Properties5
| Type | Article |
| Author | Nick Desbarats |
| Date | Nov 4, 2021 |
| URL | https://nightingaledvs.com/ive-stopped-using-box-plots-should-you/ |
| Note created | Nov 4, 2021 |
If you’re a human with normal vision, you intuitively perceive thick, wide shapes as representing greater quantities (of whatever) than thin, narrow shapes. The ‘box’ segments in each box-and-whisker shape, then, look like they represent more of something than the ‘whisker’ segments, leading to the faulty interpretation that they contain more values, or perhaps have more importance. All four box and whisker segments contain the same number of values
In a box plot, however, longer box or whisker segments don’t represent greater quantities. The four segments in a box plot each represent the same quantity, i.e., they each contain the same number of values, regardless of how long or short they are. In fact, shorter segments in box plots actually represent higher densities of values
“But I don’t find box plots hard to understand.” I hear you. I don’t find them hard to understand, either. That’s because you and I have been looking at box plots for years and our brains have learned to “think around” their design flaws. It doesn’t mean that they’re a well-designed chart type.
Have a look at the box plot on the left, then compare it to the jittered strip plot of the same data on the right:
I’m not the first to point out that box plots always make distributions look ‘bell shaped,’ i.e., like the values are clustered around the median and gradually trail off away from the median. If a set of values isn’t bell shaped (like the “Control group” above), though, box plots make them look bell shaped anyway.
- Tags:
favorite
The box plot alternative that I use most often is a strip plot:
Strip plots can be grasped by most audiences in a few seconds and can be explained with a single sentence, such as, “Each dot is the age of a study participant.”
you could move to what I call a distribution heatmap, which can handle any number of values:
Yes, a distribution heatmap ups the complexity a bit since it introduces the concept of bins (a.k.a. intervals), but bins are still considerably easier to grasp than quartiles. You also lose the ability to see how many values are in each group and there are a few other limitations, but box plots have all those same limitations (plus several others).