As a team we talk a lot about stats and data as rationale for decision making, identification of problem areas or for the assessment of campaign success. However, how this data is shared and communicated is key to its interpretation and impact.
Having recently attended a training course on the use of effective visualisations I have pulled out some of the key learnings below.
What is the impact of data visualisation:
Some real-world case studies show the potential impact of well visualised data.
For example, this chart shows the number of Polio cases per 100,000 people across the US. It shows a clear decline in cases once the vaccine was introduced. The visualisation of this data allows it to be interpreted as a causal relationship and strongly supports the argument for vaccinations.
The second example is that of the ‘climate change hockey stick’ which shows significant and unprecedented global warming since the industrial revolution.
Both of these examples also represent a political argument and show how data visualisation can carry an agenda. It is therefore important to consider exactly how data can be interpreted and what the best methods are for the most desirable outcome.
How is data visualised and interpreted?
Not all data visualisations are as effective as each other. Studies such as that by Cleveland and McGill (1984) show there to be some inherent issues with specific visualisation types in terms of their perception and interpretation.
The main culprits in these studies were pie charts and treemaps as “it is easier to perceive the order of the data by judging position against a common scale (and length) compared to processing angles, curvature and area…”
These examples represent one of the main issues with pie charts, Example A shows the blue section to be 25% of the total, however, despite Example B being the exact same chart, due to its angle it is much harder to interpret the blue section as 25%.
Pie charts are also notoriously difficult to compare, in example A can you tell whether the dark blue section has a larger share than the light blue section below it? If so, can you determine what the difference between them is?
Examples D contains the same data as example E – this clearly show how data can be much easier to interpret in bar chart format. Treemaps can make comparisons difficult, and deciphering shaded areas isn’t the most accessible means of visualising data.
So, seeing the value of good visualisations and the potential pitfalls of bad ones, we can reflect on our own practices to see what processes can be put in place to produce effective and actionable data visualisations.
How to achieve best practice
The main emphasis of the data visualisation course was on best practice for visualisation techniques – this resulted in the following list of golden rules and key questions to ask before getting stuck into a visualisation project:
Early considerations of data
- Is the data fluid or static?
- How easily can the audience access your work – mobile/desktop/print?
- Can the visualisation be interactive?
- Does the data need cleaning?
Key questions – always start with audience
- What do they actually want to know?
- What is their level of expertise/experience?
- What is likely to be well received?
- Will the audience want to drill down into the data – what is their level of interest?
- Will it be viewed whilst the creator is present?
- How often will it be viewed?
Which chart and when?
- A vs B = bar chart
- Composition = pie, tree map, 100% stacked bar
- Trends over time = line graph
- Relationship between A and B = scatter graph
- Data distribution = box and whiskers/histogram
- Flow of source to destination = chord/snakey chart
- Contribution to total = waterfall
- Placed data = map
- Less is more
- Ditch chart junk!
- Don’t overload users
- Only show/highlight where action can be taken as a result of the data
- Prose matters
- Use titles for graphs and tables – intelligent titles which tell the reader how to interpret it
- Annotations can really reinforce a visualisation if the user isn’t that invested or interested
- Tell a story with the data (i.e. therefore…what can be done about this)
- Be consistent in approach/data sets etc
- Use the same/similar colours and layouts making it faster for users to navigate
- Use colour sparingly
We currently have a tendency to produce our data visualisations in isolation, missing out in-depth discussions with the target audience throughout the visualisation process. We often provide an abundance of data (or chart junk!) to make up for lack of specificity.
We will therefore be moving forward with these guidelines for best practice, so that the holder of campaign or website data can produce visualisations which tell a comprehensive story which is specific to the target audience.