A new type of graph

[DRAFT] Introducing the Dan Chart (ok, that name won’t fly. The Dart? Darts fly!), or to name it descriptively, the aligned stacked bar chart. i have had trouble representing data where there are both a lot of categories (series), and a lot of periods (or values). The clustered column stops being useful around 5×5. The typical solution has been to use either the stacked column, or the 100% stacked column. Both of these have serious drawbacks and an easy solution.

First, observe how unintelligible the clustered column chart is:

The stacked bar char shows how the total changes over time, and tries to give you an idea of how the composition of that total changes. Even though changes in an individual category are represented in the chart, ease of interpretation depends on the order of the categories. Changes over time will be very easy to observe for the first/bottom category. You can easily see, for example that for category 1 October is slightly higher than May (0.6 vs 0.53). However, it’s takes a lot more effort to compare Oct to May for category even though the difference is bigger (0.56 to 0.39). If I asked you which category grew more from Oct to May, you’d be pretty hard pressed to answer that (without the numbers).

The 100% stacked bar chart ignores changes over time and only focuses on the change in composition. This is useful for when those changes either don’t matter, or are very small. It gives you a clearer picture for how composition changes. However, it still has the same problem that it is hard to get insights about the changes for those middle categories. It gets unwieldy around 4 or 5 categories. Consider the following example. The total data series is stationary (always sums to 1). You can clearly see that category 10 increased a lot and category 1 decreased. It’s harder to see what is happening to the interior categories, like category 5. It looks like it is staying pretty much the same. Can you notice the consistent decrease from July to Dec (goes monotonically from 8.7% to 5.6%).

These changes are hard to spot because each series is not aligned with itself. What the Dart (ok, fine. the aligned stacked bar chart) does is put white space between each series so that each series is aligned to itself. In the examples I centered aligned them, but they could just as easily chosen to align them to the bottom or the top. With this format you can easily look across the series to see how it changes over time, and make those comparisons. You also can look at a particular time period and see the composition of the total. It works with both with data that changes over time, and data that is stationary. I am not sure why we don’t see this type of graph! I’ve created a hack to make it work in MS excel (contact me if you’re interested) but hopefully this could be adapted into the program to improve chart readability!

Leave a Reply