Choosing a chart type: which chart fits your data?

Choose a chart from what you want to show: comparison, time, part-to-whole, distribution, relationship, map or flow. Ends at a named chart and its rule.

Start

Choose a chart type
For anyone making a chart for a report or slide: pick what you want to show and how much data you have, and get a chart type and the rule to check.
Based on UK government guidance (Analysis Function and ONS) and the Financial Times Visual Vocabulary. These are rules of thumb, not laws.
First write the one sentence the chart must say, such as "Sales fell in every region except the North".
Do readers need to look up exact values?
- Yes: go to step 3, Use a table
- No: go to step 4, What do you want to show?
Use a table
A well-structured table can often be as powerful as a chart, and it is the right choice when people will read off precise numbers.
Check: right-align numbers and keep gridlines light.
What do you want to show?
- Compare or rank: go to step 5, What are you comparing across categories?
- Change over time: go to step 10, How many time points?
- Parts of a whole: go to step 17, One total or several?
- Spread or relationship: go to step 23, Spread of one measure, or a link between two?
- Places or flows: go to step 31, Places on a map, or flows between states?

Compare categories

What are you comparing across categories?
- One value each: go to step 6, Use a simple bar chart
- Two to four series: go to step 7, Use a grouped bar chart
- A before and after pair: go to step 8, Use a dot plot
- More than four series: go to step 9, Use a heatmap
Use a simple bar chart
The standard way to compare the size of things. Use horizontal bars when category names are long.
Sort bars by value when rank matters. Keep the gap between bars narrower than a bar.
Check: the value axis must start at zero. Never break it on a bar chart.
Use a grouped bar chart
Also called a clustered bar chart. UK Analysis Function guidance calls two bars per cluster best practice and advises a limit of four.
Check: if readers mainly compare one series across categories, small multiples may read better.
Use a dot plot
A simple way to show the change, or the range from minimum to maximum, across many categories.
Check: order the categories by one of the values so the pattern is easy to see.
Use a heatmap
A grid of two sets of categories, each cell shaded by its value. Good for showing patterns between two categories of data.
Check: it is less effective at showing fine differences in amounts. Add value labels or a table if they matter, or try small multiples.

Change over time

How many time points?
- Two or three: go to step 11, Use a slope chart
- A few, one series: go to step 12, Use a column chart
- Many points: go to step 13, How many series?
Use a slope chart
Good for showing change as long as the data can be simplified into two or three points without missing a key part of the story.
Check: label both ends of each line directly. With many categories the lines cross into a tangle.
Use a column chart
ONS guidance suits column charts to a single series with only a few data points, and to flow data such as counts per month.
Check: the axis must start at zero. If that hides the change, use a line chart instead.
How many series?
- Up to four: go to step 14, Use a line chart
- More than four: go to step 15, Use small multiple line charts
- Parts of a changing total: go to step 16, Use a stacked area chart
Use a line chart
The standard way to show a changing time series. UK Analysis Function guidance says aim for a maximum of four lines.
Check: unlike a column chart, a line chart axis can start above zero when that helps. Keep the scale honest.
Use small multiple line charts
One small chart per series on the same scales. Readers can see each series without a tangle of lines.
Check: keep the axes identical across panels, or comparisons mislead.
Use a stacked area chart
Shows how a total changes over time and roughly what makes it up.
Check: only the bottom layer has a flat baseline, so changes in the upper layers are hard to read. Use with care.

Parts of a whole

One total or several?
Use part-to-whole charts only when the parts add up to a meaningful 100%. Not for rates, incomplete data, or survey questions where people tick several boxes.
- One total: go to step 18, How many parts, and is one dominant?
- Several totals: go to step 20, Do the totals matter, or only the shares?
How many parts, and is one dominant?
- Five or fewer, one dominant: go to step 19, Use a pie or donut chart
- More, or similar sizes: go to step 6, Use a simple bar chart
Use a pie or donut chart
UK Analysis Function guidance allows pie charts with five categories or fewer and a dominant category. ONS says ideally three or four, never more than six.
A donut works the same way and leaves room for the total in the centre.
Check: if segments are similar in size, readers can't compare them. Use a bar chart instead.
Do the totals matter, or only the shares?
- Totals and shares: go to step 21, Use a stacked bar chart
- Only the shares: go to step 22, Use a 100% stacked bar chart
Use a stacked bar chart
Compares the composition of totals across groups.
Check: keep to four or five segments per bar and group the rest as Other. Only the first segment has a common baseline.
Use a 100% stacked bar chart
Every bar runs to 100%, so readers compare proportions across groups whatever the size of each total.
Check: say the totals somewhere. A 100% bar hides that one group may be ten times bigger than another.

Spread and relationships

Spread of one measure, or a link between two?
- Spread: go to step 24, How many groups?
- Link between two: go to step 28, Is there a third measure to show?
How many groups?
- One group: go to step 25, Use a histogram
- Several groups: go to step 26, Use box plots
- Few data points: go to step 27, Use a dot strip plot
Use a histogram
The standard way to show a distribution. Keep the gaps between columns small to highlight the shape.
Check: the shape depends on bin width. Try a few widths before you choose.
Use box plots
Summarise several distributions side by side, showing the median and the range of the data.
Check: a box plot hides shape, such as two peaks. Explain the box and whiskers to a general audience.
Use a dot strip plot
Shows every individual value along one line, so nothing is hidden by summary.
Check: dots with the same value overlap. Jitter or stack them.
Is there a third measure to show?
- No: go to step 29, Use a scatter plot
- Yes, as size: go to step 30, Use a bubble chart
- Both categorical: go to step 9, Use a heatmap
Use a scatter plot
The standard way to show the relationship between two continuous variables, each on its own axis.
Check: readers tend to assume a relationship is causal unless you tell them otherwise.
Use a bubble chart
Like a scatter plot, with each circle sized by a third variable.
Check: size circles by area, not radius, and keep the number of bubbles low enough to read.

Places and flows

Places on a map, or flows between states?
- Places: go to step 32, Is the value a rate or a count?
- Flows: go to step 36, What kind of flow?
Is the value a rate or a count?
- Rate or share: go to step 33, How many areas?
- Count or total: go to step 35, Use a proportional symbol map
How many areas?
- Many areas: go to step 34, Use a choropleth map
- Only a handful: go to step 6, Use a simple bar chart
Use a choropleth map
Shades each area by its value. ONS suggests five colour bands, or up to six on a diverging scale.
Check: always map rates, not totals. Counts mostly show where the big or populous areas are.
Use a proportional symbol map
Places a symbol sized by value on each area. Use it for totals rather than rates.
Check: small differences between symbols are hard to see. Label the key values.
What kind of flow?
- Between stages or places: go to step 37, Use a Sankey diagram
- Gains and losses: go to step 38, Use a waterfall chart
Use a Sankey diagram
Shows the volume moving from one condition to at least one other. Good for tracing the outcome of a complex process.
Check: too many nodes turn it into a tangle. Group small flows together.
Use a waterfall chart
Shows how a starting value becomes an ending value through positive and negative steps, such as a budget.
Check: label each bar with its value and make the start and end totals visually distinct.

Outcomes

Use a table

A well-structured table can often be as powerful as a chart, and it is the right choice when people will read off precise numbers.

Check: right-align numbers and keep gridlines light.

You get here from step 2, Do readers need to look up exact values? (Yes).

Use a simple bar chart

The standard way to compare the size of things. Use horizontal bars when category names are long.

Sort bars by value when rank matters. Keep the gap between bars narrower than a bar.

Check: the value axis must start at zero. Never break it on a bar chart.

You get here from step 5, What are you comparing across categories? (One value each), step 18, How many parts, and is one dominant? (More, or similar sizes), step 33, How many areas? (Only a handful).

Use a grouped bar chart

Also called a clustered bar chart. UK Analysis Function guidance calls two bars per cluster best practice and advises a limit of four.

Check: if readers mainly compare one series across categories, small multiples may read better.

You get here from step 5, What are you comparing across categories? (Two to four series).

Use a dot plot

A simple way to show the change, or the range from minimum to maximum, across many categories.

Check: order the categories by one of the values so the pattern is easy to see.

You get here from step 5, What are you comparing across categories? (A before and after pair).

Use a heatmap

A grid of two sets of categories, each cell shaded by its value. Good for showing patterns between two categories of data.

Check: it is less effective at showing fine differences in amounts. Add value labels or a table if they matter, or try small multiples.

You get here from step 5, What are you comparing across categories? (More than four series), step 28, Is there a third measure to show? (Both categorical).

Use a slope chart

Good for showing change as long as the data can be simplified into two or three points without missing a key part of the story.

Check: label both ends of each line directly. With many categories the lines cross into a tangle.

You get here from step 10, How many time points? (Two or three).

Use a column chart

ONS guidance suits column charts to a single series with only a few data points, and to flow data such as counts per month.

Check: the axis must start at zero. If that hides the change, use a line chart instead.

You get here from step 10, How many time points? (A few, one series).

Use a line chart

The standard way to show a changing time series. UK Analysis Function guidance says aim for a maximum of four lines.

Check: unlike a column chart, a line chart axis can start above zero when that helps. Keep the scale honest.

You get here from step 13, How many series? (Up to four).

Use small multiple line charts

One small chart per series on the same scales. Readers can see each series without a tangle of lines.

Check: keep the axes identical across panels, or comparisons mislead.

You get here from step 13, How many series? (More than four).

Use a stacked area chart

Shows how a total changes over time and roughly what makes it up.

Check: only the bottom layer has a flat baseline, so changes in the upper layers are hard to read. Use with care.

You get here from step 13, How many series? (Parts of a changing total).

Use a pie or donut chart

UK Analysis Function guidance allows pie charts with five categories or fewer and a dominant category. ONS says ideally three or four, never more than six.

A donut works the same way and leaves room for the total in the centre.

Check: if segments are similar in size, readers can't compare them. Use a bar chart instead.

You get here from step 18, How many parts, and is one dominant? (Five or fewer, one dominant).

Use a stacked bar chart

Compares the composition of totals across groups.

Check: keep to four or five segments per bar and group the rest as Other. Only the first segment has a common baseline.

You get here from step 20, Do the totals matter, or only the shares? (Totals and shares).

Use a 100% stacked bar chart

Every bar runs to 100%, so readers compare proportions across groups whatever the size of each total.

Check: say the totals somewhere. A 100% bar hides that one group may be ten times bigger than another.

You get here from step 20, Do the totals matter, or only the shares? (Only the shares).

Use a histogram

The standard way to show a distribution. Keep the gaps between columns small to highlight the shape.

Check: the shape depends on bin width. Try a few widths before you choose.

You get here from step 24, How many groups? (One group).

Use box plots

Summarise several distributions side by side, showing the median and the range of the data.

Check: a box plot hides shape, such as two peaks. Explain the box and whiskers to a general audience.

You get here from step 24, How many groups? (Several groups).

Use a dot strip plot

Shows every individual value along one line, so nothing is hidden by summary.

Check: dots with the same value overlap. Jitter or stack them.

You get here from step 24, How many groups? (Few data points).

Use a scatter plot

The standard way to show the relationship between two continuous variables, each on its own axis.

Check: readers tend to assume a relationship is causal unless you tell them otherwise.

You get here from step 28, Is there a third measure to show? (No).

Use a bubble chart

Like a scatter plot, with each circle sized by a third variable.

Check: size circles by area, not radius, and keep the number of bubbles low enough to read.

You get here from step 28, Is there a third measure to show? (Yes, as size).

Use a choropleth map

Shades each area by its value. ONS suggests five colour bands, or up to six on a diverging scale.

Check: always map rates, not totals. Counts mostly show where the big or populous areas are.

You get here from step 33, How many areas? (Many areas).

Use a proportional symbol map

Places a symbol sized by value on each area. Use it for totals rather than rates.

Check: small differences between symbols are hard to see. Label the key values.

You get here from step 32, Is the value a rate or a count? (Count or total).

Use a Sankey diagram

Shows the volume moving from one condition to at least one other. Good for tracing the outcome of a complex process.

Check: too many nodes turn it into a tangle. Group small flows together.

You get here from step 36, What kind of flow? (Between stages or places).

Use a waterfall chart

Shows how a starting value becomes an ending value through positive and negative steps, such as a budget.

Check: label each bar with its value and make the start and end totals visually distinct.

You get here from step 36, What kind of flow? (Gains and losses).