These are pictures similar to the ones found in the paper, except they are
actual distance vs weighted distance, instead of
actual distance vs predicted distance.
Predicted distance was a rescaled version of weighted distance, based on the linear relationship between weighted distance and actual distance. Here, we give just raw weighted distance along the x-axis.
Let's start with the NFL.
The slope of this line is nearly 1, which makes sense since in the NFL teams go back and forth for every road game. In the other leagues, they do not, so we should not expect a slope of 1. Here's the NHL:
Still very linear, but definitely not a slope of 1. Here's MLB:
Again, linear, but the slope is less than 1. Here's the NBA:
Definitely the worst of the bunch. Something seems to be going on with the two groups of teams in the ovals. The small oval is all the West Coast teams. This method underestimates travel for those teams by a lot. This is potentially because the schedulers work extra hard to make efficient schedules for those teams, giving them multiple-game road trips when they have to travel long distances. But without knowing more information, that's just a guess.
I've disable comments so that the discussion stays at Tango's blog.