Small multiples - when stack bars just won't cut it . . .

Stacked bar charts can help readers make quick comparisons across categories. They can be particularly useful for breaking down responses to straightforward survey data - such as perceptions of services running on a scale from satisfied to dissatisfied - by demographic factors such as sex and age bands.

Formatting techniques, such as using white space to chunk related categories together, using axis labels to report category names and base sizes together, and thoughtful colour choices reflecting the logic of what's being charted, can all help make interpretation of stack bar charts fairly straightforward.  I've demo'd one approach to stacked bar chart design, showing how to use these and other formatting techniques in Excel 2016 to make charts to help non-technical readers out - not freak them out.

But sometimes, a stacked bar chart just won't cut it . . .

Take, for example, this technicolour dandy:

This stack column chart was published by the Home Office, U.K., in "Crime outcomes in England and Wales 2014/15" and it just doesn't work. It's clear that there are some big differences in outcomes between crime types, but it takes a lot of looking back and forth between the legend and the chart to figure out what's going on.  Most readers - probably including some who are supposed to use this chart to inform decision-making - won't have time.

We can rebuild him. We have the technology.

Mmmm . . . not really.  There are some design choices that would improve this chart:

• flipping the order of the legend entries, so that it matches the order in which crime outcomes are displayed, and positioning it to the right of the chart (with some text formatting) 
• displaying the chart as horizontal bars, rather than vertical columns (in which case the legend is positioned above the chart) would make the crime types easier to read, and remove the text that's been rotated by 90 degrees. 
• showing the overall breakdown of outcomes first, not last, would make it easier for readers to see how the breakdown for any particular crime type varied from that for crime overall. 

But even with these changes and more, it still wouldn't be good.

There are too many possible crime outcomes for this data to display well as a stack chart. Small multiples are a better option for a static report: interactive charts should be considered for online use.

Small multiples

I've worked up one option for a small multiple below, but can't say I like it much. It's built out of two charts, with transparent data series used to separate each category.

Figure 2.1: Outcomes assigned to offences recorded in 2014/15, by offence type 

I think it's marginally better than a stack bar: it's easier to see quickly which are the main outcomes for each crime type. But it feels quite cluttered and that kerning is dismal.

Back to the drawing board for this one I think. I'm pretty sure that a small multiple is the way to go, but it's going to need a different design.  More later . . .