Pie charts, some people say that you should
never use pie charts in professional data visualization, I disagree.
They say that because allegedly pie charts are very inaccurate, but
there's not real data backing that claim up.
Pie charts are actually remarkably accurate,
and also they provide us with some variations otherwise,
we'd be stuck to bar charts, and line charts.
So I think they do have the area of usefulness,
that said however, the area of usefulness is quite narrow.
For one thing parts should always total 100%,
the percentage sign does not always mean you need a pie chart.
Pie charts only work when you're trying to show parts of a whole.
I sometimes give this task to my students,
women make up to 42% of LinkedIn's workforce,
30% at Google and 33% at Facebook, sketch me a chart.
And inevitably, a couple of people sketch something like this,
which of course is a disaster.
Those parts should total 100%, and these of course don't.
And therefore, these should be three separate,
I don't know, bars perhaps, this of course, is a much better solution.
Second thing, please no 3D and
please avoid legends with pie charts, I think legends are criminal.
Often used for aesthetic purposes,
the third dimension does not really improve the reading of the data.
On the contrary, those plots are difficult to interpret because
of the distorted effects of perspective.
Let me show you what those people at Wikipedia mean.
Now this is a pie chart by none other than Steve Jobs,
and what they ask you something, what is the biggest wedge here?
And most of you would correctly say, well it's the blue one?
But what's the second biggest wedge?
And at this point, most of you will incorrectly say that's the green one,
which in fact is wrong the second biggest wedge is the purple one.
The green wedge of course being Apple.
Now this is criminal as far as I'm concerned, so please do not use 3D,
and the same applies to any other kinds of charts, unless you have data for
the third dimension, which you almost never have.
So, there's a vastly different perception of this chart versus this chart.
And if you really want to highlight that red wedge,
I encourage you to just made the rest of the wedges gray.
This would achieve the same result, without so much lying.
In case you are wondering whether there's any difference between pie charts and
well doughnut charts, there's none.
In most of the studies I saw for instance in this one, both pie charts and
doughnut charts show the same margin of error with,
by the way, bar charts being the least accurate.
And the most accurate kind of a chart is this weird square bi-chart which you can
not produce using standard PowerPoint working instruments.
Thing number three, with a pie chart you can have three to six wedges,
no more but probably no less.
Which means that no, you don't always need a chart,
you don't always need a chart to visualize just one number.
I don't have any research studies to support this conclusion, but my guess is,
that if you just show people one number their estimate will be pretty accurate.
You won't need to show them the whole chart.
So yes, you can highlight numbers, you can really highlight numbers.
But I don't see whether adding pie charts in this
example would produce any additional benefits.
And this is an example where we have just way too many wedges.
As you can see here, we can have once again, both a table and
a chart, which is a sign that there's something wrong with the chart.
Now what do you lose when we remove the table?
And their answer is we'll lose that difference between examiners and SPEs.
What if we try to highlight examiners with blues and SPEs with red?
I don't know if that works, looks like too many shades of blue on one slide.
And I don't think this works very well either.
What I think you should do instead, you should create a bar chart and
then just arrange the bar chart.
I think this is a much better idea.
This horrible example would never work as a pie chart, it should be a bar chart.
And finally thing number four, if you want to compare use bars.
Pie charts do not work if you need to compare wedges, or
if you need to compare two pie charts between one another.
Bar charts work wonderfully,
you can see on the instant which bar is bigger, and which bar is smaller.
So to conclude, parts should total a 100%, please no 3D, please no legend either,
I think legend is criminal as far as pie charts are concerned.
Three to six wedges, no more but probably no less.
And please do not use them if you want to compare things,
pie charts are not made for that.
