flow map, chord |
| Build-up of a total | Waterfall |
Colour and design fundamentals
Three palette families, three purposes
- Sequential (one hue, light to dark): for ordered values, low to high (heat maps, choropleths)
- Diverging (two hues through a neutral midpoint): for values around a meaningful zero (deviations, Likert)
- Categorical (distinct hues): for unordered categories (cap at 7)
Design rules to follow every time
- Start the y-axis at zero when length is the encoding (bars).
- One message per chart. If it needs two sentences to explain, split it.
- Direct labels beat legends when you have 5 series or fewer.
- The title carries the message, axes carry units, the source line carries credibility.
- Annotate the unusual point rather than asking the reader to find it.
- Always show units, time period and source.
- Use colour as a redundant cue, not the only cue (colour-blind readers).
3.1 Time series, change over time
Line chart
- Show how a single quantity evolves on a continuous (usually time) axis.
- Compare the trajectories of several series on the same scale.
library(ggplot2)
df <- data.frame(year = 2010:2024,
value = c(50,51,52,55,57,60,62,63,66,68,70,72,75,76,78))
ggplot(df, aes(year, value)) +
geom_area(fill = "#047B77", alpha = 0.08) +
geom_line(color = "#047B77", linewidth = 1.1) +
geom_point(color = "#047B77", fill = "white", shape = 21, size = 2.5, stroke = 1.4) +
labs(title = "Annual value, 2010-2024",
x = "Year", y = "Value",
caption = "Source: Illustrative") + theme_minimal()library(ggplot2); library(tidyr)
df <- data.frame(year = 2010:2024,
A = c(50,52,53,56,58,60,62,63,66,69,71,73,76,78,80),
B = c(40,42,44,46,48,50,51,53,55,58,60,62,64,66,68),
C = c(30,33,35,38,42,45,48,51,55,57,60,63,66,68,70))
long <- pivot_longer(df, -year, names_to = "series", values_to = "value")
ggplot(long, aes(year, value, color = series)) +
geom_line(linewidth = 1.2) +
geom_point(size = 2, fill = "white", shape = 21, stroke = 1.4) +
scale_color_manual(values = c(A = "#047B77", B = "#1A9490", C = "#F59E0B")) +
labs(title = "Comparing three series over time",
x = "Year", y = "Index") + theme_minimal()Area chart
- A line chart with the area under it filled, emphasising volume or cumulative magnitude.
- Show how a total and the contribution of each component to that total evolve over time.
library(ggplot2)
df <- data.frame(year = 2010:2024,
value = c(50,52,53,55,57,58,60,61,63,65,68,70,72,75,78))
ggplot(df, aes(year, value)) +
geom_area(fill = "#047B77", alpha = 0.3) +
geom_line(color = "#047B77", linewidth = 1.5) +
labs(title = "Area chart", x = "Year", y = "Value") + theme_minimal()library(ggplot2); library(tidyr)
df <- data.frame(year = 2010:2024,
Solar = c(2,3,5,7,9,11,14,17,21,25,29,33,38,42,46),
Wind = c(8,9,11,12,14,16,18,20,22,25,28,30,33,35,38),
Hydro = c(20,21,21,22,22,23,23,24,24,25,25,26,26,27,27))
long <- pivot_longer(df, -year, names_to = "source", values_to = "twh")
ggplot(long, aes(year, twh, fill = source)) +
geom_area(alpha = 0.92, color = "white") +
scale_fill_manual(values = c(Solar = "#047B77", Wind = "#1A9490", Hydro = "#F59E0B")) +
labs(title = "Stacked area", x = "Year", y = "Generation (TWh)") + theme_minimal()Slope chart
- Two time points, several series, the slope of each line is the rate of change.
- Use, before / after across 3 to 10 categories (KPIs, sub-regions, indicators).
- Avoid when: More than 10 lines cross, switch to dumbbell or small multiples.
library(ggplot2); library(dplyr); library(tidyr)
df <- data.frame(region = c("East","West","Middle","North","South"),
y2018 = c(24,18,20,32,28),
y2024 = c(38,27,30,45,36))
long <- df |> pivot_longer(-region, names_to = "year", values_to = "value") |>
mutate(x = if_else(year == "y2018", 0, 1))
ggplot(long, aes(x, value, group = region, color = region)) +
geom_line(linewidth = 1.6) +
geom_point(size = 4, fill = "white", shape = 21, stroke = 2) +
geom_text(data = filter(long, x == 0),
aes(label = paste0(region, ": ", value, "%")),
hjust = 1, nudge_x = -0.04, fontface = "bold") +
geom_text(data = filter(long, x == 1),
aes(label = paste0(region, ": ", value, "%")),
hjust = 0, nudge_x = 0.04, fontface = "bold") +
scale_x_continuous(breaks = c(0,1), labels = c("2018","2024"),
limits = c(-0.6, 1.6)) +
labs(title = "Slope chart", y = "Value (%)") +
theme_minimal() + theme(legend.position = "none")Dumbbell chart
Show distance and direction of change between two states for several categories. Both endpoints and the change are visible at once.
Can be used for any from-to comparison across categories.
library(ggplot2)
df <- data.frame(region = factor(c("East","West","Middle","North","South"),
levels = c("East","West","Middle","North","South")),
y2018 = c(24,18,20,32,28),
y2024 = c(38,27,30,45,36))
df$delta <- df$y2024 - df$y2018
df$mid <- (df$y2018 + df$y2024) / 2
ggplot(df, aes(y = region)) +
geom_segment(aes(x = y2018, xend = y2024, yend = region),
color = "#94a3b8", linewidth = 2, alpha = 0.5) +
geom_point(aes(x = y2018, color = "2018"), size = 6) +
geom_point(aes(x = y2024, color = "2024"), size = 6.5) +
geom_label(aes(x = mid, label = paste0("+", delta, " pp")),
vjust = -0.9, color = "#1A9490", fontface = "bold", fill = "white") +
scale_color_manual(values = c("2018" = "#94a3b8", "2024" = "#047B77")) +
scale_y_discrete(limits = rev(levels(df$region))) +
labs(title = "Dumbbell chart", x = "Value (%)", y = NULL) +
theme_minimal()3.2 Categorical comparison
Bar chart
- Compare a single numeric variable across discrete categories using length on a common axis (the most accurate visual encoding). Any “compare values across categories” question.
- Mainly used 6+ categories, long category names, ranked or sorted lists.
- Not use it when: The x-axis is continuous (histogram), or labels are too long (use horizontal bar).
library(ggplot2)
df <- data.frame(sector = c("Health","Education","Defence","Energy","Transport","Other"),
share = c(42, 38, 27, 19, 16, 14))
ggplot(df, aes(reorder(sector, -share), share)) +
geom_col(fill = "#047B77", width = 0.7) +
geom_text(aes(label = share), vjust = -0.4, fontface = "bold") +
labs(title = "Bar chart", x = NULL, y = "Allocation (%)") +
theme_minimal()library(ggplot2)
df <- data.frame(sector = c("Health","Education","Defence","Energy","Transport","Other"),
share = c(42, 38, 27, 19, 16, 14))
ggplot(df, aes(share, reorder(sector, share))) +
geom_col(fill = "#047B77") +
geom_text(aes(label = share), hjust = -0.2, fontface = "bold") +
scale_x_continuous(expand = expansion(mult = c(0, 0.08))) +
labs(title = "Horizontal bar", x = "Value", y = NULL) +
theme_minimal()Grouped bar chart
Compare values across categories and sub-groups simultaneously, by placing sub-group bars side-by-side.
Suitable for 4 sub-groups or fewer, 6 categories or fewer.
- But to compare totals, use stacked bar instead.
library(ggplot2); library(tidyr)
df <- data.frame(year = factor(c("2018","2021","2024")),
A = c(35,42,48), B = c(28,33,39), C = c(21,25,30))
long <- pivot_longer(df, -year, names_to = "region", values_to = "value")
ggplot(long, aes(year, value, fill = region)) +
geom_col(position = position_dodge(0.7), width = 0.6) +
scale_fill_manual(values = c(A = "#047B77", B = "#1A9490", C = "#F59E0B")) +
labs(title = "Grouped bar") + theme_minimal()Stacked bar chart
- Show the total for each group and the breakdown of that total into components.
- Good when totals matter and parts sum to something meaningful.
- Normalise each bar to 100% so the reader compares proportions, not totals.
library(ggplot2); library(tidyr)
df <- data.frame(year = factor(c("2018","2021","2024")),
A = c(35,42,48), B = c(28,33,39), C = c(21,25,30))
long <- pivot_longer(df, -year, names_to = "region", values_to = "value")
ggplot(long, aes(year, value, fill = region)) +
geom_col() +
scale_fill_manual(values = c(A = "#047B77", B = "#1A9490", C = "#F59E0B")) +
labs(title = "Stacked bar") + theme_minimal()library(ggplot2); library(dplyr); library(tidyr)
df <- data.frame(year = factor(c("2018","2021","2024")),
A = c(35,42,48), B = c(28,33,39), C = c(21,25,30))
long <- pivot_longer(df, -year, names_to = "region", values_to = "value") |>
group_by(year) |> mutate(share = value / sum(value) * 100)
ggplot(long, aes(year, share, fill = region)) +
geom_col(position = "stack") +
scale_fill_manual(values = c(A = "#047B77", B = "#1A9490", C = "#F59E0B")) +
labs(title = "100% stacked bar", y = "Share (%)") + theme_minimal()Lollipop chart Or Cleveland dot plot
- A bar chart with the heavy fill replaced by a thin line and a dot, reduces ink and helps long lists breathe.
- 8 to 30 categories, sorted by value.
library(ggplot2)
df <- data.frame(sector = c("Health","Education","Defence","Energy","Transport","Other"),
share = c(42, 38, 27, 19, 16, 14))
df$sector <- reorder(df$sector, df$share)
ggplot(df, aes(share, sector)) +
geom_segment(aes(x = 0, xend = share, yend = sector),
color = "#94a3b8", linewidth = 1.4) +
geom_point(color = "#047B77", size = 6) +
geom_text(aes(label = share), hjust = -0.7,
color = "#047B77", fontface = "bold") +
labs(title = "Lollipop", x = "Value", y = NULL) +
theme_minimal()library(ggplot2)
df <- data.frame(country = paste("Country", LETTERS[1:6]),
value = c(62, 41, 58, 33, 49, 27))
df$country <- reorder(df$country, df$value)
ggplot(df, aes(value, country)) +
geom_segment(aes(x = 0, xend = value, yend = country),
color = "#94a3b8", alpha = 0.7) +
geom_point(color = "#047B77", size = 5) +
labs(title = "Cleveland dot plot", x = "Value", y = NULL) +
theme_minimal()Diverging bar chart
Encode both direction and magnitude around a meaningful zero, useful for Likert surveys, profit-loss, deviations from target.
When to use: Survey results (agree / disagree), signed values, year-on-year change.
But not usefull if the data has no natural zero.
library(ggplot2)
df <- data.frame(
response = factor(c("Strongly disagree","Disagree","Neutral",
"Agree","Strongly agree"),
levels = c("Strongly disagree","Disagree","Neutral",
"Agree","Strongly agree")),
share = c(-12, -18, 10, 32, 28))
ggplot(df, aes(share, response, fill = share > 0)) +
geom_col() +
geom_vline(xintercept = 0, color = "#475569", linewidth = 0.6) +
scale_fill_manual(values = c("TRUE" = "#047B77", "FALSE" = "#EF4444"),
guide = "none") +
labs(title = "Diverging bar", x = "Share (%)", y = NULL) +
theme_minimal()Pareto chart
- Bars (descending) combined with a cumulative line, shows the “vital few” categories that account for 80% of the value.
library(ggplot2); library(dplyr)
df <- data.frame(cat = LETTERS[1:8],
v = c(20, 18, 12, 9, 7, 5, 3, 2)) |>
arrange(desc(v)) |>
mutate(cum_pct = cumsum(v) / sum(v) * 100)
ggplot(df, aes(reorder(cat, -v))) +
geom_col(aes(y = v), fill = "#047B77") +
geom_line(aes(y = cum_pct, group = 1), color = "#F59E0B", linewidth = 1.2) +
geom_point(aes(y = cum_pct), color = "#F59E0B", size = 3) +
geom_hline(yintercept = 80, linetype = "dashed", color = "#475569") +
scale_y_continuous(sec.axis = sec_axis(~ ., name = "Cumulative %")) +
labs(title = "Pareto chart", x = "Category", y = "Frequency") +
theme_minimal()3.3 Distribution
Histogram
- Show the shape of one numeric variable, where mass sits, spread, skew, outliers, multi-modality.
- A smoothed version of the histogram (KDE). Easier to overlay multiple groups.
- Used for comparing the shapes of 2 or 3 distributions on the same axis.
library(ggplot2)
set.seed(7); df <- data.frame(height = rnorm(500, 170, 9))
ggplot(df, aes(height)) +
geom_histogram(bins = 22, fill = "#047B77", color = "white") +
geom_vline(xintercept = mean(df$height), color = "#F59E0B",
linewidth = 1.1, linetype = "dashed") +
labs(title = "Histogram", x = "Height (cm)", y = "Frequency") +
theme_minimal()library(ggplot2)
set.seed(7)
df <- rbind(data.frame(group = "A", x = rnorm(500, 170, 9)),
data.frame(group = "B", x = rnorm(500, 178, 7)))
ggplot(df, aes(x, fill = group, color = group)) +
geom_density(alpha = 0.35, linewidth = 1.2) +
scale_fill_manual(values = c(A = "#047B77", B = "#1A9490")) +
scale_color_manual(values = c(A = "#047B77", B = "#1A9490")) +
labs(title = "Density plot", x = "Height (cm)", y = "Density") +
theme_minimal()Box plot
Compact summary of a distribution: median, inter-quartile box, whiskers, individual outliers.
Comparing distributions across many groups, spotting outliers.
A box plot plus a mirrored density, reveals bimodality and skew that a box plot hides.
Used when non-Gaussian distributions, multi-modal data.
library(ggplot2)
set.seed(7)
df <- rbind(
data.frame(group = "A", x = rnorm(120, 50, 10)),
data.frame(group = "B", x = rnorm(120, 55, 8 )),
data.frame(group = "C", x = rnorm(120, 48, 14)),
data.frame(group = "D", x = rnorm(120, 60, 7 )))
ggplot(df, aes(group, x)) +
geom_boxplot(fill = "#CDEAE8", color = "#047B77",
outlier.color = "#1A9490", width = 0.55) +
labs(title = "Box plot", x = "Group", y = "Score") +
theme_minimal()library(ggplot2)
set.seed(7)
df <- rbind(
data.frame(group = "A", x = rnorm(120, 50, 10)),
data.frame(group = "B", x = rnorm(120, 55, 8 )),
data.frame(group = "C", x = rnorm(120, 48, 14)),
data.frame(group = "D", x = rnorm(120, 60, 7 )))
ggplot(df, aes(group, x)) +
geom_violin(fill = "#CDEAE8", color = "#047B77") +
geom_boxplot(width = 0.12, fill = "white", color = "#047B77") +
labs(title = "Violin plot", x = "Group", y = "Score") +
theme_minimal()QQ plot
Compare a sample’s quantiles to a theoretical distribution (usually normal). Points on the diagonal = good fit.
Used for checking the normality assumption before applying a parametric test.
library(ggplot2)
set.seed(7); df <- data.frame(x = rnorm(200) + 0.4 * rnorm(200)^3)
ggplot(df, aes(sample = x)) +
stat_qq(color = "#047B77") +
stat_qq_line(color = "#F59E0B", linewidth = 1.2) +
labs(title = "QQ plot",
x = "Theoretical quantiles", y = "Sample quantiles") +
theme_minimal()3.4 Relationship and correlation
Scatter plot
Plot each observation as a point on two numeric axes, to see the relationship: linear, non-linear, no relationship, clusters, outliers.
Used when your first question “do X and Y move together?” or “are there clusters?”.
library(ggplot2)
set.seed(7)
df <- data.frame(x = rnorm(200, 50, 10))
df$y <- 0.55 * df$x + rnorm(200, 0, 5) + 10
ggplot(df, aes(x, y)) +
geom_point(color = "#047B77", alpha = 0.55, size = 2.5) +
geom_smooth(method = "lm", color = "#F59E0B", se = FALSE, linewidth = 1.1) +
labs(title = "Scatter plot", x = "Variable X", y = "Variable Y") +
theme_minimal()Correlation heat map
- Show the correlation matrix of many numeric variables at once. Diverging palette (red to blue) makes positive vs negative correlations pop.
library(ggplot2); library(reshape2)
m <- cor(mtcars)
df <- melt(m); names(df) <- c("v1","v2","cor")
ggplot(df, aes(v1, v2, fill = cor)) +
geom_tile(color = "white") +
geom_text(aes(label = sprintf("%.2f", cor)), size = 3) +
scale_fill_gradient2(low = "#EF4444", mid = "white", high = "#047B77",
midpoint = 0, limits = c(-1, 1)) +
labs(title = "Correlation heat map") + theme_minimal() +
theme(axis.text.x = element_text(angle = 45, hjust = 1))Pair plot (scatterplot matrix)
- A matrix of pairwise scatter plots between every pair of numeric variables, with histograms on the diagonal.
# install.packages("GGally")
library(GGally)
ggpairs(iris, columns = 1:4, aes(color = Species, alpha = 0.6)) +
ggplot2::labs(title = "Pair plot")3.5 Composition, part-to-whole
Pie chart
Show how a single whole is split into parts.
When to use: 5 categories or fewer, parts sum to 100%, audience expects a pie.
A pie chart with the centre removed. The hole can carry a total or a KPI.
Same rules as a pie, plus the centre is useful for a total or label.
library(ggplot2)
df <- data.frame(region = c("North","South","East","West"),
share = c(42, 25, 18, 15))
ggplot(df, aes(x = "", y = share, fill = region)) +
geom_col(width = 1, color = "white") + coord_polar(theta = "y") +
scale_fill_manual(values = c("#047B77","#1A9490","#0EA5E9","#F59E0B")) +
theme_void() + labs(title = "Pie chart")library(ggplot2)
df <- data.frame(region = c("North","South","East","West"),
share = c(42, 25, 18, 15))
ggplot(df, aes(x = 2, y = share, fill = region)) +
geom_col(width = 1, color = "white") + coord_polar(theta = "y") +
xlim(0.2, 2.5) +
scale_fill_manual(values = c("#047B77","#1A9490","#0EA5E9","#F59E0B")) +
theme_void() + labs(title = "Donut chart")