MikePadgett.com

Who hates all the pies?

— The pie chart is a common tool for displaying statistics graphically, but the limitations of the format can often hinder rather than facilitate communication…”

I was reformatting statistical content the other day and I came across a herd of lazy pie charts.

As they grazed there on the fertile page all fat, polychrome and insouciant, I remembered that we need to choose our graphical data models very carefully.

As Wainer (2009) so brilliantly reminds us,¹ some models when incorrectly applied can often hide the statistical message they were designed to communicate.

Source: Wikimedia Commons

The original pie

Consider the “original” pie chart produced by William Playfair, the Scottish pioneer of information graphics.² The pie communicates to us the proportions of the Turkish empire located in the respective continents.

This pie does not communicate the actual figures or percentages other than the area equal to 100%, nor does it represent the Turkish proportion of each continent, for that is obviously a different statistic.

What did Playfair want us to infer from his chart?

1. The Turkish empire covers three continents
2. The overwhelming majority of Turkish empire is in Asia
3. About a quarter of the Turkish empire is in Europe and somewhat less in Africa

What is not immediately inferable from the chart:

1. The total size of the Turkish empire
2. The actual percentage proportions in each continent
3. The size of the Turkish empire relative to other empires or the continents themselves

So the pie is almost purely narrative, communicating the array of values, their proportions to each other and the dominant value. It is this latter dominance that the pie communicates most effectively.

When the eye is bigger than the belly

Corpulent pie charts require a lot of visual digestion. If we pass from Playfair’s example to something more florid, the product is much less eloquent.

Here, the pie’s only real strength – the communication of dominance – is rather wasted in the proximity and plurality of the values.

Indeed the creator of this chart did not want us to infer dominance, he/she merely wanted to communicate the values and when represented all those colours and labels, we have to do a lot of work to digest them.

The next pie takes us to the opposite extreme.

Why do we need a pie at all? Most of us can imagine well enough how dominant a figure like 98% tends to be against the remaining 2%.

And nobody sums up better the rest of what needs to be said than that arch-pie hater Stephen Few of Perceptual Edge, a one-man continent of information design.³

Few appends the arguments against three-dimensional and multicoloured presentations: such ‘innovations’ offer us no solution for pie indigestion. Indeed shading and tilt are merely extra layers of crust to cut through.

Cutting to the chase

Returning to my work, I remembered to put myself in the position of my users: so instead of a pie chart, I employed a horizontal bar chart, much like this one:

The advantages of a horizontal bar are numerous, not least among which are the increased legibility of the horizontal alignment and the muting of any dominant-subordinate relationships.

1. It is pretty much impossible to accurately read a pie chart value unless it is a recognisable fraction (for example, a quarter or a half);
2. Colours and 3D rendering can make values appear confusing or inaccurate
3. Pies only really enable us to infer the dominance of some values over others, such that
4. Pies will often cause us to infer different or unintended meanings.

Footnotes

1. In his fascinating book Picturing the Uncertain World: How to Understand, Communicate, and Control Uncertainty through Graphical Display by Howard Wainer (2009) Princeton University Press. Online review at American Scientist §
2. See the Wikipedia article on William Playfair §
3. Stephen Few’s article at PerceptualEdge.com §