2014-12-08
In case you were
wondering what donut and pie charts have in common, it is that neither is a
very good form of data visualization. Let's run through some of the common
problems. In the pie charts above, which
is more, the grey slice or the blue slice?
Which shade of blue you ask? The
one opposite the grey will be a greater challenge. So let's go with that. Now, how much is that difference, if there is
any? Lastly, consider the same for the
donut chart. Is the donut chart easier or more difficult to see the
difference?
Hover over the image
above and you will see the answer to these questions. The unhappy donut chart
was made from the same source as the pie chart. Pie charts may represent the
share of a whole, fairly intuitively, but they are lacking tools for analysis because
it is difficult to calculate differences in area when compared to other forms
of visualization. It takes time to interpret them and sometimes they can lead
to the wrong conclusion, entirely.
1000 Words
So why bother with a
chart at all? Why not just list the data
out in a table like this (which was used to make the pie chart):
Charts can offer an
immediate understanding of the data without having to read each and every
value. In fact, reading a table is far slower than reading a pie chart. So it's
got one up on that, anyway. But let's
take the table above and convert it to a bar chart. I'll match the colors to the pie chart so we
can compare the grey and blue slices again:
It is now quite a
bit easier to see how these categories line up and what the difference is
between them. Part of the reason why we
can read charts so quickly is because we are visual by nature. Many studies of data visualization have
focused on preattentive attributes, these are characteristics that can be
interpreted before conscious thought with a sub-second response before it is
stored in short term memory. Less than
250ms is a common benchmark for considering an attribute to be preattentive.
Some of these attributes are:
- Line Length
- Line Width
- 2D - Position
- Orientation
- Curvature
- Shape
- Hue
- Saturation (Intensity)
- Enclosure
So, for example,
when you look at a pie chart, you preattentively process the differences in
color and shade (hue and saturation). It is easy to tell one slice from
another. And that's about it. Comparing the size of one triangle to the next is more
difficult and less accurate than comparing lengths as we saw in the bar
chart. Area is not preattentive, while
line length is. Let's do another
exercise - what is the difference between the circles below? Hover over to see.
Now, let's try this
comparison using length. What is the difference in line lengths below?
That's right, they
are the same as the circles, but it is far easier to see. Again, comparing sizes in shape is much more
difficult than line length. This is just
the way the human eye works. Some
preattentive attributes are better at encoding numeric values than others. This is why we can safely say, there is always a better choice than a pie chart. When we make the decision about how to create
a report, we can lean on those attributes that work better for the task at
hand. Shape is a great way to
distinguish groups, as in a scatter plot.
Line length and 2d position are great ways to display numeric values, as in bar, column and line charts.
And pie charts are a great tool for demoing challenges in data
visualization :)
m
Reference and further reading:
Labels: DataViz
