Software packages for charting data often offer menus; i.e., lists of charts that are available. Excel offers a menu; Tableau has Show Me! Elementary statistics and graphing books offer recipes; i.e., guidance on when to use the chart types that appear in common menus. The recipes suggest that one use pie charts for parts of a whole, column or bar charts for comparisons, line graphs for trends, histograms for distributions and scatter plots for relationships. Similar menus and recipes appear on the Web; for example, the Extreme Presentation Chart by Andrew Abela or the Chart Chooser by JuiceAnalytics. These recipes reinforce the idea that the choice of graphs is limited to those on their menus.
One of most common mistakes in designing graphs is choosing the wrong graph type for the data. There's no doubt that recipes are very useful for novices to charting and help prevent the use of inappropriate chart types. However, there are several things that are worth keeping in mind when consulting menus and recipes:
1. They discourage “thinking outside the menu.”
2. They may include chart types that should never be used.
3. They don’t include some very useful chart types.
Today we’ll concentrate on my first objection: discouraging “thinking outside the menu.” You’ll hear more about my other objections in upcoming posts. The common graph forms found on menus may not provide the best way to display a particular data set so that the display answers the question of interest.
I’ll illustrate my point with two examples. I received an email from Beth Lisberg Najberg, an information designer with Beginnings in Chicago, asking me which of the two charts below depicting insurance premiums I preferred, or if another presentation would be better. The email also noted a key observation: although it is recommended that insurance premiums for the oldest age group be no more than triple those of the youngest, in this data set the ratio was much higher. She specifically asked: "Is there a way to show that $314 is nearly 4.5X the base premium?” Whenever I design a graph, I think about the message I want to get across; this clearly was the point that she wanted to be most prominent in the presentation.
I took a look at the two graphs that she had sent and it was obvious that neither made that point. They didn't because they were designed with the menu mindset. While there's nothing wrong per se with line graphs and bar graphs - both are widely used and effective - I don't like either of the graphs above for this set of data. Why? The line graph has lines connecting the premiums. This suggests that there is a continuous range of premiums. There are not. The bar graph has spaces between the bars. This suggests that there are some ages for which no premiums are available. Also not true.
Here is a possible redesign which was also drawn using Excel. To eliminate the spaces between the bars, highlight the data series, choose format selection and then set the gap width to zero in Excel 2010. Use Excel’s Help to see how to set the gap width in other versions.
This figure shows that there is a premium for every age between 0 and 65. It shows that the premium for under 18 years covers more ages than do other premiums. It labels the lowest premium and shows that the highest is 4.4 times the premium for ages 0-18. It also shows where three times the lowest premium falls. Then the title reinforces the message rather than using a generic title like “Rate Spread.”
This is certainly a simple graph that is easily understood. It came about by thinking, “What am I trying to show and how do I show it?” rather than “Which of the charts on the menu should I use?”
My second example also comes from an email I received from someone who knows that I am not a fan of pie charts and even less of a fan of multiple pie charts. The labels have been altered to provide anonymity. The email just said, “Care to improve this graph?”
Figure sent to me for improvement
Pie charts suffer from perceptual problems (we’ll be saying more about this in later posts) and it is difficult to make comparisons across multiple pie charts. Again, let’s think about what we want to show and about how best to show it.
This diverging stacked bar chart clearly shows that that Brand A comes from a more equally distributed audience than the other brands. It facilitates age comparisons much better than the multiple pie charts and enables the reader to see the age distribution of each of the brands. The intensity of the color increases with increasing age, lessening the need for referring to the legend. R code for diverging stacked bar charts is available in version 2.17 or later of the HH package. See ?likert. I plan to post a Tableau workbook that explains how to create diverging stacked bar charts in Tableau by the end of the year.
Avoid the menu mindset. Think about the best way to communicate your story without limiting yourself to the charts on your software’s menus.
