In Bad Graphics – Stacked Pyramid Chart, I critiqued a popular infographic display type, the pyramid chart. In this post I will repeat the favor for funnel charts. These are not the funnel charts which are also called tornado charts, and in some circles are used to construct population pyramids (see Tornado Charts and Tornado Charts and Dot Plots elsewhere on PeltierTech.com). These are stacked 3D abominations which sometimes follow a flawed analogy to a physical funnel, just like the pyramid charts in my previous example follow a flawed analogy to an actual pyramid.
The first example of a funnel chart comes from the FusionCharts Free Chart Gallery.
This example shows the flawed analogy I’ve alluded to, which indicates that some quantity goes from a large number to a small number, starting with website visits, only a fraction of which result in downloads, and only a fraction of those result in inquiries about purchase, etc. It seems to me that a physical funnel should make everything that goes in the top come out the narrower bottom. If it didn’t work that way, your kitchen would become a disaster.
This chart indicates a filtering process, but a more accurate representation might show X percent going forward and 100-X percent dropping out. The chart’s analogy breaks down, however. In their zeal to make the chart more closely resemble a kitchen funnel, there is a conical section and a cylindrical extension on the narrow end of the cone. Within this cylindrical section, the values decrease in magnitude, but the diameter is constant. So the false impression that diameter relates somehow to value has broken down.
A second version of this chart is hardly better. It has lost the cylindrical extension at the bottom, but all of the conical sections have different slopes of their sides. This is almost a stacked parabolic chart, which I really hope never to see.
Let’s review the problems with 3D charts of this type.
- The value of each section is portrayed by its thickness, while the apparent cross-sectional area, width, and volume of each 3D section overwhelms the reader’s impression of its thickness.
- The apparent cross-sectional area, width, and volume of each 3D section change arbitrarily, more as a function of position within the stack than of value.
- The 3D appearance adds unnecessary clutter (chart junk).
- The chart makes comparison between segments impossible.
- The chart do not allow easy comparison between time periods.
- The whole funnel chart rationale is based on a faulty analogy to an actual funnel.
The pyramid chart shows a similar faulty analogy to a true pyramid. The base is the most important part of a pyramid: without a strong base, the pyramid will soon crumble. In the example below (from UberBI), the ‘Accessories’ product line is depicted as the base of the company’s sales. While the profit margin for accessories may be greater than for the main product, the ‘Accessories’ product line cannot support the whole company, and like a pyramid with a faulty base, a company based on a nonessential product line will soon crumble.
The following FusionCharts example doesn’t even follow the false analogy that each slab is a subset of the slab above. Instead, the data points are aligned arbitrarily, in alphabetical order.
Their 2D version of this chart is not much easier to read than the 3D chart above.
FusionCharts offers more examples, but I think they’ve shown the features of their charts quite well in the charts I’ve shown.
Dundas has a very good white paper, Dashboard Best Practices, which is available online in both html format [link no longer available] and as a pdf document [link no longer available]. Unfortunately, their chart gallery is filled with examples that show how not to apply best practices. They proudly display examples of bad funnel, cone, and pyramid charts.
Similar examples of bad charts abound. Search the internet for computer graphics, business intelligence, data visualization, and dashboards, and you will find many more bad examples than good.
So what’s a good alternative? Keeping it simple is always a good strategy. If you have a single time period to report, a bar chart works nicely:
There is no need to overwhelm the reader with multiple colors, no reason to add 3D effects or any of the seductive lighting and shading effects now too readily available in modern software packages. Note how easy it is to compare values in this chart. Data labels can easily be added to show a relevant percentage.
To present a time series you could use a line chart:
Because the ‘Purchased’ series is so small relative to the other series, you should add another chart showing this series by itself.
Another way to analyze this kind of data is to normalize each series by its value at the beginning of the time span covered by the data, to compare the percentages each series changes over time. The labeling in this chart should be laid out better, but you get the idea.