Long-time readers of this blog know I’m a big advocate for using data visualisations to better narrate the ‘stories’ that are hiding inside organisational data. My interest was therefore sufficiently piqued when I came across this article on the Bulletin Expert Contributor Network. And while it doesn’t make any wild discoveries, I enjoy the way it provides seven ways to improve data visualisations.
I fully agree with the author’s statement that we often spend too much time and effort on improving our data manipulation and analytics skills while neglecting the visualisation side of things. Having good visualisation skills is crucial as it enables us to convey insights to non-technical, business-oriented decision-makers in a user-friendly and memorable format.
Keep it simple
While I hate the ‘stupid’ added to this common acronym, I do like the way the author uses the data-to-ink ratio as the measure to drive this point home. Way back when I did my data visualisation course with Stephen Few, we spent a lot of time focused on this and it is still very relevant today.
Choose the right chart
You must use the appropriate chart to get the right message across. The article shows which chart types work best in several practical scenarios. Although they mention line graphs under continuous nominal data, I would explicitly call out the use of line charts for time-related data.
While the article rightly mentions that you can use pie charts for small amounts of categorical data and for comparisons, we were instructed never to do this as the eye-brain coordination is bad at comparing shapes that aren’t horizontally or vertically aligned. We were also instructed never to use any 3D charts for the same reason.
Visualise one aspect per chart
If you try and display too many aspects on a single chart, the messages can often get mixed up. Having said that, some of the better visualisation tools let you easily use three aspects of a single variable on a single chart. For example, you can use a bar graph to represent the counts of a particular variable, and you can use the widths of the bars to denote another aspect, and the colour of the bars for a third aspect. You can even add average and median lines to the bars to give an indication of spread so long as it is all clear and easily understood.
Spice up your axis range
This part of the article is well presented with illustrations on how the appropriate use of the axis scale can be used to emphasis key differences, especially if those are relatively small in comparison to a larger measure, such as totals.
The article refers to this as spicing up the axis. I think that’s taking it a bit far. This is only about using the appropriate scale.
Use transformations to emphasise change
Again, there are nice, concise explanations with illustrations on how data transformations, such as logarithmic scales, can be used to emphasis small differences.
Scatter your points in your scatter plot
Another catchy heading to indicate how you can use different icon types and opacity when you have a lot of points scattered closely together. In more advanced visualisation tools, you can also use icon type, icon size, and colour density to convey related meaningful information about the variable being plotted.
Pick your palette wisely
This plays directly into the data-to-ink ratio as well when it comes to the attractiveness of the visual and the ease with which it can be used. When we first start designing visualisations, we tend to use bright and contrasting colours. This approach might be fine for a once-off proof of concept, but most people will get put off by the overly bright colours especially when using the graph multiple times per day. A softer, pastel-type palette with enough contrast is often softer on the eye and more comfortable to work with in the long run.
And then there is the type of device the visualisation is targeted for that must also be considered. Something that displays well on a 19-inch screen might not work so well on a smartphone, and vice versa.
My two cents
Overall, this is a nice and sensible article which highlights the key aspects to consider in visualisation. It does so succinctly, arranged under some catchy headings. I’m always glad when key information gets refreshed and re-presented to keep on reinforcing the points, as well as to present them to data scientists that are new to the field.
