Data Viz Whining - Scrabble Edition
It's hard to express how much I hate this plot from this otherwise enjoyable video.
There's just so much junk on the chart:
- The trends mean nothing unless you want to imply that before the change things were already trending towards upwards towards what would become "the post-QI normal". Why would it be upward initially, then downward/flat after? Did Scrabble stop improving suddenly? Or maybe...
- The sample is fucking tiny. You have thousands of matches over decades well recorded. Why are you only looking at weekly averages for a single year?
- Using the simple uncertainty calculation built into ggplot really oversells the confidence of the trend - 20+ point variations week-on-week make a <3pt uncertainty academic at best.
- There are clear high-leverage outliers in the data, which obscure the real story when you focus on means and trends - adding QI to the dictionary made it less likely for players to get shafted by bad luck - it made it a more consistent game, rather than a simply higher scoring one. Bulk averages are far less interesting than the distribution - which the plot could be highlighting with the exact same data.
Par for the course with 538, I know, but still annoying to see it referenced elsewhere.