clock menu more-arrow no yes

Filed under:

Shooting Percentage Estimation Errors

New, comments
I recently found location data for missed shots going back four years, so I decided to compare shooting percentage estimates with and without missed shots. When missed shots are included, the high shooting percentage in the slot decreases substantially in size:
We can see where the estimate error is largest by subtracting the dataset that includes misses from the one that doesn't:

So by failing to include missed shots, we've substantially overestimated the shooting percentage in the high slot, 20 to 30 feet out from the goal. We can also express the error as a percentage:

This gives a different result - shots from the blue line are much less likely to go in than we would otherwise think. If we had location data for shots that get blocked (we only know where they were blocked, not where they were shot from), we'd probably find that we're making an even larger overestimate.