NHL on brink of intelligence explosion

[billybudd]

Willie Mitchell, Rob Scuderi, Robyn Regehr, Matt Greene are just some of the defensive defensemen the Kings Cup winning teams have used. In fact Dean Lombardi is on record as saying he prefers pairings where astay-at-homer is paired up with a puck-mover. So the narrative that they're moving away from that is a bit wrong. Nor is speed and skill exactly the term I'd use to describe the Kings, especially not their 2012 team.

Yeah, that's what stood out to me. "4 lines of skill" doesn't apply to Los Angeles. It's revisionism to say that's how they won those cups.

It's also a bit rich that they name drop Scuderi as a relic, despite playing a pretty significant role in a championship for one of their "forward-thinking" teams. Not that Scuderi's effective at this point, but don't muddy your own narrative if you don't have to.

 

[Aceboogie]

Quickly looking at your data Brooks, it would appear that PDO correlates closer than Fenwick does. Has anyone looked at the R^2 comparing PDO to win percentage ranking vs Fenwick to win percentage ranking? I performed this analysis for GF% vs Fenwick vs Corsi and the R^2 values decreased in the order mentioned. Curious to see where PDO fits in. Perhaps the key to winning is actually get a couple of snipers and Tuukka Rask! Of course, that would also raise the question: how sustainable is PDO? Again, looking at the Avs only, neither Fenwick nor PDO appear to be very repeatable, but I'd be interested to see how teams that don't have such volatile roster turnover like COL recently fared season to season.

Of course it correlates the best, PDO is luck. It is extremely rare to have a consistently high pdo, unless you are a good team.

Lucky teams do well, but you can't control luck

 

[BB1133]

Quickly looking at your data Brooks, it would appear that PDO correlates closer than Fenwick does. Has anyone looked at the R^2 comparing PDO to win percentage ranking vs Fenwick to win percentage ranking? I performed this analysis for GF% vs Fenwick vs Corsi and the R^2 values decreased in the order mentioned. Curious to see where PDO fits in. Perhaps the key to winning is actually get a couple of snipers and Tuukka Rask! Of course, that would also raise the question: how sustainable is PDO? Again, looking at the Avs only, neither Fenwick nor PDO appear to be very repeatable, but I'd be interested to see how teams that don't have such volatile roster turnover like COL recently fared season to season.

For why PDO doesn't correlate and why it's unsustainable, here is a good article about PDO.

If you have an inflated PDO and poor possession numbers, you're usually a paper tiger...aka the Avs. The other side of that is great possession but poor PDO... a team like the Devils. They should've been in the playoffs last year with the 5th best possession ranking, but Martin Broduer played too many games and they couldn't win a shootout. Next year the Avs should regress some while the Devils will rise.

The best example for team shooting % part of PDO is the Ducks. They've sustained a top 5 team shooting % for 4 years in a row. Their problem has been inconsistent goaltending and below average possession % (28th, 21st, 20th, and 16th in those years)...so they've never been taken seriously as true contenders by the analytics community. Dellow wrote about it at the start of the playoffs .. http://www.sportsnet.ca/hockey/nhl/ducks-lean-too-heavily-on-bottom-six/ and another good article on the Ducks the year before http://www.fearthefin.com/2013/3/18/4119114/sharks-vs-ducks-is-anaheim-the-best-bad-team-in-hockey

One could argue that the Ducks higher shooting % is due to their skilled depth, but overall shooting %'s generally normalize.

 

[CDN24]

Well, of course, with Don Cherry's incoming retirement, you can expect an "intelligence" explosion!

Joking aside, I've been looking forward to better possession stats. Will be interesting.

For example, with that kind of data, we'll likely be able to improve our analysis of giveaways and takeaways (which are, as of now, subjectively measured anyways).

For example, Karlsson takes the puck in his own zone, rushes it up the ice and then gives it away in the opponent's zone. For comparison, let's say a plug 4th liner gives the puck away in neutral zone, which then results in an instant scoring chance against.

Both plays are recorded as giveaways on the stat sheet. But one is starting from his own zone, ends up the offensive zone and might create a scoring chance in the process. The other starts in the neutral zone, finishes in the defensive zone and results in a scoring chance against.

With better possession data, we'll be able to actually measure the positive impact of puck carriers and discriminate them much better from the real turnovers machines.

The subjectivity is the problem that will arise when we make advanced stats more advanced. Things like CORSI seem be a good evaluation of a team and predictor of who will fall off but did we need Corsi to realize the Leafs record was not sustainable last year when the were giving up 50 plus shots a game. Comparing 2 indivisuals is harder because they are playing on different teams for different coaches in different systems. All players on good teams tend to have better possession numbers, they are not all good but a rising tide floats all boats.

When you get into things like zone entry vs dump it gets more subjective as there are times when the dump in is the right play, once you try to measure did the guy make the right play it gets more subjective. In that line I think that in the example above Karlsson vs 4th line plug, the Karlsson play is probably more dangerous. he turns it over in offensive zone and is caught up ice, opposition gets a chance at odd man rush or at worst a forward playing D. The 4th line plug gives it away in the neutral zone with 2 D behind him ready to react. I watched a Sens game live last year where Karlsson was directly responsible for 3 goals against on plays where he was hehind the oppositions goal line. He also had 3 points ended up even for the night + - wise. Ottawa outshot the opposition (Mtl) by a lot and lost 5-4 in OT. Karlsson probably had great posession stats that night but he was the reason Ottawa lost.

 

[Czech Your Math]

Lastly, why is it that the only way for the supposed regression to take place is for the team's point total ranking to drop to its Corsi ranking? Why can't the team's Corsi ranking regress back up to match the team's point total? For the Avs this season, I believe it will be somewhere in the middle. After a year of development, I can hardly see this young core getting significantly worse.

I think it's more likely a team "regresses" to the quality suggested by GF/GA ratio or Corsi or w/e, than for a team to suddenly "learn" to increase their GF/GA ratio or Corsi or w/e to the level suggested by their previous outperformance.

Now it's possible that the Avs may not be dependent on such a strong Corsi figure for success, but we'll just have to wait until next season and see what happens.

 

[CarpeNoctem]

The problem with using advanced stats in hockey is that they will likely often ignore common hockey sense. For instance, a player in on what appears to be a good scoring chance may take a shot from a bad angle because his team desperately needs a line change.

Ultimately, hockey may have a lot of quantifiable data but will it be measured accurately in terms of quality? Who's to say what a good shot decision is without considering the player? A good slapper is a nice way to score but not when you're only 5 feet away!

 

[smackdaddy]

The problem with using advanced stats in hockey is that they will likely often ignore common hockey sense. For instance, a player in on what appears to be a good scoring chance may take a shot from a bad angle because his team desperately needs a line change.

Ultimately, hockey may have a lot of quantifiable data but will it be measured accurately in terms of quality? Who's to say what a good shot decision is without considering the player? A good slapper is a nice way to score but not when you're only 5 feet away!

Because the data averages out the percentages making up for different types of player strengths. Doesn't matter if it's a slapper 5 feet away or 10 feet, the percentages will tell a coach where to put players for best chance of scoring so the players can make those shot decisions themselves. After all, they're the professional on-ice decision makers.

 

[Viktri]

You could just as easily look at goals against. In those same years the champs finished:

1, 17, 6, 2, 2, 1, 1

Even better, why not look at the number of playoff game wins? In those same years, the champs finished 1st in playoff game wins every year. 100% predictability!

Goals and possession measure different things - there's correlation but the point is that you're trying to measure something more micro (ie: shots) rather than something more macro (goal, which can be caused by shots).