Thread by @JFreshHockey

My thoughts on PDO (since it's come up):

- Flawed because not all teams have equally talented finisher/goaltenders
- Useful because luck is by far a larger driver than talent
- Useful because it's very hard to just "guess" goaltending and finishing talent for a team...

- Useful because while goaltending and finishing do have a talent component (obviously) they are also extremely variable, especially from season to season.
- Useful because it has been shown that season-to-season and mid-season to mid-season it does regress towards 100

We know that skater finishing and goaltending isn't that far from random from season to season. If a model factoring in xG can barely pick out player-level talent, how should we expect the layman trying to figure out if his team's #s are sustainable to pick it out at a team level

These graphs compare year 1 PDO to year 2 PDO (and components). If team shooting or goaltending "talent" played a primary or even major role in deciding PDO, that red trend line would be a lot steeper than it is.

Using a very simple linear regression based on this data I put together a rudimentary estimation of year 2 PDO components based on year 1 results. The variance is evidence that there is a talent component but it is not very pronounced.

Using the same method just from the 41 game mark to the 82 game mark of a season you still see significant regression towards 100 overall.

Not every team can be expected to regress to 100. But they can be expected to regress /towards/ 100 and not all of us are gifted with the innate ability to parse out finishing and goaltending "talent" at a team level. So on-ice percentages will continue to be useful.