We do lots of data visualisations at Sandtable. Whether we are doing exploratory data analysis (EDA), cleaning up data or analysing simulation outputs against reference data, there is always a plot to be made.
Personally, I try to visualise as much as I can. In this way I build up my intuition about time series evolution or a distribution of an attribute. It serves also as a face validation of my modelling efforts. It is especially useful when I fit my model against several reference data series, which is the usual case for ABM models.
I also just like plotting interesting data, to build a story into a chart and move it from a bare data presentation (an exploration task), to communication as an infographic (an explanation task).
You might say that the majority of data visualisations aren’t interesting at all, that they are actually pretty dull, and in many cases that would be true. Nevertheless, there are some things that need to be done right.
For instance, the visualisation needs to be presented in the right context, with data expressed in tangible units, e.g. £ or $. The period over which the data is shown (in a time series) needs to be chosen to show what is of interest. I can show the variable of interest alongside something meaningful. I can also play with scales and the colours of dots and lines.
Moreover, whether it is a simple exploratory chart or a more complex infographic, it’s important to be mindful of your audience, including your colleagues. Don’t cut corners and create a badly thought out or a very complex figure and push the effort required to understand it onto the reader. One person’s time saved is many people’s people time wasted. Personally, my motto is, if you share a chart, always make it publication ready. Once it has left your computer, and been posted on email or Slack, you have no control over it. Name it. Give it labels. Make it clean.
Driven by a desire to make my plots more accessible, I went recently for a data visualisation course with Andy Pemberton (@andypemberton) at The Guardian (@guardianclasses). The workshop wasn’t about technical skills and tools but it was led from a more conceptual level perspective. Below, I summarise my main take-outs from the workshop.
1. The most common mistake
The desired workflow should look like this:
Analyse data -> Form an idea (headline, message) -> Visualise
Most of the time we shortcut from data to visualisation. Whenever you are struggling to explain the content of a figure or to answer a “so what?” question, it’s a sign you have skipped the middle bit.
2. Headline
If you have skipped the idea, message formation bit, you most likely haven’t thought about the headline. Quoting Andy Pemberton: “The headline develops the argument to bring someone on the other side of the bridge”.
The headline provides the narrative, the way to read the chart. Its key characteristics are: relevance, timelessness and exclusivity. The headline helps to connect you with your audience, expresses emotions and shows you care. Don’t use question marks, and write plain English.
I still struggle to digest this point on a day to day basis, however, now that I have started forcing myself to think about a figure title or a headline I spend more time refining what I actually want to communicate.
3. Make it about one thing.
An audience can only catch one thing at the time. If someone throws multiple objects at you, you will catch at best a single random one. Make sure which one you want to be caught. Less is more here.
4. It’s not a dashboard
Visualisation is not a dashboard, whose function is to report the state of a system. Visualisation is explanatory, it communicates a story.
5. Make it real, and attach it to feelings.
Your holy grail is to make a chart that communicates through emotions.
For instance we all know house prices in UK have been growing like crazy. How crazy?
“If everything else you bought increased in price at the same rate as house prices since the 1970s..”
– A chicken would cost £51
– A bunch of six bananas would cost £8.50
If you want communicate through feelings your tools are:
Headline, relation to a common thing, colour, size, structure, format.
6. Think about the emotion you want to communicate when you make plotting decisions
When something is big, make it huge. When something is small, make it tiny.
7. Location of the chart
The chart is the proof of your headline. If your graphic is a part of a bigger structure, place it towards the end. Its function is to summarise, to nail your argument.
8. Tone – Benefit – Audience Triangle.
There is no perfect visualisation, there are always tradeoffs to be made. Those tradeoffs are described by tone-benefit-audience triangle.
– Audience: Who is the end user of your work? Think beyond your immediate client, others will use it down the line as well.
Your audience will define the scope: start the process with the audience and then go back to data. Bring in what is relevant.
– Benefit: By knowing your audience you are better equipped to understand what will make a difference to them. Think about which data story is relevant to them.
– Tone: hot vs. cold
There is a bias to create dull, cold graphics. Most charts are on the cold side. They breed a false sense of security. They risk making people fall asleep behind the wheel.
Around the time of the last financial crisis, when the gaps in modelling of economics were laid bare, there was a quote that summarises the complacency of modellers: “no one was ever fired for running one more linear regression”.
I think there is an analogy here: no data scientist was ever fired for making one more dull plot. Remember, being right in your analysis is not enough. You want to communicate it, you want someone to digest your message and catch the point.
Summarising your findings, through a comparison on a chart or a headline is a game of balance. You have to be brave to use it.
9. Colour is emotion
Colours are digested as emotions. Think about the traffic light system, red and green colours. They are used in stock market dashboard charts, red for drop and green for growth.
The other common function of a colour is to highlight. Remember also that the fewer colours needed, the better readability of your chart.
10. Use Plain fonts
Fancy fonts may be not available on every device.
11. If you need to cut the chart, cut it with a thin line.
The thinner the line, the more prestigious it feels.
12. Animations outperform static charts.
13. Ask for help
Asking for help is a super skill I constantly forget about. Getting feedback from other people will transform your work and minimise the chance you are making a plot for yourself.
I created below two examples of what I consider a bad and good figure. The good figures are clean, labeled with relevant units and has a setting the context title.
Example 1, bad figure.
Example 1, good figure. The good figure has a context setting title and y-axis label that describes its units. The is no need for the x-axis label because by using month names x-axis is self descriptive here. The figure is clear, without too many grid lines. The color of the line is aligned to the subject. Fonts are easily readable.
Example 2, bad figure.
Example 2, good figure. Again, the title sets the context and provides the narrative. Both axes are named this time. There is a lot of going on here so the spectrum color scale is used to express the lines’ order. The initial and final lines are bold to highlight the starting and ending years. Legend provides information to individual lines and is not intrusive into the plot. Text and arrows give narrative to the cyclic pattern seen in data.
I hope this will be helpful to you. Below, you can find several links shared during the workshop to give examples of good and bad data visualisations
Should you check your email? An ingenious illustrated flowchart
Gartner’s hype cycle reveals the internet of things is the hottest digital trend right now
The change a baby makes to your life, as seen through personal data
This exquisite infographic tracks two decades of global migration
This Bill Gates infographic reveals the world’s deadliest animal. (The answer will surprise you)
https://www.nytimes.com/interactive/2017/07/17/us/politics/trump-appointments.html
Comments are closed.
Categories
Archive
- September 2019
- August 2019
- July 2019
- June 2019
- March 2019
- February 2019
- December 2018
- August 2018
- April 2018
- February 2018
- September 2017
- August 2017
- July 2017
- June 2017
- May 2017
- April 2017
- March 2017
- December 2016
- November 2016
- October 2016
- September 2016
- August 2016
- July 2016
- June 2016
- May 2016
- April 2016
- March 2016
- January 2016
- December 2015
- November 2015
- October 2015
- September 2015
- June 2015
- May 2015
- April 2015
- March 2015
- September 2014
- August 2014
- June 2014
- May 2014
- April 2014
- March 2014
- November 2013
- September 2013
- June 2013
- May 2013
- September 2012
- June 2012
- May 2012
- April 2012
- March 2012