a small (multiple) makeover for a big range problem

Today’s post is about a common challenge: when one data series is so large relative to the others that a single scale makes it nearly impossible to see any details. Consider the following line graph. It displays state and local revenue by transportation mode, which I created using data from the Bureau of Transportation Statistics 2018 Report.

What can you conclude about this data?

I can see that Highway represents the largest portion of total transportation revenue, earning much more relative to the other modes. But figuring out how revenue has changed over time for a given mode is challenging since the lines look flat on this scale. Let’s look at a couple alternative views of this data.

I could improve visibility by pulling apart the four lines into distinct graphs—creating a small multiple chart. The separate charts allow me to set custom y-axis ranges to better display the trend for each series, as shown in the following visual.

Taking a step back, I can now make observations that were difficult to see in my original graph. Highway revenue has recovered to pre-recession levels, while Air transportation has experienced a slower recovery. These details weren’t obvious before, so from a visibility standpoint, this is an improvement. 

However, while this solves one problem, it simultaneously introduces another. In the transition from a single graph to four, where I’ve varied the y-axis min and max, I lose the ability to compare the relative sizes of each transportation mode. My small multiple chart requires some careful attention to detail and mental effort to process the data plotted along the different axis start and end points. It is exactly because of this, that a general guideline with small multiple charts is to have a consistent axis across panels. Unfortunately in this case, however, I can’t implement a single scale without sacrificing my insights.

How might I account for the drawbacks of an inconsistent scale, while still preserving the ability to see important details? Let’s look at one more view of this data.

Vertically stacking the charts and arranging with the lowest dollar volume at the bottom to the largest at the top, and directly labeling the start and end points is one potential compromise. While I haven’t completely solved the problem of visually comparing relative sizes—which I’d argue is secondary to seeing how revenue has changed over time for a given mode—I did design with that ability in mind. In this iteration, I can quickly scan down each panel to put into perspective that $78B is greater than $4B. This makeover is an example of the tradeoffs we often face, which is why it’s important to be specific about what you want to communicate and keep that in mind when you build each graph. Some other changes I’ve made in this final version: I added a recession bar to easily compare pre- and post-recession values and annotated takeaways, so someone processing this chart can quickly glean the key points.

An unexpected bonus of the stacked version: it doesn’t differ too drastically from the original. Sometimes creating a makeover that resembles the original can work to your advantage, especially if you need buy-in from others to make the change. I’d call that a win! I can see details, compare relative sizes—sort of—and outline specific takeaways to get my audience focused on the right things.

Employing a small multiple view to split data apart certainly isn’t the only solution to the big range problem, but it’s one that I find myself using frequently. For additional small multiple inspiration, pop over to the community, where that’s the focus of this month’s SWDChallenge: leave datapoints on examples you like, lend feedback, or participate by creating and sharing your own small multiple chart!

You can download the Excel file to see how I created these graphs.