A chart's a chart, no matter how small

Sparklines are minimalist by nature - often without labels on the axis. Their intention is to show at a glance the trend for that series.

But minimalism doesn't equal less thought. Nor should we accept our software's defaults on how these mini-charts are formatted and scaled. For example, Tableau defaults to using the same scale for each line chart, whereas Excel's sparklines default to independent scales.

In this week's Makeover Monday, Andy Kreibel produced this lovely dashboard based on the Chicago taxi dataset.

Click to see the original Tableau dashboard. Very Few-esque!
 
Within the dashboard, Andy uses sparklines to present the changes in monthly trip volumes over a number of years, for the top 10 areas of Chicago in terms of volumes:

Each sparkline has its own independent range for the Y axis, set to minimum and maximum value for each area. By choosing independent Y axis ranges, the variations within in each sparkline are accentuated so you can see more clearly the changes in volume per month for each area.

However, something that has been bothering for me a while (since working on a dashboard that also had sparklines with independent Y axes), is how this might be perceived by the reader.

Typically, if I see a series of charts that look the same, with no additional information about the scale, on first glance I would expect them to be the same.

Having the same scale allows for ease of comparison between each chart. Once I understand the first chart, I can subsequently flick between the charts to look for comparisons and anomalies. If the ranges change, then all I can do is set my context to the latest month's figure for that series and see if it has gone up or down.

By using the minimum and maximum for each series, some sparklines may misrepresent the trend. Consider these three series, where the sparkline's Y axis has been set to be independent from each other:

Sales per monthJFMAMJTrend
Product A00091011
Product B201020102010201920202021
Product C000182022

Stephen Few, a proponent of sparklines, has a very useful article on this issue (PDF), and the article's subsequent discussion contains this table for making decisions on how to scale your sparklines:

So based on this reading, and some other playing around, here's some variants on Andy's original sparklines for your consideration:

As with most decisions in data visualisation, the answer is: it depends. The basis of your decision should be: the message you need to convey; and the audience's expectations- do you want them to compare between series? Would they expect to?

Out of the options presented above, I would opt for the percentage change with the start date of December 2014. This complements the story I think Andy was trying to tell.


PS. Since starting this post, I saw this design by Kevin Manuel for the same Makeover Monday challenge, which uses the sparkline design I would choose. Nice!

Click to see the original Tableau dashboard.