Functional design for graphs

Good graph design means that information will be communicated quickly and accurately. By carefully designing the components of your graph, you can ensure that the graph is easy to understand both textually and visually.

Use appropriate divisions and scale

The axis scale, divisions and aspect ratio should be chosen to show data in a way that is contextually appropriate and visually effective. These elements should neither exaggerate nor minimise the data – for example, by simply stretching a graph to make it taller (to accentuate differences in values) or wider (to minimise differences).

The scale should minimise wasted space so that the data fill as much of the data region as possible. Axes should not be extended beyond the data region unnecessarily.

The quantitative axis for column or bar graphs should start at zero. Dot plots and line graphs for time-series data do not need to start at zero.

Show intervals for missing data

The distance between intervals (tick marks) on an axis should always be consistent with the difference in quantity that they represent. For example, equal intervals of time should be shown as equal intervals of distance in tick marks along an axis.

Each interval should still be labelled when data for the interval are missing. Indicating a break without labelling the missing intervals can be confusing and even misleading for readers (Few 2008a). Breaks in measurement or missing data should always be shown as such (e.g. a break in a line graph).

Label the axes

Both axes should be clearly labelled with categories and values. Align labels horizontally, where possible, including on the vertical y axis if the label is short and space permits. If the y axis label is aligned horizontally, place it at the top of the axis; if it is aligned vertically, place it midway down the axis.

The appropriate units should be shown in the axis title in parentheses, rather than on the axis itself:

Any multipliers in axis labels should be used consistently. It is always preferable to scale units so that multipliers are not needed. If multipliers are needed, the following form should be used:

The multiplier always comes at the end of the unit. To avoid confusion with magnification – for example, in photomicrographs – no brackets are used for the multiplier. Use the symbol × to indicate multiplication, not the letter x.

Quantitative axis tick marks should extend the full width or height of the graph so that values can be more easily determined. Keep them light in weight and tint (e.g. 0.5 pt at 50% grey). However, if the graph is particularly simple or small, tick marks may not be necessary.

Ensure that data points are clear

Points on scatter, dot or strip plots should be clear and of appropriate size. Open and closed symbols of different shapes are usually the easiest and clearest way to distinguish between points and series. Balance the point size with overlapping density so that all data are visible.

Lines should be clearly distinguishable from each other by varying line weight, colour or style. However, be careful that these differences do not result in visually emphasising some lines over others.

On line graphs, it is best to avoid ‘node symbols’ (a symbol such as a star or shape placed on the data point); they add visual clutter. Consider using vertical bar graphs instead of line graphs when the reader is required to identify individual data values.

Columns, bars or boxes should be clearly distinguishable from each other by varying colour, tint (steps of 20%) or pattern. As for lines, be careful that these differences do not result in visually emphasising some data items over others. Make sure there are not too many categorical subdivisions, dot types or lines – the more cluttered the graph, the more it becomes unattractive and difficult to understand.

Vertical or horizontal bars that represent discrete groups or categories should be separated from each other with white space. A bar to white space ratio of approximately 1:1 is best (Few 2012).

Leave less or no white space between clustered bars for subordinate categories of the highest-order groups (e.g. side-by-side bars for men and women across each Australian state and territory). Clustered bars should sit close together, making it clear to the reader that these bars are related. Try to show no more than 3 or 4 clusters for each group – readers will find it difficult to compare the same subcategories across groups for more categories.

Add supporting information

Include contextual information if this is appropriate and will not distract from, or obscure, the data. Examples of supporting information are mean, target, baseline, deviation, regression, best-fit or trend lines, confidence intervals, acceptable ranges, and comparison with a previous time period.

When needed, the legend should be placed in a prominent area, such as the bottom left corner of the graph, listing the items in the same sequence as they occur in the graph. It may be more effective to label the data directly (e.g. by adding labels to each line at the right-hand side of a line graph), provided that the data items are distinct enough. This eliminates the extra time it would take a reader to understand a legend.

Consider error bars and confidence intervals

Confidence intervals or error bars are lines or shading extending from data values to show variability or uncertainty in these values. Wide or long error bars indicate more uncertainty.

Error bars are not used consistently across scientific disciplines. This means that readers’ ability to recognise and understand error bars will differ.

Accordingly, if variability or uncertainty in your data values is not a focus of the graph or your data message, it may be best to present the data without error bars. If variability or uncertainty is a focal point, make sure these aspects of the data are discussed in the text, and that error bars or shading are explained in the notes section underneath the graph. This enables you to explain the role of error in the results, rather than leaving the reader to potentially misinterpret the graphical display of error.

Shading designs may convey error to readers more accurately and less distractingly than lines protruding from individual data values (Correll & Gleicher 2014).