Quote:
Originally Posted by SeeGeeWhy
I'd like to see a plot of ... the last 10 completed seasons, with ALL teams on the plot, with instantaneous PDO on the Y-axis, and Games Played on the X axis. Does PDO REALLY regress to 1000 for all teams? And does that hold true across multiple seasons? I'd like to think that a team that drafts well from a limited pool, hires good coaching will tend to accumulate pieces in their system that have a higher degree of performance and less variability to their performance... which is what this bullspit "stat" is all about.
|
Yeah, that would be an interesting data set. But before you call it a bullspit stat, consider whether it's the stat itself, or the conclusions that are being drawn from it. It's a weird stat because the assumption is that past performance IS NOT an indicator of future performance, and thus that things will tend to average out. But if someone runs the numbers as you suggest and finds that there is a correlation between past performance and future performance in this regard (which I think there would be), then it's still a useful stat, because it tells us that shooting percentage is the result of systems and players, and not the unpredictable element that has been suggested. The big failing of PDO right now is that it has zero predictive power, because most of us doubt that underlying assumption about regression. But if you understand the patterns of regression, you could then use it to project how likely it is that a particular team's goaltending or shooting percentage improves or declines throughout the season. And that would make the stat a heck of a lot more useful than it is now.
I'd expect you'd find that there's solid correlation between past and future results in save %, and a somewhat less solid but still noticeable correlation in shooting %. That's just a guess from a cursory look at past seasons.