On 24 June 2011, Business Insiders featured the display below
as their chart of the day—a page that even made LinkedIn Today
(“the most shared news on LinkedIn,” they say). Let me therefore
similarly discuss it in this blog entry as my own chart of the day,
but for very different reasons: this graphical display exemplifies
several shortcomings typical of the charts produced these days.
First of all, the length representation (columns) is inappropriate.
For evolutions, lines are a better choice—something the author
of this chart probably realized at some point so he or she added
the two dashed lines, which I assume embody linear regressions.
These regression lines are rather pointless, though, as they are
not used to show the rate of change: the two numbers displayed
are just the difference between the last column and the first one.
In any case, a length representation demands that the scale start
from zero. The truncated columns shown here suggest changes
of about +345% and -45%, respectively; that is, they exaggerate
the actual changes by a factor five. Although the exact numbers
are displayed, viewers will most likely walk away from this chart
with a distorted view of the actual data. This chart is a visual lie.
Finally, several details can be improved. If you do use columns,
at least remove the distracting shadow. Simplify the two scales:
include fewer values. Question the usefulness of the grid, too,
given that the change is displayed explicitly already. And write
a more accurate label to describe the data: the “total minutes”
are actually not shown anywhere in this relative representation.
One positive point nonetheless: the two data subsets are labeled
directly on the display, not in a separate key defining what blue
and red mean here. These labels help keep the display intuitive.
A more effective display—one that would highlight the evolution,
be true to the data, and exhibit a higher signal-to-noise ratio—
might look something like this (variations are of course possible).
For displays of relative values, such as this one, we might argue
whether a logarithmic scale would be a better choice, as it avoids
compressing decreases compared to increases (it would display
a two-fold reduction and a two-fold growth at equal distances
from the 100% reference line). The trouble is, a log scale is not
very intuitive for most people: it is, in a sense, an acquired taste.