This entry is part of the Chart Index, the reference library for the Chart Design Field Guide.

A choropleth map encodes a quantitative variable as the colour of geographic regions. Every county, state, or country becomes a shape; the colour of the shape is the value. The form is intuitive (most readers have used coloured maps since school) and has been the default for spatial data since Charles Dupin shaded the départements of France by literacy in 1826.

It is also one of the most-misused forms in modern data visualisation. The pathologies — encoding count instead of rate, using rainbow palettes, letting large sparsely-populated regions dominate the visual weight — are well documented and still everywhere. A clean choropleth is a careful choropleth.

What it is

A choropleth maps geographic regions to filled polygons (using a map projection) and a quantitative variable to colour. Each region is shaded based on its data value; a colour-scale legend translates colour back to value. The reading task is spatial-pattern recognition: where are the high regions, where are the low, what is the geographic structure?

The hero chart above shades the eight Australian states and territories by the share of dwellings with rooftop solar. The pattern is the point: the sunny, detached-housing states of the north and west are dark, the cooler south-east is light, and Tasmania is lightest of all. The quantised legend at the bottom left turns any shade back into a number. Randomise the values and the fills snap to their new classes — and notice that the ramp itself re-fits to the new range, which is the first thing to distrust in any choropleth.

When to use it

Choropleths are the right choice when:

  • The data is geographic and ratio-based — rate per capita, percentage, density, average.
  • The reader's question is "where is this high, where is this low, what is the spatial pattern?"
  • The regions have comparable underlying populations or the value is already normalised.
  • The geographic boundaries are meaningful — postal codes for delivery analysis, states for political analysis.
  • The audience expects a map for spatial data.

When not to use it

  • Raw count data. See the callout above. Use a dot-density map, hexbin, or normalised choropleth.
  • Region areas matter more than region populations. Large sparsely-populated regions dominate visual weight even when their actual contribution is small. Use a cartogram (area-distorted map) or a hexbin instead.
  • Very few regions. A choropleth of 5 regions is a bar chart with a map projection. Use bars.
  • Mixed-resolution data. When some regions have many observations and others have few, the values are not equally reliable. Either filter to high-confidence regions or annotate the uncertainty.

Design principles

Always normalise

This is the single most important rule. Shade by per capita, per area, percentage, or another rate — never by raw count. A count of customers per postcode map is a population map; customers per 1,000 residents per postcode is the actual story. The two maps below use the same solar installations; only the denominator differs.

By count: rooftop solar installations, thousands
New South Wales and Queensland are darkest because they have the most houses. This is a population map wearing a solar costume — it says nothing about how likely a given household is to have panels.
By rate: share of dwellings with rooftop solar, per cent
The same installations divided by dwellings. Now South Australia and Queensland lead, New South Wales drops to the middle, and the story becomes climate and housing stock rather than headcount.

Choose the right scale

Sequential scale for one-directional data (low to high). Diverging scale for data with a meaningful midpoint (above/below average, positive/negative change). Categorical scale only for true categorical data (e.g., region type). Never rainbow. The hero chart is sequential because a solar share has a natural zero and runs one way; a year-on-year change in that share has a natural centre, and the ramp should say so.

Diverging: change in solar share over twelve months, percentage points
A diverging ramp pinned symmetrically about zero, so a cream region means no change and the two hues mean opposite directions. Western Australia and the Northern Territory grew; Tasmania and the ACT slipped back. A sequential ramp would have hidden the sign.

Class deliberately, and fix the domain

A choropleth's ramp is almost always classed: the values are cut into a handful of bins and each bin gets one colour. The number of classes is an editorial decision. Few classes make a bold, readable map that hides detail; many classes approach a continuous ramp — Tobler's classless choropleth — and let neighbouring regions differ by shades the eye cannot name. Just as important is the domain: if the ramp stretches to fit whatever data arrives, the darkest colour always exists somewhere, and a map of trivial differences looks exactly like a map of dramatic ones.

Three classes: a coarse but honest map
The solar rates cut into low, middle and high. The reader can name every region's class at a glance, and the legend is three swatches. The cost: Queensland and South Australia are now indistinguishable, and so are Victoria, the ACT and the Northern Territory.
Seven classes on a fixed 0–100 domain
The same rates, but the ramp now runs from 0 to 100 per cent rather than from the minimum to the maximum in the data. The map goes pale — correctly. No state has solar on more than half its roofs, and a ramp that stretched to fit the data would have painted Queensland as if it had.

Use perceptually uniform palettes

Viridis, magma, or LAB-tuned single-hue ramps. The default ramps in spreadsheet tools are perceptually non-uniform — equal data steps produce unequal visual steps, distorting the reading. Use perceptually uniform palettes; the sequential teal used throughout this entry is one example, with seven stops spaced evenly in lightness so that each class step looks like the same size of jump.

Show the legend clearly

Every choropleth needs a colour-scale legend showing the value range. Include the minimum and maximum values; for sequential scales, include a midpoint. For diverging scales, make the midpoint explicit and meaningful. On the figures in this entry the legend sits over open water, which is the right place for it: never let it cover a region.

Annotate notable regions

In a national map, a handful of regions usually carry the story. Label them directly — the city with the highest value, the region with the unexpected low. The unlabelled regions provide context; the labelled ones provide narrative. With only eight regions, labelling all of them costs nothing; with eighty, label the extremes and let the legend do the rest.

Beware the large-area bias

Colour is read in proportion to the area it covers. On a map of Australia, Western Australia is a third of the continent and the ACT is a speck, so whatever colour WA carries dominates the impression and whatever the ACT carries is invisible — regardless of how many people, customers or votes each contains. When the quantity being mapped belongs to people rather than to land, a cartogram makes area carry the data instead.

Resident population as a choropleth, millions
Western Australia's three million people cover a third of the canvas in a mid teal; New South Wales' eight and a half million are squeezed into a corner. The eye weights the map by land, and land is not what the data is about.
The same population as a Dorling cartogram
Each state becomes a circle whose area is its population, held near its true position. The south-east corner now dominates the picture, as it does the country, and the territories are small but present rather than lost.

Pair with a sorted bar chart

For precise comparison, complement the choropleth with a small inset bar chart of the top-N regions. The map shows spatial pattern; the bars show precise comparison. Two charts, one story.

Anatomy

The composition of a choropleth
An anatomical guide

A choropleth's anatomy is geographic polygons + colour-scale legend + selective labels. The reading task is spatial; the scale legend is the precision lookup; the labels provide narrative. Tasmania is left without a datum above to show the neutral no-data fill — a region with no value must never be painted as if it had the lowest one.

  • Cartogram — area-distorted map. Better when geographic area misleads.
  • Hexbin / H3 map — uniform grid cells. Removes the irregular-region problem.
  • Tilegram — equal tiles per region; every region visible whatever its area.
  • Dot-density map — for raw count data; one dot per N events.
  • Bivariate choropleth — two variables encoded as a 2D colour scale; advanced technique.
  • Value-by-alpha — choropleth modulated by a second variable; covered in a separate entry.

Reading list

  • Snow, J. (1855). On the Mode of Communication of Cholera. The original public-health map.
  • Brewer, C. (2003). Cartographic Design for Color. The science of map colour palettes.
  • MacEachren, A. (1995). How Maps Work. The foundational reference for thematic cartography.
  • Tobler, W. (1973). Choropleth Maps Without Class Intervals? — the case for continuous scales.