Nonsense Charts

Daily Dose of Excel shows us an excellent example they found of awful chart design. The chart concerns male and females accepted to various colleges, the class size, number admitted, and number of applications.

It's unclear what the 'take home' is supposed to be - that some colleges have lots more applicants, that the acceptance rate varies, that the male/female ratio varies?

It fails immediately because you're being asked to compare areas - when comparing data we tend to look at just one dimension - e.g. length of a bar. In this case we tend to consider the diameter more than the area, leading to underestimating the difference between colleges. Secondly, the male/female comparison fails because the scale is so small that drawing anything meaningful, even within a single college is difficult.

I would also argue that the dimension used is wrong - a more interesting dimension would be % females accepted vs. % females who applied. I'm also unsure what the class size and admitted dimensions show - admitted is presumably just males plus females. Daily Dose of Excel asks "could (should) this chart be done in Excel?"

Not one to turn down a challenge I roughly estimated the data in the chart and added some college names (obviously not the real ones). It wasn't an easy thing to chart and this isn't perfect. The raw size of the applicant pool is important to show, as is the % who actually got admitted. The ratio of males/females is interesting (though I still think it's the wrong dimension). I declined to show the class size information.

A major difficulty is the very large applicant pool of (what I call) Princeton. Maybe this would be one of those few occasions where an axis with a break in it may work. Equally, for most colleges, the admitted proportion is so much smaller than the applicant pool, that it makes comparing college to college difficult - I resorted to adding the data as text as well. Lastly, there's no way you could break the admitted bar down further to show female/male, so I pulled that data out below the main chart.

For almost any application, pie charts are not the best way to show data (that whole 2D thing again), but when you only have 2 slices, it can work - I think it works okay in this situation - you can certainly quickly scan across and see where the proportion of males/females is very different and where it's more 50/50... Anyway, suggestions on a postcard for how it could be done better.