Guidance on Best Practice in Statistical Presentation

Main Content

Download this document in pdf format (421Kb)

Drawing Charts: Best Practice

Why draw a chart?

Charts can be an eye-catching and effective way of presenting your data. They provide a strong visual impression of distributions or trends, and can add interest to pages of text or tables. They are particularly suitable if you have a small number of values to display, and it is the comparison between the data that is important, rather than the values themselves.

When not to draw a chart

Charts are mostly for demonstration (i.e. display) purposes rather than as a source of reference material. They are not usually suitable for displaying large numbers of values, and it is also difficult to read precise values off a chart. The place for storing precise data is in a reference table.

Types of chart

There are many forms of chart but this webpage considers three main types:

	Bar charts. Suitable for presenting sizes of related measurements. For example, average salaries of different ethnic groups.
	Line graphs. Suitable for presenting a series of values which change over time. For example, monthly inflation rate figures.
	Pie charts. Suitable where it is important to show each value as a share of the total. For example, share voting for each party at an election.

There may be occasions when you wish to use other charts - scatter plots, for example - but be aware that general readers may be less familiar with these and that you may need to give extra explanation.

Although modern software makes it easy and tempting to draw flashy three-dimensional charts, these should be avoided. They may look good but in practice can be difficult to interpret properly, and can be very misleading.

Many aspects of good practice are common to all charts, but each type also has its own specific requirements:

Design principles for all charts

Clear title (supported by relevant footnotes if appropriate).
Include information on the geographic area covered, the time period, and the units of measurement.
Indicate data source.
If colours are used they should be easy to distinguish.
If possible, label lines and sections of the chart directly rather than in a separate key.
Charts should be easy to read. If you can't portray the data clearly then either split them into more than one chart, or use a table instead.
Textual summary. A good chart should tell its own story, but the key message(s) of your chart should also be summarised in words.
Only make a chart as large as is needed to get the message across. Charts which are too large are harder to take in at a glance.

Bar charts

May be either vertical or horizontal bars.
Axis should start at zero and be clearly labelled.
Bars should be of equal widths - otherwise their visual impact is distorted.
Bars should be in an appropriate order. Order of size is usually best unless there is an obvious natural order such as age or time.

Line graphs

Axis should start at zero and be clearly labelled.
Multiple lines acceptable (eg for different countries). However, too many lines, or lines which criss-cross, may be hard to read.
If there are gaps in the data series, the sections of line adjacent to the gaps should be omitted or dashed. Drawing a solid line implies the values are recorded, which could be misleading if the true value was fluctuating.

Pie charts

Avoid having too many slices in a pie chart - half a dozen is ample. If necessary you can group some of the smaller slices into 'Other'.
Place the slices in an appropriate order - order of size is common.
Start the first segment at 12 o'clock (or some other multiple of 3 hours) and work round clockwise.
Be wary of displaying multiple pie charts. Although users can read a single pie to appreciate relative proportions, it becomes a much harder exercise if they are comparing several pies. In such cases appropriate bar charts (or a table) may be preferable.

Bar chart: example

Bar chart example

Source: Social Trends 2005, Office for National Statistics

Line graph: example

Line graph example

Source: Social Trends 2005, Office for National Statistics

Pie chart: example

Pie chart example

Source: Social Trends 2005, Office for National Statistics (shading has been edited)

Bad practice: the 3-D chart

Bad practice 3D chart example

In which year was the population of Greater London highest? The true answer is 1939, but the 3-D distortion could mislead you into thinking it's 1961. Moral: avoid 3-D. It looks clever but is likely to confuse your audience.

Note too that the colours on this 3-D chart are poor for effective black and white printing.

A confusing bar chart

Confusing bar chart example

The volume of burglary within Melton is only 1721 incidents compared to 5835 for Charnwood, however this figure accounts for 18% of crime within Melton Borough. Thus, burglary is a relatively more important issue for Melton Borough than for Charnwood Borough.

Source: Leicestershire County Council 2005, with kind permission.

A first look at this chart raises questions:

The title suggests that Melton is the worst area, but Charnwood has the highest bar and Melton's is one of the shortest. How can this be?
The vertical axis is in thousands, but the values written above the bars are percentages and don't appear to have any relationship with the height of the bar. What do they mean?

From the summary and footnote we learn that the height of the bars represents total number of burglaries, and the percentages show burglaries as a proportion of all crimes in each area. However, this isn't intuitive - remember that a good chart should tell its own story. A better solution might be to show the count and rate data on two separate charts, again with appropriate commentary referring to the significance of each measure. Alternatively both count and percentage data could be presented in a table.

Another unusual feature of the chart is that the widths of the bars vary according to the size of the percentage values. This is overly subtle and serves no useful purpose, especially as the percentages are stated anyway. Much simpler would have been to make all bars the same width.

This chart doesn't work as a printed or online copy, but in the authors' defence it was prepared for a presentation to subject experts, and the detail was explained at the time. This could have been appropriate for the setting, demonstrating that the suitability of a chart partly depends on audience and context. As a general rule though, particularly in publications, your general aim should always be to keep your charts clear and simple, minimising the need for explanation.

Bad practice: the truncated axis

Compare these two, both showing the same data:

Bad practice - truncated axis one example

The left hand chart has a vertical axis starting at £90m, whereas the right hand one correctly starts at zero. Whereas the true picture is of four firms with relatively similar profits, the left hand chart gives the initial impression that Firm D has generated nearly four times as much profit as Firm A. This could give a very different impression to potential investors!

There are occasions when axis truncation is appropriate. For example, if the value of a variable tends to be stable over time you may wish to zoom in to a small section of a line graph axis to highlight more subtle changes that are occurring.

However, axis truncation is often done to give a deliberately misleading impression, encouraging readers to think that the situation is better or worse than it really is. This might be done by someone with a biased agenda. An independent statistician, however, should seek to show their audience the true situation and let them make their own judgements. So in most cases axes should start at zero.

You are here: