Are pie-charts evil or just misunderstood ?

OK - I admit it: when I see a pie-chart in business analytics, my blood pressure rises and, yes, I am apt to tell the analyst exactly what I think of the monstrous, hard to read, waste of space and ink they created.

I am most definitely NOT the first person to suggest that pie-charts are over used and close to useless.  Google 'Pie Charts are Evil' and see for yourself.  This is an area where career analysts tend to agree, yet, in practice, pie-charts are very, very heavily (ab)used.  In the hope that I can influence even a handful of people to create fewer of these eyesores I'm adding my voice to the argument.

Let's look at an example and see how it might develop.  I want to display share of sales for my company vs. the rest for the last month.  Something like this:
example 1
It's not very pretty is it ?  Perhaps if we made it 3D ?  (Just 3 clicks in Excel)

example 2
And "explode" it - that even sounds cool (4 more clicks).

example 3
Cool-er, but I want to see 'our' portion at the front so rotate it. (Another 4 clicks)

example 4

Now let's play with the 3D settings, surfaces and add a shadow (about 12 more clicks).

example 5

This is starting to look very cool !  It is a little hard though to see what the share is for 'Vendor G'.    Bit more than 25%?  Less than 33%? But, we can fix this!  With just 2 more clicks we have data labels.

example 6
Now that looks really cool and it's a... monstrous, hard to read, waste of space and ink .

Let me list the ways I find it offensive:
  1. Graphical display is supposed to help us visualize data.  This display manages to visualize "Our share of sales last month was 28.8%" .  Just how much visualization do you need for that ? 
  2. It turns out that we (humans not just analysts) are quite bad at estimating proportions from area.  This was an issue in the very first example.  And there is nothing to help us - pie charts have no scale to reference to so you're stuck estimating based on a few proportions you know well (halves, thirds, quarters).
  3. Guessing the actual share got harder and harder as the chart got 'cooler'.  Going 3D, exploding the section and rotating so that we no longer had at least one part of our segment at 12 o' clock  all made it harder to read.
  4. Adding shadow, highlights and texture probably did not do much harm, but it certainly did not help visualize our one and only fact. (Yes, I do know there are 2 numbers in the chart but as one is easily calculated from the other, essentially we have one useful piece of information)
 Expand this to multiple values and we can examine another issue.

example 7
Is it at all obvious to you which pie segment is related to which Vendor?  Note that I did not devise this color scheme for the example, it's the default that Excel provided when I added more data to the previous example.  If we are trying to spot categories by color they have got to be very high contrast and when we get to 8 different colors its very hard to tell some of them apart.  (The color blind could give up long before 8). In this case, I can figure it out from the thin orange segment which is clear to me.  The others are then charted (and labeled) in sequence so I can figure each one out.  But, does this not seem like an awful lot of work for your reader ?  

What if we add category labels to the pie segment too ? 

example 8
If you think this is an improvement, you should probably stop reading now and let's just agree to disagree.

By the way, those folks that did stop reading at the last sentence are so wrong !

Most of the blogs and articles I have read on this subject, quite a few, but probably less than 1% of the voluminous published material, spend a lot of time telling you why pie charts are wrong but less on suggesting alternatives.  Let's try and address that.

How about this as an alternative to example 8 ?
example 9
This just works.  A simple bar chart immediately fixes 2 of our issues.  
  • We are no longer trying to estimate proportion from area (which we are bad at) but from length (where we are substantially better).  Even without the data labels you could make an intelligent guess as to actual values, more so if I added a few more grid-lines and scale labels.
  • The category label is directly (and much more neatly) tied to its visualization.  No need for multiple colors, no trying to figure out which number relates to which category.
Perhaps its not as "cool-looking" and if you really wanted to, you could go 3D, switch out to pipes rather than bars and apply a texture or shadows.

example 10

Personally I don't think this adds any value for people who actually want to read and use it.  If you are publishing to people who don't want to read your material but may be impressed by the look of the graphic, perhaps it has a place ?

How about this as an alternative, a simple table.  Yes, its primitive, I know, but it really does work, try reading it.
example 11
  Or how about this, a "cooler" version using Excel's in-cell data bars?
example 12
Let's wrap this up by going back to our very first example.  We wanted to show share of sales for 'our' company.  On my screen (your's may well be different) this uses roughly a 2" x 3" space (5cm x 7.5 cm) to display only one useful number 28.8 %.  If that's all the data you have to display and you have such a space to fill, well, maybe, it's ok to use a pie chart.  But can't you display something a little more useful ? 

Example 13
 Same space, much more data and very importantly, much more useful.

So, back to the question at the top of this post.  Are pie charts evil or just misunderstood?  To my mind, they are 99.9% evil and misunderstood - if in doubt, don't use them.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.