PDO Regression to the Mean, or why you should ignore Shooting Percentage
by Hawerchuk on Dec 20, 2011 7:01 AM EST
in Statistics
Email
Print
I'm still not sure everyone has completely internalized the lessons of PDO (shooting percentage + save percentage), so I wanted to look at the relationship between past and future PDO. This plot shows how much future PDO regresses to the mean relative to PDO though a certain number of shots (both for and against.) For example, PDO through the first 1000 shots regresses 87% to the mean over the rest of the season - a team with a 1030 PDO through 1000 shots expects a PDO of 1004 for the rest of the season.
Basically, whatever you think you know about your team's supposed ability to maintain high shooting and save percentages, they are very likely to crash back to league average regardless of how many shots you've observed. Internalize this chart and you can make a lot of money betting against people who are convinced there's mysticism in scoring goals.
17 comments
|
1 recs |
Do you like this story?
Comments
What about on the flip side?
Yes, you’d expect low shooting percentages to bounce up over time. But I’m curious about defense, since I follow the Bolts and they had a 991-ish PDO in 2010 (best I can tell) and have a 983 PDO now, almost all to do with a poor save rate. I’d expect that 983 PDO to bounce up, but my assumption is that they’ll regress toward a PDO lower than 1000 due to their lack of depth in the net and on defense. Or is that still bad logic?
I love Casey Fossum. Now try and take me seriously.
Follow @steveslow
by Steve Slowinski on Dec 20, 2011 8:00 AM EST reply actions
I think this looks at both regression downward and regression upward.
Red Line Station and @RedArmyLine, featuring coverage of the most frustrating team in the NHL
I believe in next year.
by red army line on Dec 20, 2011 8:41 AM EST up reply actions
The basic idea is that the true talent of shooting percentage and save percentage is so minimal that every team can expect its PDO, over a sufficient sample size, to regress (upward or downward) to 1000. Teams with abnormally high SV%‘s and SH%’s are highly unlikely to have more talent in those areas than other teams. Instead, the true talent lies in shot quantity generation/suppression, as it is at least somewhat stable over time.
by ranford4life on Dec 20, 2011 8:55 AM EST via mobile up reply actions
Yeah, I know...I get the theory.
I guess what I’m asking is a bit more nuanced.
For lack of a better comparison, consider BABIP (batting avg. on balls in play) in baseball. Over large samples, pitchers will almost always see their BABIPs regress to around league average: .300. That’s just how it is…there’s little skill involved, and BABIP is mostly a result of luck and random fluctuation.
But on either extreme, some pitchers regress toward BABIPs that are slightly different than league average. For example, reallyreally dominant pitchers have been shown to suppress BABIP to a slight degree (think .290 instead of .300), and Triple-A pitchers that don’t belong in the majors get blown up (think .310-.320 BABIP).
I’m a baseball stats guy first, so apologies for the imperfect comparison. I guess what I’m asking is, is it possible that on either end of the extremes, the regression isn’t as strong? This probably shows up more with players than teams, but it just seems inaccurate to suggest that the Bolts will regress all the way up to a 1000 PDO (or close to it) by the end of the season. They had a below-average defense last season, are playing with a backup goalie as their starter, and they’re without their best defender for possibly the entire season.
I love Casey Fossum. Now try and take me seriously.
Follow @steveslow
by Steve Slowinski on Dec 20, 2011 9:16 AM EST up reply actions
Correction, you mean “really old backup goalie as their starter” – but I understood what you meant.
There are absolutely extremes in hockey too. Gabe has shown Ilya Kovalchuk’s shooting talent as one of the few that actually stand out from the norm. On the team level, however, we rarely see it. Mostly because teams will also address problems like poor goaltending by trading for an average goaltender.
To be fair to Rollie, his current save percentage is over 20 points lower than his career average. He’s not a top ‘tender, but he’s not as bad as he’s played this year. Then again, the dude is fourty-frickin-two years old and may have finally hit the age cliff.
Ryan Popilchak
Matchsticks & Gasoline, Artic Ice Hockey, &Hockey Prospectus. My twitter handle is @sprtopinionated
Yeah, I was actually referring to Garon, since Roli has been riding the bench recently.
But agreed…I’m expecting Roloson to regress and improve over the season, but I guess I see his “true talent” Sv% as being lower than his career average right now. He’s definitely getting old, and it seems likely that if he’s slumping this bad, he’s become worse than he once was.
The trading point is a good one…yeah, you’d expect teams to regress/improve in an area because they address the need over the course of the season. But I think that also complicates matters, because then you’re implying that the team wouldn’t regress back to the mean as much simply through regression alone.
So if a team has a 980 PDO, how much regression should you expect if they don’t do anything to improve the team? All the way to 1000, or do you assume there’s a reason why their PDO is so low to begin with? I imagine the answer to these questions depends a) how far along into the season we are, and b) the specifics of the team/situation. But I think it’s interesting to consider.
I love Casey Fossum. Now try and take me seriously.
Follow @steveslow
by Steve Slowinski on Dec 20, 2011 9:51 AM EST up reply actions
I don’t think anyone’s really talking about regression if it’s only a small difference.
If Vancouver or Boston have a PDO around 1005 or even 1010, yes, they might be a bit lucky but an above average goaltender should give you above average goaltending. By a very small margin, but still. OTOH, if the Bolts ended the season with a PDO of 991, I don’t think anyone would be surprised.
by Simon Lamarche on Dec 20, 2011 10:24 AM EST up reply actions
Okay, that's about what I expected, but I wanted to make sure I wasn't assuming things that the research wasn't showing.
I love Casey Fossum. Now try and take me seriously.
Follow @steveslow
by Steve Slowinski on Dec 20, 2011 11:07 AM EST up reply actions
Steve,
I don’t think we have a big enough sample (6 seasons x 30 years) to draw conclusions about regressing to a different mean. And teams will act aggressively to change their mean on the goaltending side. Certainly for short periods of time teams will have different expected outcomes.
Cool, thanks for the clarification.
I kinda figured that it’d be a tough thing to show in the data, since it’s confounded by all sorts of things and we only have so much data.
But I think eventually, that’s the sort of thing we’d love to be able to do, right? Essentially have an xPDO, regressing each team individually based on evaluations of their goaltending/defense…taking the smaller paint brush and fine-tuning the larger scale picture of regression.
I love Casey Fossum. Now try and take me seriously.
Follow @steveslow
by Steve Slowinski on Dec 20, 2011 12:47 PM EST up reply actions
Mostly because teams will also address problems like poor goaltending by trading for an average goaltender.
As a fan of the Oilers I am shocked to hear that teams can make trades during the season to address weaknesses.
Everyone knows rock attained perfection in 1974. It's a scientific fact.
Writer for The Copper & Blue and a frequenter of the time waster that is Twitter.
by ryanbatty on Dec 20, 2011 1:03 PM EST up reply actions 1 recs
Pay no mind to what’s said here. We both know mid-season trades to add quality to the lineup are unpossible!
Twitter|Cult of Hockey|OilersNation|LeafsNation|Hockey Prospectus|ESPN Insider
Mail: jonathan (dot) willis (at) live (dot) ca
by Jonathan Willis on Dec 20, 2011 1:33 PM EST up reply actions
Boston is especially amusing actually because between 2007-2008 and 2010-2011, their shooting percentage has swung wildly every year — bottom 5, top 5, bottom 5, top 5.
But right now, they’re not only the runaway PDO leader for the second year running, their PDO has actually gone up from their league-leading number of last year.
Sorry for the late comment, but how did you run this for the larger values in the graph? Did you use multiple seasons?
Driving Play - The Blog with Three First Lines
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
