When being correct is not always correct

[Alexander_Thor]

When you start getting into the Data Visualization field you quickly learn that there are good visualizations and there are bad visualizations. Most scorned are probably the horrible pie chart and its cousin the donut chart. Should we follow Stephen Fews advice and save the pies for dessert or is there a time and place for sub-optimal visualizations?

With data visualization celebrities such as Edward Tufte and Stephen Few being very vocal in their crusade against bad visualizations the rest of the industry has started to follow suit. BI vendors have slowly adopted and almost everyone is promoting data visualization best practices now a days.

I'm not saying they are wrong, a bar chart or line chart for time series are always a better option than one or several pie charts when the core objective is to compare data points.

But is that always the core objective?

Sometimes we build QlikView apps for very large audiences, apps they might only use once in a while, apps that aren't critical for them to perform their job and sometimes we build apps that contain downright “boring” data. It’s still an important app; the users would more than likely gain additional insight from the data or the app would help them perform their job more efficiently.

Looking at myself I know there are probably several applications that Qlik has deployed internally that could help me in my job. They aren’t critical for me, I would probably only look at them once a quarter or less but still I don’t open them at all.

These would be apps with “boring” data, apps built according to every best practice in the book. Absolutely no pie/donut charts, muted downplayed color series from http://colorbrewer2.org/ and consisting to 99% of bar charts and data tables. They are in no shape or form bad apps, they are built around solid best practices, every data point is correct and every visualization carefully selected to achieve the maximum efficiency but still I can’t get myself to spend more than 15 seconds in them.

What they are lacking is attention.

Attention vs Accuracy

I want to make the case that sometimes it’s appropriate to sacrifice a certain degree of accuracy for attention. Sometimes you need some sex and sizzle to get your users to care at all.

For example, humans are naturally drawn to rounded objects versus squared yet our brains are not wired to quickly grasp the sizes of a pie chart. Despite the logical part of my brain telling me that the bar chart would be a more optimum medium to display the information my eyes are still drawn to the pie chart.

The Bar Chart makes it easier to compare the individual values against each other but for me the pie chart is more inviting.

This holds true for pie charts and it also is true for maps. In this day and age as soon as we have an address or a location in our data we are compelled to put it on a map. Why? Most of the time the geospatial dimension is totally irrelevant to our analysis but we still squeeze a map in every application that we can. Because maps engage the users - you can put almost any boring data set on a map and I would still explore it, I will most likely not gain the most knowledge out of the map BUT my interest has been sparked and I might explore the data and the app further.

So should we go wild and crazy, sprinkle every app with pie charts, donut charts, bubble charts or maps?

Absolutely not, while at the same time we need a certain degree of attention to get our users to take interest in the data; we also have a responsibility to represent the data in the most accurate way possible. But what good is the data if people won’t take any interest in it at all?

I say it’s okay to stray from the path of best practice as long as you are aware why you are doing it.

Michael Anthony has previously blogged about Progressive Disclosure which can be used to overcome the initial attention hurdle while the rest of the application can focus on delivering as accurate representation of the data as possible.

TL;DR Pie charts - bad. But sometimes good.

[#2]

I have never been able to put to words the reason why I still put in a pie chart or two in my qv af

Pps... But you described the raison perfectly. Good post, i completely agree!

[stevedark]

Nice post.  I will allow the occasional pie chart into my apps - but where I draw the line is 3D effects.  These tend to not only be sub-optimal but can actually mask the actual values.

[kouroshkarimi]

I understand the principle of your argument, but I'm afraid I don't agree. There are other ways to make your application more attractive - I try to be more liberal with colour schemes, UI elements (e.g. pop-ups), images and text.

If data is boring then adding superfluous elements to a visualization just makes it not only boring, but also hard to read.

[Marcio_Campestrini]

Nice post. It shows in a great way why put (eventually) a pie chart into an appplication.

[ThornOfCrowns]

Your example of 90/10% would be better illustrated by a single number, I think, but if you say Pie charts with > 2 values bad, then I'd agree wholeheartedly.

There is a middle way between Tufte/Few and McCandless and it is important to remember that the idea behind a visualization or dashboard is to conver meaning and tell the story of the data. I'd never agree that it's "sometimes appropriate to sacrifice a certain degree of accuracy for attention" How far is a certain degree?

[Alexander_Thor]

Hey James,

For me a certain degree is just enough to get your user/reader to actually use your application.

A lot of the Tufte/Few dogma is based around the fact that the data is critical or highly relevant to the user and that just isn't the case. Let's say you deploy to 1000 users, maybe out of those 1000 you will have 100 users that will depend on your app for their day to day. Maybe another 500 that could benefit from it and 400 that that's "meh another dashboard" so the core point for me is how do you attract those 900? For the 100 that use it in their day to day they will/have use it regardless.

And if we would never sacrifice accuracy for attention we could remove all charts types but bar-, line-charts, scatter plots and tables .

[Alexander_Thor]

Hey Kourosh,

Ye I agree, you should not something that's superfluous. But there is more than one way to visualize information.

And I have to disagree to a point around UI. Nowadays it's mandatory to have a useable and beautiful UI.
We have reached that level of maturity in the industry where a "good" UI is the norm and a "bad" UI is just distraction where if you look only a few years back it was the other way around.

[ThornOfCrowns]

Hi Alexander,

thanks for the reply.

To use your example, if you're deploying to 1000 users and you're trying to get 90% to even look at the app, you're fighting the battle at the wrong point, I think. No small amount of obfuscation / sizzle ( delete as appropriate ) will bring a large percentage of that 900 across.

I know that's an extreme example, but in general, the best way, in my opinion, to ensure your dashboard / viz is used is to win the 'hearts and minds' of the end users before they get to the "meh..." stage.

[Alexander_Thor]

Yeah it's a extreme but also a real scenario.

Sometimes I use the lovely P&L statements as a good example. It's an extremely important piece of information for every employee within a company. It drives all kinds of soft values like transparency, culture and alignment. But oh god it's a boring piece of data, show me a another P&L table and I will turn to harakiri.

So how do we create a P&L statement that's engaging to a wide user base but still accurate? The pie chart is most likely NOT the way to go and I'm not advocating it's use as a go-to chart as it 99% of times aren't.

Actually that could be a fun competition - how to you turn a P&L into a engaging piece of information for a entire company. Hmm...

[kouroshkarimi]

Hi Alexander, thanks for the reply. I totally agree, you should never sacrifice good UI / UX for anything. When I look at the early demos available on the QlikView site compared to what is being produced now... it's night and day. Thankfully Qlik is starting to promote best practice!