Why Tufte Is Flat-Out Wrong about Pie Charts
speakingppt.com
speakingppt.com
Tufte is not only about data viz. and powerpoint bashing. Integrity is central to his work.
I would not trust someone who present me a graph like in the second example (whether as a pie chart or as a bar chart). There is no way to know if the classes have the same attendance. Hence his conclusions could be completely wrong.
The "2 largest distributors control 65% of the market" examples haven't really been thought through... If the point is to group companies B+C together and to group companies A+D+E+F together, why show them separated at all?
The "African-Americans are more heavily represented" example is particularly egregious to me... My first though when looking at it: what do clocks have to do with this? Even knowing that they're pie charts and not clocks, it's far from the best way to illustrate the point. I'm not sure immediately off hand what would be better, but lots of little pie charts is very confusing, especially when the density varies. In this case, if all we're concerned with is the percentage of African-Americans, that's a single value that needs to be shown, whereas the pie chart unnecessarily shows two values.
I think that kind of thing is the point of Tufte's books... If you don't care about some of the data that your graph is showing, why show it at all? It only distracts from the data you do care about. In the case of the two-item pie charts littering that page where you only care about one of the two percentages, a pie chart is a bad idea since it always includes the other data that you don't care about, thereby creating more visual clutter and making it harder to understand.
The second dataset could be done better with stacked bars (eg, https://i.imgur.com/oGzIKED.png). The author seems to think stripping temporal data is a virtue here, for some reason, when it clearly isn't.
The third dataset should be presented with a heatmap on a state (or county) level, a la https://projects.fivethirtyeight.com/mortality-rates-united-...
For one thing, the dataset does not do a good job of conveying uncertainty, specifically that caused by low-population counties. The second-highest mortality rate county - Buffalo, South Dakota - has a population of 1,912 people, and the "City" which is the county seat is remarkably the least populous county seat in the US with a population of 14. The various mountainous locations (appearing to be mostly ski towns) in Colorado which top the list of lowest-mortality counties have similarly low populations.
For another, size and population density are often misleading. That small, orange blotch in the lower left? That's LA County, with a population greater than the 10 lowest-population states combined (including such large areas as Montana, North and South Dakota, Alaska). Information in these low-population areas is over-represented, while the largest cities are under-represented.
I suppose those caveats are just something to be aware of. Because I don't know of a good solution for to how to accurately and concisely convey this sort of information.
I'd prefer a heat map, instead of having to look at many little pie charts.