The two extremes of visualizing data are poor and excessive design. On the one hand, there are poorly designed charts that contain unnecessary elements that add visual clutter, have poor color selections and so on (see Stephen Few’s example). On the other hand, other charts are over-designed with various embellishments added purely for aesthetic effect (again, Stephen Few’s example to the rescue!). Both types of graphs make it difficult for the audience to extract information that the chart is meant to represent without sacrificing accuracy of the data.
A recent paper on visual embellishments in bar graphs empirically evaluates how different design choices affect accuracy of data extracted from a chart. The authors test the following bar charts against a standard graph (see g in Figure below):
- Rounded ends on rectangle bars (see a)
- Triangles instead of rectangle bars (see b)
- Capped ends on rectangle bars (see c)
- Overlapping triangles instead of rectangle bars (see d)
- Quadratically increasing triangles (see e)
- Bars that begin below the origin (see f)
Authors find that only the capped charts (c) and bars extending below the origin (f) are as effective at representing data accurately as the standard bar chart (g). For relative comparison between bars and their numerical representations, only the chart where bars extend beyond the origin (f) do as well as standard bars (g). The take-away? Stick to the standard bar chart. And, in the words of the authors:
Bar chart embellishments do indeed have an impact on how well the data within the chart can be communicated. For nearly all tested chart embellishments, even small changes like rounding the top of a bar, led to higher error rate.
So, while striving to avoid the extreme of poor chart design, make sure you don’t end up on the opposite end of the spectrum. Simply designed charts are often the best ones.