What a chart does
A chart turns a table into shapes. A value becomes a length, a position, an angle or an area, and the reader compares the shapes instead of reading the numbers one at a time. The guide to presenting statistics from UNECE, the United Nations Economic Commission for Europe, sorts what charts can show into five kinds: comparison, change over time, distribution, correlation, and the share of a whole.
Two questions decide the choice:
- What should the reader see? Which group is largest, how something has changed, how the values are spread, or whether two quantities are related.
- What are the data? Categories with no order of their own (fruits, ways of travelling), numbers that have an order (number of siblings, years), many measurements of one quantity (heights, times), or pairs of measurements (two values for each person).
Seven charts and what they show
| Chart | Shows | Data |
|---|---|---|
| Bar graph | counts or values compared across categories | categories |
| Vertical line chart | how often each value occurs | a few possible whole numbers, such as number of siblings |
| Pie chart | how a whole is divided | parts that together make up the whole |
| Line graph | change along an ordered axis, usually time | measurements at different times |
| Histogram | how a measured quantity is distributed | many measurements of one quantity, in classes |
| Box plot | five numbers that summarise a distribution | many measurements of one quantity |
| Scatter plot | the relationship between two quantities | pairs of measurements |
Choose by the question
The table below is the same rule table that marks the Choose the chart questions on the tables and charts worksheets. More than one chart can fit the same situation: bars fit any comparison of categories, and a few measurements over time can be shown with bars as well as with a line. Where a chart does not fit, the reason is given.
| Data | Bar chart | Pie chart | Line graph | Histogram | Box plot | Scatter plot |
|---|---|---|---|---|---|---|
| Categories compared, with no whole they share | fits | does not fit | does not fit | does not fit | does not fit | does not fit |
| Parts of one whole, each unit counted once | fits | fits | does not fit | does not fit | does not fit | does not fit |
| Answers where each person could choose several | fits | does not fit | does not fit | does not fit | does not fit | does not fit |
| One quantity at different times | fits | does not fit | fits | does not fit | does not fit | does not fit |
| Many measurements of one quantity | does not fit | does not fit | does not fit | fits | fits | does not fit |
| Pairs of two quantities measured on the same unit | does not fit | does not fit | does not fit | does not fit | does not fit | fits |
* Bar chart, with a caption saying that several answers were allowed.
Why the others do not fit
Categories compared, with no whole they share
- Pie chart. The values do not add up to one whole, and a pie chart shows the parts of one whole.
- Line graph. The categories have no order, so a line between them shows nothing.
- Histogram. A histogram divides a measured quantity into classes; categories are not measured.
- Box plot. A box plot summarises many values of a measured quantity; categories are not measured.
- Scatter plot. A scatter plot needs two quantities measured on each item.
Parts of one whole, each unit counted once
- Line graph. The categories have no order, so a line between them shows nothing.
- Histogram. A histogram divides a measured quantity into classes; categories are not measured.
- Box plot. A box plot summarises many values of a measured quantity; categories are not measured.
- Scatter plot. A scatter plot needs two quantities measured on each item.
Answers where each person could choose several
- Pie chart. Each person could choose several answers, so the shares add up to more than 100% and there is no one whole to divide.
- Line graph. The categories have no order, so a line between them shows nothing.
- Histogram. A histogram divides a measured quantity into classes; categories are not measured.
- Box plot. A box plot summarises many values of a measured quantity; categories are not measured.
- Scatter plot. A scatter plot needs two quantities measured on each item.
One quantity at different times
- Pie chart. The values at different times are not parts of one whole, and a pie chart loses their order.
- Histogram. A histogram shows how values spread across classes, not how one value changes over time.
- Box plot. A box plot summarises the values and loses their order in time.
- Scatter plot. There is one value at each time, not two quantities measured on each item; a line graph joins the values in order.
Many measurements of one quantity
- Bar chart. A measured quantity is divided into classes that meet, so the bars touch: that is a histogram.
- Pie chart. A pie chart has no axis, so it cannot show how the values spread along a scale.
- Line graph. The values are not in order of time, so a line between them shows nothing.
- Scatter plot. A scatter plot needs two quantities measured on each item.
Pairs of two quantities measured on the same unit
- Bar chart. Each item has two numbers that belong together; this chart shows one quantity and loses the pairs.
- Pie chart. Each item has two numbers that belong together; this chart shows one quantity and loses the pairs.
- Line graph. The items are not one value in order of time; a line through them draws a path that is not in the data.
- Histogram. Each item has two numbers that belong together; this chart shows one quantity and loses the pairs.
- Box plot. Each item has two numbers that belong together; this chart shows one quantity and loses the pairs.
An example: the survey with several answers
Forty pupils were asked Which pets do you have at home? and could tick every answer that applied.
The data behind the figure
| Category | Count |
|---|---|
| Dog | 14 |
| Cat | 12 |
| Rabbit | 6 |
| Fish | 4 |
| No pet | 10 |
The answers were dog 14, cat 12, rabbit 6, fish 4 and no pet 10. That is 46 ticks from 40 pupils, because some have more than one kind of pet. As shares of the pupils they are 35%, 30%, 15%, 10% and 25%, which add up to 115%.
A pie chart cannot show this survey. The circle is one whole of 100%, and a pupil with both a dog and a cat would have to sit in two slices at once. Taking shares of the 46 ticks instead gives a circle that adds up, but then the dog slice is 30% of the ticks, and a reader reads it as 30% of the pupils having a dog. The share is 35%.
The bar graph above shows the survey correctly: one bar per answer with the number of pupils who chose it, and a line under the chart saying that more than one answer was allowed. Turned into shares of the 40 pupils, the bars add up to more than 100%, as they should.
Common mistakes
- A pie chart of a question with several answers. The shares add up to more than 100%, and there is no single whole to divide (the example above).
- A bar axis that does not start at zero. A bar’s length is its value. With the axis starting at 80, the bars for 85 and 90 are only 5 and 10 units long, and 90 looks twice as large as 85 although it is less than 6% larger. UNECE’s guide advises starting the axis at zero and marking a break in it when part of the scale is left out.
- Three-dimensional charts. In perspective, the parts nearest the reader look larger than they are. According to UNECE, three-dimensional charts rarely add value and often distort the data.
- A line between categories that have no order. The line between two points on a line graph shows a change from one to the other. Between dog and cat there is no change to show; that is a job for bars.
- A relationship read as a cause. A scatter plot can show that two quantities are related, not that one causes the other. See correlation and causation.
Going deeper: lengths are read more accurately than angles
Two charts of the same table can be read with different accuracy. A reader of a bar graph compares lengths along a common axis. A reader of a pie chart compares angles and areas, and UNECE notes that many readers find angles harder to compare than bars or lines: two slices of nearly equal size are hard to tell apart, while the same difference is plain between two bars. The guide also notes that many statisticians advise against pie charts, but that they should not be ruled out completely, because they show well how large one category is relative to the whole.
For bars and pie slices alike, the guide gives the same advice about order: sort the categories by size to make them easier to compare. That applies to categories with no order of their own. Where the categories have an order, such as age groups or years, that order stays.