What is data visualisation and when should it be used?
Data visualisation refers to the graphical representation of information and data. It’s a way to visually convey quantitative results, assisting users explore, understand and communicate at higher speed and accuracy. Such visualisations are useful when using a pictorial view uncovers meaningful relationships otherwise not easily observed from a bland table. We must remember however, just because we have numbers doesn’t mean we should transform these into illustrations as sometimes (for little data) the numbers directly are more efficient and impactful.
Why is it important?
For companies looking to harness complex statistical data analysis for business decisions, data visualisations are a real asset. At the outset, they appeal to our brain, aid in data exploration, remove complexity to create meaning, reduce time to insight and help drive data-driven cultures.
Our brains process perceptive information quicker than cognitive information
According to neuroscientists at Massachusetts Institute of Technology, the human brain can process an image in just 13 milliseconds. We are hardwired to prioritise visual information over other incoming stimuli and can do so at a much greater speed (partly because we can simultaneously process multiple images, rather than one word at a time for example). As such, using graphs and charts to convey large complex data is far easier and more effective than pouring over figures in spreadsheets or report papers. Cornell University states if a scientific claim is presented in pure words or numbers, 68% of people will believe the information is accurate and truthful. However, if you put a simple graph with the claim, the number rises to 97%. Another interesting point as to why visuals suit our brain lies in our inability to comprehend large numbers. While we understand big numbers intellectually, they can often seem like abstractions making it very difficult to conceptualise the data. Data visualisation assist us to better relate and ‘feel’ the size as simply saying a large number may not have the desired impact.
Visual data exploration assesses the quality of your data and highlights important aspects
It is essential to ‘look before you leap’ as much statistical analysis is underpinned by assumptions of the data. There are many examples of multiple datasets having the same summary statistics but vastly differing in shape. Often analysts ignore this step and jump right into running the data through a regression or clustering model which can overlook important features as the data doesn’t fit the assumptions of the model. This initial visual step can immediately identify outliers, show the shape and illuminate correlations to enable the correct application of data science techniques.
Good data visualisation turns complexity into clarity to create meaningful information
The importance of this point increases as we see the backgrounds of the audience expand. For many non-technical end users these visualisations will be the only interaction they have with data as they remain unfamiliar with the intricacies and nuances embedded back of house. Insights are worthless if they can’t be comprehended or consumed in a useful way. A good visual helps get the point across and allows many to understand a complex problem in a simple way. Communicating data findings to key stakeholders is a priority challenge, placing the ability to digest and translate this information into effective business decisions at the forefront of the value proposition.
Good data visualisation also helps digest masses of data in a short amount of time
As the data deluge continues there is a narrowing time for information processing. To combat this, data visualisations enhance users’ ability to interpret high volume by making it easier and simpler to recognise key trends and patterns. This, in turn, leads to faster response times and more effective decision making. The powerful advantage of pattern recognition over non-visual displays has propelled the shift towards popular self service dashboards as they help users quickly consume the most relevant information to continue to perform their role and the tasks at hand.
Appealing visualisations aid in driving a data-driven culture
As mentioned in my previous post, when data is represented in a visually appealing manner it grabs employees’ attentions. Whether it be through exotic charts, interactive graphics or new tools, data is brought to life which builds momentum through engagement. In saying this, it’s imperative to keep in mind the end goal is to enhance the data through design, not draw attention to the design itself.
Principles to follow
It came as no shock when the study Blinded by science concluded “Most people will believe whatever you tell them as long as a chart is involved.” Given this unyielding power, it’s our duty to design and use them correctly by applying principles of best practice, namely those drawing on human perception and cognition. Data visualisation should speak for itself, telling a story and improving the understanding of the audience. Unfortunately this is not what we see today. Despite many charts containing intelligent information, their presentation confuses rather than enlightens which results in the intended message being lost.
Communicating with data is no easy task but if you keep the following principles in mind, great analysis can be combined with great storytelling.
- Know your audience. What do they stand to learn and how will they use the piece? Tailor the design, emphasising the message to align with their priorities and level of expertise.
- Choose the right data. While this may sound obvious, sometimes different datasets can illustrate and support your point more effectively.
- Identify the right visualisation. What type is best suited to convey your message? Design for comprehension and don’t be afraid to refine until it clearly communicates your intention. Understand the difference between qualitative and quantitative data, knowing which visuals work best with each and why the following are my recommendations:
- Changes over time – line
- Magnitude – column/bar
- Ranking – ordered column/bar
- Distribution – histogram
- Correlation – scatterplot
- Part-to-a-whole – stacked column/bar or treemap
- Deviation – diverging bar
- Spatial – basic choropleth (rate or ratio)
- Don’t distort the data. It’s crucial to avoid presenting a misleading view as accuracy is the most important aspect of your design because it exudes trust. For example, manipulating axis ranges can be deceitful and show artificial fluctuations resulting purely from design.
- Use colour with a purpose. Colour should be used beyond aesthetics to communicate information, allow you set the mood and focus attention on particular features to tell the story. Used effectively, colours assist users to rapidly identify patterns and understand insights (heat maps are great examples of this).
- Use labels. Adding brief, relevant text allows viewers to focus on the key message rather than trying to understand the chart in its entirety. Words are not written to hijack attention but more to remove any guesswork on behalf of the user. Always have a title and don’t have text at an angle (it breaks the viewer’s neck!)
- Less is more. Insights are best consumed in a clear and simple way and excess information or ‘chart junk’ will act as a turn-off. Data should be reduced to the most important points as the clearer and more succinct the message, the more memorable it will be. Avoid the temptation to use fancy or complex layouts for visual appeal.
- Make it quick and easy to interpret at a glance. Visual data relies on cursory analysis which is measured in seconds, to be digestible it must be effortlessly interpretable. More importantly, in static visualisations, it’s beneficial to lead the viewer’s eye to key points. Common tactics involve mimicking movement in a ‘F’ pattern (which is how people read; top left to right then down) or using colours to direct their gaze across the page.
- Ensure it tells a story. According to studies in the Heaths ‘Made to Stick’, stories are more effective than statistics because data isn’t as memorable as stories and stories are more persuasive than statistics. The end goal of your data visualisation should always be an action, tie your message to business goals and ensure your image involves a strong narrative.
More specific pointers include:
- Avoid pie charts – Humans can’t make accurate estimates or comparison of angles, instead length is more effective (column/bar).
- Avoid the double ‘y’ axes – As the scales are arbitrary, we are tempted to compare meaningless magnitudes which can deliberately suggest false relationships (instead use a single y axis chart with the ratio as the metric).
- When using tables, align right – This assists the user to scan and perform row to row comparisons with ease and simplicity.
- Avoid 3D – Unless a third dimension is actually being plotted, it only makes comparisons more problematic.
- Avoid spider charts – The viewer tends to focus on the shape which is determined by a non-ranked category order (instead use a ranked column/bar to compare values).
- Compare areas or volumes as a last resort – When comparing two or more dimensions at once we rarely make accurate estimates as we instinctively judge the lengths or widths in the first instance.
- Understanding different metrics for comparison will yield different results – Additive difference, relative change, multiplicative change, log of the multiplicative change, etc.
- Watch out for the Simpson’s paradox – When a trend appears in several different groups of data yet disappears or reverses when these groups are combined.
Where to in the future?
Data visualisation combines art, science and technology. Advances in virtual reality and augmented reality have unlocked and unleashed new capabilities allowing datasets to be explored and analysed in a way even more fit for purpose. Companies can now create 3D intuitive environments with the ability to digest higher dimensional data and identify previously unavailable and unimaginable patterns. Manipulation of the data takes place through interactions (dragging, zooming, combining, etc.) which further accelerates the articulation of datasets and time to insight.