# Trolls, Elves, Billygoats/Pies, Bars, Lines

Yesterday, we talked briefly about Seth Godin’s (flawed) three laws for graphing, and mentioned that he had reengaged these issues in a follow up post.

He illustrates his “No Bar Charts Law” with the following pie and bar charts:

The pie chart is designed to make the point that Trolls are where the action is; Godin believes that this graph gets the point across clearly and easily, while that same point is lost in the below bar chart.  Although Trolls are the highest bar, Seth says, the bigger change in the Billygoats category  and the lack of emphasis for the main point may draw away the audience’s attention from trolls.

## TROLLS AND ELVES AND ORCS, OH MY!

Of course (as pointed out by the Juice Analytics Blog), Seth’s comparison is a little unfair–his bar chart has three times the data that the pie chart has.  Of course it’s more complicated and unfair.  In fact, the three column bar chart is a better graph to show three years of data than three successive pie charts.  I made the three pie charts below using (roughly) the same data as Seth’s bar chart above, and they illustrate some of the big problems with using multiple pie charts.  Can you tell how the Elves slice changes over time?  By how much does the dragons slice change?

Others have attempted to improve on Seth’s charts.  Taking the route of simplicity, Jon Peltier presents the Pie Chart data in a simple bar graph:

Stephen Few also had a variation on the same theme, with another clean, clear, simple bar chart showing the same data.  Stephen’s looks a little different because he kept the emphasis on the troll bar (a nice touch) and used the percentage data that Seth put in his pie chart instead of the raw numbers Jon used.

Honestly, I like these simple bar charts better than the simple pie chart.  It easily shows the relative size of the categories while still emphasizing the dominance of the troll category.  But as Jon says, “I guess it’s a matter of taste” — if you know your audience prefers pie charts go with the pie chart.

I’m a little less enamored with Jon’s three-panel bar chart to display the time-series data.

Although I think the three bar charts are better than the triple pie charts, they still have the problem of making it difficult to easily compare across years.

Stephen’s multi-year bar chart only shows two years and changes the data to make the point that multiple bar charts side by side can emphasize change.

Jorge Camoes also changes the data to make a point about using a bar chart to (literally) highlight trends.  Jorge puts only the most important categories in color, and those categories show the declining importance of trolls and the increasing importance of elves.

## MY CONTRIBUTION TO THE MENAGERIE

As for me, I actually prefer an entirely different style of chart for this data: the line chart.  Although admittedly opinions about these charts are mixed in the office, I think this style of chart preserves the advantages of the pie and bar graphs while being (in my opinion) much more easily readable.  The slopes between the year categories show the trajectory of change much more clearly than either the pie or bar charts, while the vertical distance between the categories shows the relative magnitude.  Sure, things get a little messy at the bottom, but that just makes the point that there’s not much difference between the low-frequency categories.  What do you think?

Update (7-16): Based on some nice feedback from Jon in a comment below, I’ve made some changes to the line chart.  They include labeling the lines (and ditching the legend), emphasizing (through size and color) the time series we care about (Trolls & Elves), and deleting the grid lines (which, for this data, were unnecessary to get the picture).  The line graph above might be one I put into a report, while this one is better suited for a presentation making the point that Trolls and (to a lesser extent) Elves are the important categories to focus on.

His comments bring up another point–he doesn’t like how the line chart looks.  That’s fine–neither do some of my coworkers!  Although there’s a lot science in choosing which chart to use to tell your story, there is still room for some art and personal expression.  As long as the the audience can clearly get the point of the story, there is room for slight differences in how we tell it.

## One comment on “Trolls, Elves, Billygoats/Pies, Bars, Lines”

1. Charting of these time series was difficult. My split bar chart suffers from difficulties in comparing values across years, but Seth’s point was that Trolls were big, and I was showing that Trolls were big three years in a row.

A clustered bar like Jorge’s may be best, even though the rule says use a line chart for time series. I tried a line chart, which ended up looking much like Geoff’s, and I didn’t like it (and no offense, but I equally dislike Geoff’s).

Here’s how Geoff’s line chart may be improved:
1. Give prominent colors to the trolls and elves, and color the other series an unobtrusive gray. This follows the lead Jorge took with his clustered bar chart.
2. Move labels from the chart to the series. It’s hard to do for the clump of data at the bottom, but it’s probably less critical down in the weeds anyway.