Today Andy Kirk and I duel it out in the annual Data Debate. We tackled big and small issues in the field of data. The recording will be available here. This post contains links and further reading on the topics covered.
Are politicians emboldened to say whatever they want? If they are, is that changing society’s attitude to data? Is social polarization accelerating this trend? I fear it is.
Sir David Norgrove, Chair of the UK Statistics Authority, was driven to write to Boris Johnson in September warning him of his “clear misuse of national statistics.” William Davies writes on this in the Guardian, using GDP as an example of the end of statistics. Simon Kuper offers a glimmer of hope with advice on how to take on the populists in the FT. However, his main argument is to ditch the facts and lead with emotion and the story. Sure, that might win people over, but if there’s no role for information and data, have we lost the fight?
The case for animation
My main argument is that animation is exceptional for presenting data in a dramatic way. It enables storytelling and creates tension, surprise, and drama.
The master was Hans Rosling, who astounded us in 2006 with his first TED talk on health data, and later with his Joy of Stats
But I see a lot of people fixate on the chart they want to build rather than focus on the question or the data itself. Watch this video. The end result is a chart you’d never find this in a chart directory. But, if your question is: In which years does B outsell A, it’s a wonderful way to represent it (it’s not the only way, but it does work).
Also consider the two charts below. One shows drought index in the US and the other shows US Road Fatalities. The structure of the data is the same (month, year and state). Only one is a successful chart (the drought chart on the right). Consulting a chart directory, there’s a risk you’ll pick the small multiple and publish. But of course, the data itself drives the appropriate chart type.
Big Ass Numbers: awesome!
Which of the above dashboards conveys a headline better? The one with the BANs or without the BANs?
I came to appreciate BANs through the writing phase of The Big Book of Dashboards. I consider them the Headlines for your dashboard. A well defined set of BANs will capture your KPIs in a way you can interpret instantly. Clever use of colour or other visual indicators can show, straight away, whether they are above or below target. Once the headline is digested, then you can decide if you need to devote time to interpreting the charts in the rest of the dashboard.
This was fun! What I like about defending “bad” charts is that it emphasizes the lack of “rules” in dataviz. There are only guidelines and any binary argument of right or wrong is a failed endeavor.
My defense is simple. If your primary goal is to show the total of all categories, then a stack is fine. Yes, that comes at the expense of being able to accurately see the changes over time of the individual categories. If the primary goal is to see the individual categories, a stack is a poor choice. I believe you should consider primary/secondary goals in all assessments of charts. Remember: All charts are a compromise.
The same defense applies here. If the goal is to give the gist of the data, as they tried to in the BBC article on Apple’s Tax Bolthole, then you’re ok with bubbles. A bar chart will provide more accuracy, sure. In the case of the BBC article, I’m not convinced readers miss out by their inability to accurately see if Apple is precisely 30% bigger or smaller than another bubble.
Welcome to the Sweet Spot. What’s the Sweet Spot? As Tableau’s Evangelist, I need 3 things to do my job well:
I need to know Tableau inside out
I need to know my day-to-day job
But I also need to know how data impacts and is impacted by the world around us.
Keeping on top of all of these gets me in my Sweet Spot. To stay there, I read a lot of stuff. And now: It’s time to share it, fortnightly, with you all.
This week – 3 things to challenge assumptions about AI (hint: robots ain’t coming to kills us. It’s much more boring, in a good way, than that). Thanks to everyone for feedback on switching up the format of this, and for links you’ve been sending me. I’ll include some of those in future mails.
AI is being hyped more than Big Data was. That is deceptive and misleading; it’s time to refocus.
Why read and share? AI is really exciting. But the media swoons over Musk, Zuckerberg and others and their sci-fi prophecies. This brilliant article by Chad Steelberg urges us to dismiss the hype and focus on the reality. Think about our own roadmap: the Data Interpreter and Recommendations are machine learning, in our products now. That’s not killer robots: it’s just a branch of AI doing mundane (but still amazing) stuff to make our jobs go quicker. Ignore the attention-seeking tech titans, and instead get on with augmenting your own intelligence.
Superintelligence: it’s going to destroy us and take over the galaxy. Or is it? Watch this and decide.
Why you should read and share this? Personally, I think talk of Superintelligence is an interesting philosophical game, but closer to science-fiction than reality. The Future of Life Institute do think about this stuff, though. Watch their overview. Then decide: when talking about AI to customers/colleagues, should you focus on mundane machine learning applications of killer robots? [Hint: it’s not robots.]
Want an overview of the current opportunities and risks of AI? This podcast has them all.
Why you should read and share this? AI is coming, and it’s part of Tableau’s roadmap. It’s up to all of us to educate ourselves about the pros/cons and realities. We all need to get on top of it. This podcast is a superb overview of the benefits and risks from the BBC. It’s balanced and doesn’t succumb to the usual hype. It features lots of diverse opinions, including a great interview with Cathy O’Neil, author of Weapons of Math Destruction.
Happy reading! More to come in a couple of weeks. I’ve been toying with ways to do this (LinkedIn, here, YouTube). What do you think? Do you like the content? How would you best like it delivered?
A great pleasure in life, is conuming amazing content across the web. Here’s 3 great things I’ve read in the last two weeks which help me connect data with things happening in the wider world. I hope you like them!
Why should you read this? First because it’s data-driven diversity and second because it teaches us how to have better conversations with others.
tl/dr: An Uber exec resigned over a sexist comment (Paraphrased: “More women on the board? They’ll never stop talking”). The Economist looks to see if any data supports that: it doesn’t. However, the data does reveal fascinating differences between gender’s conversational styles and goals. This is great information for all of us to empathise with others.
Why should you read this? This is about iteration with data: the only way to insight
tl/dr: It is simply not possible to get the best articulation of your data without exploring it and teasing out the best view through iteration. This article excellently describes the iteration process and how it helped develop the best possible end result.
Why should you listen? Because AI and machine learning aren’t good enough to threaten humans yet.
tl/dr: Fitness gadgets suffer from Dead-End Dashboard problems but they’re getting better. This podcast pits them against a real fitness trainer. The conclusion? Like most AI applications, they’re far from fully functional and for the foreseeable future. The point is that for now, augmenting human ingenuity is the right path forward for AI and ML.
Let me know your thoughts – do you have any comments? Do you want to see more posts like this?
Last week, Stephen Few critiqued lollipop charts and I wrote a post in their defense, in which I claimed I coined the term “lollipop chart” when I first wrote about them in 2011.
Stephen Few claims he’d seen them several times before I made them in Tableau. I accept that’s entirely reasonable. I remember at the time I thought my idea was original, and if I’d seen them prior to my post, I hadn’t consciously registered them.
Stephen also sent me a link to the lollipop graph, on Wolprham Mathworld. One of his readers had googled the term lollipop chart back when I wrote my post. The purpose of the lollipop graph is different to that of the lollipop chart, but coining cutesy names is certainly not something exclusive to me (tadpole or barbell graph, anyone?
Which all renders my claim to inventing them hanging on by a thread! I probably wasn’t the first to make a lollipop chart. I wasn’t the first to come up with a cute name. Maybe, just maybe, I can claim the ever-shrinking privilege of coining the lollipop chart!
Problem: A bar chart with many bars of a long length are unpleasant to look at.
I believe lollipops create a visual experience which is easier to look at. In his post, Steve used an example using 7 bars with a large range. I didn’t create the lollipop for that situation. While I believe a bar or lollipop works for that kind of data, criticizing the lollipop without addressing the original intention is disingenuous. I did consider fat bars and thin lines, too, in seeking a solution. Fat bars are just too inky, and the thin lines don’t have enough definition. Lollipops are an attractive compromise to solve the problem.
*Did I invent lollipop charts? Alberto Cairo credited me with their invention in his book The Truthful Art, and I’m not going to argue against that! The images on my original post are no longer available, unfortunately.
When designing objects, be they hotel room taps/faucets, iPhones, or cars, the creators grapple with the concepts of affordances and signifiers. These terms were introduced into design by Don Norman, author of The Design of Everyday Things, based on earlier work by JJ Gibson.
What are these and how can we apply them to our dashboard design?
An affordance is something an object (or dashboard) can do. A tap/faucet can run hot or cold water, for example.
A signifier is an indicator of some sort. In our tap example, this might be red/blue dots signifying which way to turn the tap to get hot or cold water.
How many times have you tried to use a tap in a bathroom and not known which way to turn it for hot or cold? This is a frustration of modern life: apparently a sleek design is more important than signifying (red/blue) the affordance (hot/cold).
Dashboards have affordances and signifiers. How you implement them will influence their success. Let’s use an example. I’m going to use an excellent dashboard by Eric Brown. It allows you to compare gestation periods of different animals.
Let’s play a game. Here are the rules: take a look at Eric’s dashboard and, without using your mouse, identify all the ways in which you can interact with the dashboard?
How many did you count?
There are eight intentional affordances Eric built into this dashboard. Did you spot them all?
How many of those eight affordances have a signifier?
Here are the affordances:
1, 5 and 6 are drop down filters. Drop-downs are a staple of interacting with dashboards, and web pages. But why put the three filters in different places?
2, the light bulb, is a hover-help tooltip. Hover over the light bulb and you see a description explaining how the compatibility score is calculated. That’s great if you’re familiar with the “hover-for-help” trope, but if you’re not, then, well, it’s just a light-bulb. How would you know it contains an explanation?
3 and 4 allow you to click on the animal to highlight it in the scatterplot and see more details.
7 is the colour legend. If you click on one of the colours, it highlights all animals in the scatterplot in that category. Does the dashboard tell you you can click on the legend?
8 allows you to click and see the datasources.
Note – did some of y ou think you could click on the silhouettes of the whales in the top right? If you’re a Tableau expert, you might have thought you could. But, no, that is just a legend. It has no interactivity.
That’s a lot of stuff you can do with this dashboard. But only some have signifiers.
We could solve the problem by ensuring every affordance has a signifier. Here’s what that could look like:
Here are the main changes I made:
Move all drop-down filters into one place
Added instructional text to the scatterplot and colour legend
Removed the hover-help tooltip and placed the calculation explanation in the bottom right, above the data source links
I could now claim to have fixed the “problems” with Eric’s dashboard. Anyone now coming to the dashboard with no prior training in it, or dashboards in general, now has a signifier for every affordance. If they invest the time in reading the dashboard, they will be able to interact fully.
Should I always create a visible signifier for every affordance?
No. Sometimes you are designing dashboards for an internal audience. They may be familiar with dashboard interaction, or you can train them. In which case you could remove the signifiers.
The thing you need to do when designing a dashboard is consider your audience, and how you can communicate to them that they can interact with the dashboard. Skilled users know to click and experiment, or can be trained to do so. New users don’t have that knowledge or confidence. Your job is to make these decisions consciously, not by accident.