In early January this year Stephen Few introduced the concept of Bandlines. He identified a useful extension to Sparklines, making use of shaded or coloured horizontal bands to provide more information on the context of the trend line itself. See Stephen’s article on Bandlines and the thinking behind them for a detail description.
The Sparklines are ideal for showing individual trends in a small amount of screen real estate, and we use them extensively in dashboards, typically in a ‘visual table’. By definition Sparklines are small, and to make the trend easily readable, they are typically scaled individually so that each Sparkline uses the whole vertical axis. This means they do not give any impression of the scale of the numbers involved across different rows. It’s possible to use a common scale, and while sometimes that works more often it means many of the rows with smaller values are excessively flattened.
Bandlines address this by introducing horizontal shaded areas depicting the lower, middle and upper quartiles and the median represented by a line. The user can determine the context of the bands. The two most common examples would be plotting recent trend in the context of a longer period, or plotting individual rows in the context of the overall set of data being displayed.
We think Bandlines add real value, so hats off once again to Stephen, and we’re pleased to say that Bandlines are now available in the current version of XLCubed (see here for more detail).
The screenshots below show two examples, displayed in two colour schemes.
The charts depict historic margin by store. The ‘Banding across all stores’ charts show the 30-day trend for the individual store, set in the quartile context of the data for all 11 stores in the table. We can see that for the Gilroy store in row 1, while the margin has varied, it remained in the top quartile when set against all stores for almost the whole period.
The ‘Banding by store, 90 days’ charts show the individual 30 day trend, set in the context of the previous 90 days for the individual store. This helps provide much more historical context, but the line itself still focuses on the more recent trend. Stockton is probably most noteworty here as across the 30 day period it has dropped from the top quartile into the 1st quartile across the whole 90 day period.
We’d love to hear your thoughts (and also which colour scheme works best!), we will also be adding Sparkstrips in the near future so watch this space.
9 Replies to “Bandlines in XLCubed”
I think bandlines are visually distracting and increase the ink-to-data ratio beyond a reasonable amount. Sparklines alone are much better in my opinion.
I think it depends on the data and the audience. In some cases the bands offer a lot more to analysts, and we’ve been really pleased by some customer feedback in market pricing situations. For sure vanilla Sparklines will always have a place too.
Nice! However, it’s a bit confusing that the legend is in the “wrong” order relative to the rest of the graph.
If light grey or red represents lower values than dark grey or blue, why not use put them lower in the legend too?
Good spot – the legend for the coloured chart example was the wrong way round. It should be as you describe & matching the grayscale example. I’ve updated the blog to reflect that.
The dots confused me at first, since they represent high/low values. Few uses dots for outliers (values greater or less than 1.5 * midrange distance from 25th and 75th percentiles), which provide an additional level of context, particularly if all of the values fall in one band—the main benefit of bandlines over sparklines with just the IQR band.
Good move. Not to steal your thunder, but Excel (free) add-in “sparklines for excel” (SfE) has this feature since 2008…
However, I prefered to use only one band in order not to overload the tiny chart. Stephen Few will certainly notice (some day) that vertical bands also have an interest… Cant wait to read his article…
I think you may be referring to the normal bands (a colored band showing Standard Deviation or other data values), this is something that has also been in XLCubed Sparklines since the beginning. Bandlines are a new feature based on Stephen Few’s blog (http://www.perceptualedge.com/articles/visual_business_intelligence/introducing_bandlines.pdf) they show the Sparkline in context of several bands to show the value distribution.
Not to be pedantic, but it’s customary that the first quartile is the lowest and the fourth the highest.
Thanks for that – we’ve corrected our correction!