Please, no pie in the cube!
I really like Guy in a Cube. If you don’t know Guy in a Cube and you work with Microsoft BI tech – specifically Power BI – you should definitely subscribe to the channel. They make day-to-day boring tutorials and tech updates fun and upbeat and that is no small feat. Being able to talk about topics like how Power BI works with Active Directory in a way that is fun, entertaining, engaging and informative is a true talent. I particularly like the videos made by Patrick LeBlanc. His pace and on camera manners are excellent and he makes very complex concepts appear clearly. Really, usually the only thing that Guy in a Cube need to improve is to have, every now and again, a girl in the cube. I'm sure there are excellent BI professionals who aren't guys.
However, Patrick’s video from November 8 about drillthrough in Power BI had a 10 second bit in it that I couldn’t disagree with more.
It starts at 1:25.
Over here I have a map, and I have a pie chart. Love the pie charts. For you guys that hate the pie charts, that’s your problem. If my end user asks for a pie chart I give ‘em a pie chart, right? You give ‘em what they want
Well, I’m sorry Patrick, but I will have to disagree.
What is it I disagree with? Well, firstly, I disagree that it’s a question of love and hate. I don’t have particular feelings towards methods of visualisation or particular chart types. I ask myself one question about any visualisation – is it telling the story of the data well? If it does, it belongs on the dashboard. I have found, through experience, that pie charts never do. More on that later on.
The second thing I disagree with is the reactive approach to users. I’m a consultant. This means companies pay for me to come by and provide expertise where it is missing. Almost by definition if I’m in an office, it is because I know more about the problem and solutions at hand than the other people in that office. If my end user asks for a pie chart, I ask why. I also do that if they ask for a bar chart, a scatterplot, or a heatmap. I ask why because I want to understand their considerations, and then add them to the list of considerations I already have when choosing the best way to display quantitative data. And then I determine what the best display for the data is, and I use that, and I take them through my list of considerations. Because I made my decision not based on love or hate of this chart type or that, but based on a careful study of what is the story the data is telling, and how to best present that story, I can explain my decision calmly and rationally.
If after that the end user insists that I use their preferred type of chart, I explain the disadvantages of their favourite flavour. I also suggest that my knowledge of the domain is a result of research, quote relevant research, show examples, and I might even throw in a comparison to asking a lawyer to put in particular language in a contract, where the lawyer suggests an alternative saying that the particular language suggested might not be optimal. Obviously, if after all this back and fourth the customer is adamant, then yes, they will get the chart they asked for – people deserve to get the labour they want for their money – but it would not be a simple case of ask and get. I will not be doing my job properly if I didn’t mention that the visualisation selected isn’t an optimal one.
In the introduction to Show Me the Numbers, Stephen Few reminds us that with the ability to create charts with ease, that came with personal computers, we also got a surge of bad charts created by people who were never trained in ways to display quantitative data. I think this is increasingly correct in the age of self-service BI systems. In the age of Power BI and its competitors, users need to be trained in data visualisations when they create reports and dashboards to be used across the organisation. The first opportunity to do so is often given to the consultant assisting with the BI implementation.
One might ask why is this so important. So the data will be in a slightly less readable format, so what? I’ll try to answer that, and in doing so I will strive to show why pie charts aren’t very likely to be telling the story of the data properly.
We use charts and other visual representations of data because they tell the story better than words and texts. It’s almost impossible to detect a trend for a single measure over 25 months. But a line chart with 25 data points makes that something you could do at a glance.
Just as an example, check out the photo below (Source), and consider how complex the story it tells, using one image, is.
Visualisations, therefore, only add value when they allow us to unlock the insights that the data hides better than actually reading the data. It is very seldom that a pie chart is able to do that, and here’s an example I use to show why.
I have simulated a fake survey where people voted on their favourite Dr. Who actor from the new series. Below are the results in a grid. Measuring the time it takes, use the grid to determine: who is the most popular Dr. Who actor? Who is 3rd favourite? What’s roughly the ratio between the least favourite one to the 3rd favourite one?
How long did it take you to answer these? This usually happens in about 5 seconds. The answers are, of course, David Tennant is 1st, Peter Capaldi is 3rd and Jon Hurt has about a quarter of the votes Capaldi has.
Now, I kept the data EXCATLY the same, but replaced the labels, so that my mock survey now represents favourite pizza toppings. I do this because it’s a good way to help you ignore what you already know about the data as you approach it. Again, time yourself as you’re answering these 3 questions: What’s the most popular topping, what is the fourth popular one, what’s the ratio between the least popular one and the most popular one? If you exceed the time it took you to perform the task using the grid, stop. It means the pie chart is less useful than a simple table of data.
I have yet to meet anyone who managed to do the pie chart data extraction faster than the grid. If you have – leave a comment!
If you still don't know the answer, or unsure of it, look at the same data in a bar chart format. Try to answer the same questions: What’s the most popular topping, what is the fourth popular one, what’s the ratio between the least popular one and the most popular one?
Notice how the data just lends itself. Looking at the chart is already knowing the answer. We immediately see Mushrooms lead, sweetcorn fourth, and pineapple has about a fifth of the votes mushrooms have.
This is the goal of visualisation. It’s not about loving or hating a certain chart. It’s about making the data lend itself to the user, and making their lives easier by giving them a clear understanding of the data they have. It’s about telling the story. Just like Guy in a Cube are awesome at telling very technical stories in a compelling manner, our job as BI/Analytics/Big data/Data science consultants is, among other things, to display the data in a way that tells the story best.
All the data visualisations were created using Pyramid Analytics 2018, a very interesting release which reinvents the much acclaimed Pyramid Analytics BI tool. However, the bar chart was created using Power BI. Please feel free to join the conversation in the comments!
Luis Quedas :)
This is very interesting read, and made me care even more for charts I use. Poor pie chart, not my number one go to for simple visuals anymore. Thanks
(CONT...) Last comment about the pizzazz – I know it was said half-jokingly but it’s a very common misconception. Doing things so that they’ll look “flashy” or “shiny” or “full of pizzazz” is marketing and sales talk. All major BI vendors do it, all of them are dead wrong to do it – especially those who sell products meant for self-service (sorry Microsoft). People who need this to work want to see the data in the best possible way, not to have the shiniest, most vivid toy. They don’t need to be engaged by the data – they are engaged by their need of it.
(CONT...) But not all is lost for the pie chart. Where the magnitude of the category that stands out matters, two options are available: One is to use an indicator to indicate that this category is over a certain threshold. Say, light up a red dot when sales in the US drop under 80 percent of total sales. If a closer look at the standing out category’s part of the whole is necessary, then we get the one scenario where pie charts can be useful – we view our category as one slice, and all the rest as another. When a pie chart only shows two values, its shortcomings don’t matter as much. So for example if creating a set of dashboards to sales team managers, we can have a pie chart for every manager where they see the part their team is of the whole. The again, we can just display “32%” as a text which would work just as well.