Bar graphs are the most recognisable and common method for displaying data that represent differences in magnitude across groups, categories or time. Groups can be ranked or ordered (e.g. 10-year age bands, ranging from birth to 90 years old), or unrelated (i.e. nominal). The height or length of side-by-side bars is automatically understood by readers as relative size.
Data in bar graphs can be absolute values, percentages, proportions, ratios or rates.
Orientation of bars: horizontal vs vertical
Either horizontal or vertical bars can be used to display data that are grouped into categories or a small number of time intervals. However, there are data relationships where one orientation is preferable to the other:
- Vertical bars should always be used when they represent data over time.
- Horizontal bars should be considered when the labels for groups or categories do not fit easily in a horizontal alignment along the x axis.
If you are including bar graphs for both time-series data and categorical data, consider reserving vertical bars for the former and horizontal bars for the latter. This differentiates these data types, and provides a consistent association between data relationships and graph type, which will assist reader understanding.
Horizontal bar graphs
Comparing magnitude across separate groups or categories
Data that describe proportions or magnitudes of a measure or measures across different populations, groups, items or categories are most easily understood by readers as a bar chart. Bars are the most familiar shape to readers for understanding differences in data values, and the height or length of side-by-side bars is automatically understood as relative size:
Visualising single part-to-whole relationships
Horizontal bar graphs are effective for readers to judge part-to-whole relationships, particularly when the bars are aligned lengthwise. The inclusion of the horizontal axis in a bar chart further aids comparisons of magnitude.
Pie charts are often used to display proportions of a total measure. The problems associated with this graph type are discussed in What to avoid in graphs. The graph below shows how a traditional pie chart is reconfigured into a more readily understood bar chart (Few 2012):
Vertical bar graphs
Vertical bars may be used for time-series data when specific data values need to be emphasised. It can be difficult for readers to identify values for a specific time point on line graphs, which instead highlight trends or directional changes in data values over time.
Visualising data values over time, for small time series
Vertical bars are useful for comparing data values across a relatively small number of time points (around 8 or less). Using vertical bars spaced along a horizontal (x) axis of time points draws on readers’ intuition to think about time horizontally and linearly, from left to right:
Visualising differences from a baseline
Displaying deviations in data values from a baseline (e.g. the first year of measurement or the year before a new environmental management system was introduced) can be a clear and powerful way to communicate changes in a measure over time or across groups.
In (vertical) deviation bar graphs, data values are shown as differences from a meaningful baseline, rather than raw values of measurement. As with all bar graphs, bars should start at 0 on the y axis. However, the horizontal x axis no longer intersects at the base of the y axis – the value of 0 may be positioned some way up the y axis to accommodate negative deviations from the baseline.
Bars (i.e. data values) above the x axis indicate positive differences from the baseline; bars below the x axis indicate negative differences from the baseline.
Data values can be absolute deviations from a baseline or percentage changes from the baseline.
Histograms: visualising frequency distributions
Data that describe the frequency or ‘counts’ of a measure across each of its possible values or intervals are typically presented as a vertical bar chart known as a histogram – for example, the number of days per year with temperatures across the range of –10 °C to 50 °C.
Vertical bars are used instead of horizontal bars for this type of data because the intervals along the horizontal axis are related. That is, they are the range of a single measure rather than the discrete groups or categories described for horizontal bar graphs. Like time-series data, readers tend to understand measures that are displayed on a left-to-right plane as part of the same sequence:
As for time-series data, vertical bar graphs are recommended for displaying frequency distributions for measures with relatively few intervals, or when the reader needs to identify individual values. For measures with a large number of intervals, and where authors want to convey the shape of a distribution, a line graph known as a frequency polygon is recommended.