Thursday, April 17th, 2014

The Leafs and the Numbers

35

As most readers know, I am more than a little sceptical about the usefulness of hockey statistics, but I’m always willing to take another look at the subject. The most interesting team in the league for those who put a lot of stock in the numbers is probably the Toronto Maple Leafs. The data guys were all pessimistic about the Leafs before the season and they continue to be pessimistic about them now despite their good start.

The biggest single problem that I have with the analytics is the narrative that has emerged to support them. I don’t think that narrative describes hockey. It goes something like this: “Attempted shots – shots, missed shots and opponent’s blocked shots – measure puck possession and the key to winning is puck possession.”

Toronto’s “puck possession” numbers are awful – they are outshot too often and by too much. The assumption is that they have been blessed with an inordinate share of puck luck. Therefore they aren’t likely to keep winning. The argument is not entirely without merit. There is a correlation between shot differential and winning percentage (although it is not particularly powerful and it is not clear how meaningful it is.) Over the course of a season good teams tend to outshoot their opponents and bad ones tend to get outshot.

That said, I don’t think hockey is a puck possession game and in any case I don’t think attempted shots measure puck possession. Teams have essentially the same number of possessions per game, and I doubt if there is much difference in actual possession time. Losers might actually have the puck more!

(I think the key is speed and puck movement. The team that moves the puck the most quickly is the team that gets the most and best chances. There are two separate factors at work in a hockey game. First, there is the team’s ability to control the play – making sure the puck moves slowly toward their own net and moves quickly towards the opponent’s net. That’s about denying open ice to the opponents and creating open ice for the good guys.

A Taylor Hall quality player – I think he is probably the best player in the West – obviously helps a team accomplish those things, but he is still only one skater out of five. Where Hall really makes a difference is when the Oilers do manage to create some ice. That is the second factor. Does the team have finishing ability? When the team gives Hall the puck and open ice, he’s usually going to do something good with it and create a scoring chance. But when the Oilers as a group do not slow down the opponents or find a way to move the puck quickly, their stars become non-factors.)

While “puck possession” numbers do correlate with winning at a league level, there is no apparent link at the game level. Good teams tend to outshoot their opponents, but teams – good and bad – actually do better in games when they are outshot.

The Leafs? In the past three years plus 11 games, their record when they outshoot opponents is 31-44-10. When they are outshot, they are 74-50-15. I understand the qualms data analysts have with the Leafs, but “They get outshot too often” isn’t a particularly compelling argument when you look at numbers like that. When they outshoot opponents the Leafs play like a 69 point a year team. When they are outshot, they are a playoff team with 95 points.

To my eye the Leafs are pretty good without the puck, they do pretty well generating speed through the neutral zone and they certainly have finishing ability. I expect their team shooting percentage to fall somewhat as the season goes on and I don’t think they are a team good enough to get 105 points – their current pace – but I do think they will fight for a playoff spot all year long.

Even if their puck possession numbers suck.

Be Sociable, Share!

Comments

35 Responses to “The Leafs and the Numbers”
  1. James Mirtle says:

    You think they’re not a 105-point team; the data says the same. The question isn’t whether the data means nothing – it’s how much can it tell you.

    Shot differentials actually correlate a lot higher with winning than many think – factor out score effects and some of these metrics can be rather predictive. But they’re just indicators, not absolutes. I think where people run into trouble is when they equate second last in possession with second worst team in the league. It’s just one element of their game, and they have strong other ones that make them, as you say, in the playoff mix.

    But they don’t mean nothing. They add something to the conversation. And the accuracy of some of the data guys’ predictions in 2011-12 in particular was pretty remarkable. They’re outforecasting much of the MSM, if there was a scorecard on such things.

    • Tom says:

      I don’t disagree with much of your response, but most of it is also beside my key point. I don’t really care about predictions. I want to understand the logic and I expect a little consistency.

      If shots really do matter, why doesn’t it show up at the game level? Why don’t teams that outshoot their opponents win more games? Can anyone explain the splits for the Leafs?

      • James Mirtle says:

        It does show up at the game level, but special teams and score effects also alter the outcomes there obviously. If a team has eight power plays to three penalty kills, they will outshoot. If a team goes up 4-0 in the first period, even if they’re way ahead on the shot clock at that point, they will get outshot the rest of the way.

        If you look just at even strength play, and more specifically in score close situations, and bring in more data (i.e. the attempts), we’re talking about a very high correlation, especially relative to other stats available. This also has been shown to be a much more repeatable skill within a season than others statistically measured.

        But if you dumb this all down to its most basic level, with just all situation shot differentials (which is NOT what those making the arguments against the Leafs do, by the way), the answer is teams do win more games when they outshoot their opponents. The record for teams that have out shot this season alone is 88-58-21, or a 97-point pace. Teams that have been outshot in games are on a pace of just under 85 points. (The results are almost identical last season: 96/86.)

        But I’m guessing your question is about this one specific, anomalous team, over the last 60 regular season games. The answer on a basic level is Leafs shoot well and have gotten very solid goaltending. The trouble with teams that have won that way in the past (get outshot, score on 11+% of their shots and save 92+% of them) is that’s been both a hard way to consistently win and not the portrait of teams that have succeeded in the NHL playoffs.

        Where’s the inconsistency here?

        I think a huge part of the problem with some of these analytics is some of the message carriers are a bit overzealous and the message gets incredibly warped and dumbed down. If you read people like Tulsky, Desjardins, etc. the logic is very easy to pull out and the trends are hard to miss in the data. This isn’t just “getting outshot = bad” therefore “Leafs = bad,” although that is an element of their even strength play that likely has to improve to become that 100+ point team we’re not sure they are and that they should be aiming to be. There are also more reasons than simple possession stats why analytics folks think Toronto may not be able to keep this up: PDO, special teams performance, injuries, etc.

        If you throw all of that out and bring it to just shot differentials at the game level then, sure, you can say you’ve found a flaw in the argument. But you wouldn’t be addressing the argument.

        Me, I do care about predictions. Predictions are a clear sign the data is actually indicating something meaningful, about both teams and individual players (although I’d argue this is true to a lesser degree). The last thing we need is another midseason article lauding the 2011-12 Minnesota Wild for being first in the NHL if there’s a very obvious flaw in how they won games to start that season.

        • Tom says:

          But if you dumb this all down to its most basic level, with just all situation shot differentials (which is NOT what those making the arguments against the Leafs do, by the way), the answer is teams do win more games when they outshoot their opponents. The record for teams that have out shot this season alone is 88-58-21, or a 97-point pace. Teams that have been outshot in games are on a pace of just under 85 points. (The results are almost identical last season: 96/86.)

          In 2011-12, outshot teams went 597-493-156 (93 points) while outshooting teams went 589-459-138 (90 points) and 20 out of the 30 teams did better when they were outshot.

          In 2010-11, outshot teams went 627-421-137 (96 points) while outshooting teams were 558-421-137 (87 points) and 23 of the 30 teams did better when they were outshot.

          In 2009-10, outshot teams went 588-420-160 (94 points) while outshooting teams went 580-461-127 (90 points) and 19 of the 30 teams did better when they were outshot.

          The rationale for using shots as a proxy for goals is that over time, it is the volume of shots that matter. Yes, there is a difference in shot quality, but those differences wash out. We assume that if a team shooting percentage is excellent it is because they had good fortune and not because they had higher quality shots or were better shooters.

          If this was true I would expect that outshooting teams to have a persistent clear advantage, and I would expect that a significant majority of teams every year would do better when they get more shots. There is no persistent advantage and most teams do better when they are outshot. Why? Am I missing something?

          Me, I do care about predictions. Predictions are a clear sign the data is actually indicating something meaningful, about both teams and individual players (although I’d argue this is true to a lesser degree).

          I think we are talking about two different types of predictions here. I am very interested in any testable prediction. Baseball analytics are very valid because if you tell me how many hits, walks, and extra base hits a team got, I can predict with some certainty how many runs they scored. If I tell you how many EV shots, PP play shots and shots taken when I am leading or trailing by two goals or more, can you tell me how many goals I will score with any degree of reliability?

          But I’m guessing your question is about this one specific, anomalous team, over the last 60 regular season games.

          It isn’t just the last 60 games though. It is three plus years of being really lousy when they outshoot opponents and really quite good when they do not. How can that be squared with the puck possession narrative? The Leafs are an extreme case, but given enough games, I would expect every single team to do better in games when they get the most shots. Why wouldn’t that be the case?

          It seems to me – correct me if I am wrong – that the consensus view of the Leafs is that last year they did it with mirrors and that this year they would regress. The quants are not just saying the Leafs will not get 105 points. They are saying they will not maintain the pace of last season. If they are right, the Leafs probably will not make the playoffs.

          So far (add caveat about small samples) the Leafs have progressed, not regressed.

          • James Mirtle says:

            I think any team with two goalies posting save percentages of more than .930 would make the playoffs. Pretty sure every analytics person would, too. But would they predict that’s what would happen for the Leafs going in? Or that they’ll shoot better than 12 per cent when no team in recent memory has done so?

            The thing is shooting and save percentages are not very reliable. If you want a metric to predict those, you’ll probably be left wanting. Possession numbers are fairly reliable, with definite in-season and season-to-season correlation.

            As for the outshot question – if you look at season shot differentials overall, they correlate much more highly with being a good team. Your system accounts for outshooting a team 30-29 the same as Chicago outshooting Toronto 40-20 when clearly there’s a different quantitative value there.

            And the possession statistics account for that. Along with score effects, situational play, etc. It’s easy to say “this doesn’t work” when you ignore the depth of the analysis actually being done and dumb it down to just outshooting. You also don’t address the issue of postseason success, which is really what all teams should be aiming for.

            As I said, this movement in analytics is about zone time and offensive possession more than simply pucks actually hitting the goaltender. The correlation between Fenwick Close and winning and outshooting records and winning is remarkably different. That’s why we don’t use outshooting records to assess teams.

          • James Mirtle says:

            So far (add caveat about small samples) the Leafs have progressed, not regressed.

            They haven’t looked very good in doing it… this team could easily be 7-5-0 or 6-5-1 right now. Does the narrative then change? Because those are much less impressive records, projection wise.

      • James Mirtle says:

        Taylor Hall, by the way, is probably the best player in the West based on these statistics, if you consider his team, linemates and usage.

        http://www.extraskater.com/players/dashboard?season=2012

    • Mark says:

      concentrating on the 2011-12 season is interesting.

      There was one glaring example of overachievment that year that I remember offhand – and that was the Panthers winning their division and putting in a solid season all around with the 14th best record in hockey.

      And that performance was fully backed up by their Fenwick Close, which had them at 13th.So according to Fenwick Close, those 11-12 Panthers were 100% legit, and not an overachiever at all.

      Yet by more traditional metrics like Goal Differential (they were in the bottom-10 at -24 that year), and individual player projections and talent analysis, it seemed pretty obvious to everyone that the Panthers were clearly a fluke, and would inevitably regress.

      And of course, they did regress, right back to the bottom of the standings the next year, and so far this year as well.

      So even in that year you mention that the possession stats were “remarkably” accurate, we can see that they were completely blind to the most glaring anomalous performance in the league that year.

      • James Mirtle says:

        There are a lot of other factors there, which people in the stats community would also look at in making forecasts.

        One, last year’s team led the league in man-games lost to injury. Florida also went from top 10 in save percentage to dead last, as Theodore and Clemmensen predictably fell back from very good seasons.

        But the biggest reason the analytics community didn’t like the Panthers chances of repeating was the fact they were in 17 shootouts and 25 extra time games overall. They went 17-5-18 in one-goal games – which no one would say was repeatable. Add in the fact they were in a brutal Southeast Division…

        Again, if you boil things down to a one number = one answer solution, you will be wrong at times. I predicted Florida would finish 14th in the East prior to last season and I look at these numbers a lot. Possession is important. It’s not everything.

        • Mark says:

          Fair enough, and if you predicted them to drop in spite of their solid Fenwick Close than that’s a feather in your cap and shows that unlike many, you seem to consider the possesion numbers with the right amount of balance.

          I do notice that none of the (forgive the term) “advanced stats guys” were willing to look signifantly outside the leafs’ fenwick close numbers from last year when projecting this year’s Leafs team, though. And that’s where the notion of bias might come in.

          In fact, it’s hard not to see the rather huge uptick in possession stats reference over the past year as a result of specifically what it says about the Leafs recent success.

          • James Mirtle says:

            No, it’s not bias. Every indicator on the Leafs had them taking some sort of step back. Very high shooting percentage, high PDO, low man games lost, high record in one-goal games, etc. In addition to one of worst possession ratings in the six years the data is available.

            Rob Vollman predicted they’d be one of the worst teams in the league. I had them about 90 points and on the bubble.

            Perhaps the rise in possession stats reference in Toronto is higher due to the Leafs, but this was always going to catch on more at some point. I made the decision to start using them more after analytics guys predicted the eighth seeded LA Kings were likely candidates to win the Stanley Cup.

            How on earth could you ignore that?

          • Mark says:

            Seems like bias to me.

            Especially when you point out SH% and PDO as separate indicators from the shot metrics themselves, when they are the same indicator. You have decided that shots are always the true talent indicator, and goals are merely a function of shots x league average sh%. That’s fine, even if arguable – but don’t double count your indicators. Because the shot totals are the only thing that says the leafs aren’t a very good team. All other team (gf, ga, pp, pk, sv%) and individual indicators (player scoring and save % track records) indicate a quality team over 60 games now.

            As for the Kings – their playoff performances the last two years have everything to do with Quick turning superhuman for the two runs, and not much to do with their shot metrics. When Quick is human, that team is mediocre. When Quick is superhuman, they’re great.

          • James Mirtle says:

            I didn’t say most of that.

  2. beingbobbyorr says:

    I thought MoneyPuck was merely about trying to judge how much individual player A contributed to W-L vs. player B (and, therefore who is/isn’t worth X dollars). . . . . Now they’re trying to tackle the grand task of predicting team performance (where a team is a system, and, as Dr. Russell Ackoff told us, systems are more dependent on the interactions of their constituent parts than the properties of said parts)? Are QoT factors really that well understood to make such predictions, or are the hockey analytics people just doing regression analysis on past data and hoping no present/future team deviates too much from the least-squares-fitted lines?

    • James Mirtle says:

      All of the hockey analytics arguments I began reading years ago were team focused. This isn’t new, and PDO is probably one of the oldest stats out there. Other than the goaltender, one individual player in hockey doesn’t have a tremendous impact on the overall data.

      That’s not to say the individual analytics aren’t interesting – they’re great, especially for putting players’ roles in context.

      I’m not sure if you’re talking about in-season or season-to-season regressions though. The in-season ones are much more compelling, as some teams can become huge outliers 20-30 games into a year and most come back down, sometimes dramatically.

      • Tom says:

        I’m not sure if you’re talking about in-season or season-to-season regressions though. The in-season ones are much more compelling, as some teams can become huge outliers 20-30 games into a year and most come back down, sometimes dramatically.

        One of the other things that niggles at me is when we consider season to season regressions, is the change in the roster. Usually it is only a few players in and a few others out, but that understates the effect. Injuries have a big effect too. If a team loses 100 man games to injury one year, and 300 the next, the team actually iced may be quite different.

        • James Mirtle says:

          For sure. Those changes have to be accounted for and that’s usually why the numbers are better for in-season projections.

          The Leafs were also remarkably healthy last season, although aspect analytics folks pointed to as something that may not continue.

      • beingbobbyorr says:

        I’m not sure if you’re talking about in-season or season-to-season regressions though.

        I was thinking season-to-season regressions (over MANY past seasons), just because, (a) as you imply, the in-season numbers don’t become meaningful until the sample sizes get big enough (and by then, the season’s almost over), and (b) I thought MoneyPuck was trying to look for universal truths, not contemporary trends/outliers.

        • James Mirtle says:

          Actually the in-season ones are more useful early in the season as opposed to late.

          Moneypuck doesn’t purport to look for any one certain thing but answers in data. They’re not always absolute answers.

      • Mark says:

        I think you’re off here.

        It seems to me that the only reason the individual analytics aren’t used is because they are unreliable.

        Theoretically, though, the individual analytics are much more important than the team analytics, and the fact that we have such a hard time discerning the individual possession “true talent” levels really is an indictment of the possession stats overall, and not an affirmation that team stats are sufficient.

        I’m also interested in seeing if even in those in-season outliers that Corsi/fenwick is able to spot might be just as sufficiently explained by other methods – either team stats like simply goal differential being out of whack with win/loss record OR any individual significant short term individual over or underachievement Ii.e. key players performing well outside of recent career norms). Are we sure that those in-season outliers spotted by Corsi/Fenwick can’t also be seen by other methods? and does Corsi/Fenwick miss in-season outliers that other methods are able to spot?

  3. Thomas says:

    One thing I never see discussed in connection with hockey analytics is the fact that numbers vary widely from building to building. Shots are probably not quite as variable as other stats, but things like hits and takeaways seem to be fairly subjective. Until data collection is standardized, I think hockey stats have to be taken with a degree of caution.

    I don’t know if it is the dumbing down James talks about, but mostly I hear stats people talk and think their insights are pretty rudimentary. I have to believe if teams were serious about using stats, they would have much more sophisticated measures than the ones I hear discussed so much.

    It reminds me of grad school in the 90s when French social theory was in vogue. So many words to say not much at all.

    • beingbobbyorr says:

      numbers vary widely from building to building

      The hockey analytics people are very aware of the building-to-building variance due to subjectivity of local eyes-in-the-sky, and either (a) discard or (b) attenuate the stats from each team’s 41 home dates. The theory being that road games will present a smearing of the building-to-building variation, and thus render comparisons closer to apples-to-apples.

      There’s a fan on the LA Kings message board who’s always pitching his grand ideas about using various technologies (RFID chips inside pucks and skate blade holders, sonar, etc.,) to track the puck and the players, thus turning subjective judgments into more objective measurements, thereby freeing the eyes-in-the-sky to focus on more subtle nuances/judgments (that technology can’t do). I’ll have to ping him to see if he’s made any headway in creating some formal document to toss at hockey people who can act on those ideas.

      mostly I hear stats people talk and think their insights are pretty rudimentary.

      I’m guessing you were trying to say that you perceive the MoneyPuck people believe their stats are insightful/revolutionary, while you think they are rudimentary/elementary? I admit to being a little bewildered myself at claims that “the GF-GA ratio correlates well to winning percentage” is considered news (duh!), but beyond that, I find their studies pretty interesting (see Tyler Dellow’s recent series about shots/shift) even though I’m not personally interested in playing armchair GM. Besides, compared to TV sportscaster clichés like “he arrives hard on the puck” and “he’s good in the dressing room”, the aforementioned “GF-GA ratio correlates well to . . . ” is downright profound.

      I have to believe if teams were serious about using stats, they would have much more sophisticated measures than the ones I hear discussed so much.

      Whatever NHL teams are doing with advanced stats, it’s strictly black-ops / behind-closed-doors stuff (for now), because we’re still in the Jurassic era and there is nothing but (a) embarrassment or (b) loss of trade secrets for anybody drawing an NHL paycheque to stand on a soapbox and proclaim that they’ve discovered the magic beans. i.e., I don’t believe for one minute the public assertions made by the current TML braintrust, or (when he was their GM) Brian Burke, that they have no interest in this stuff.

      Even as/if some magic formulae emerge through years of consensus, it’ll take still more years (decades?) until such stats make it into http://www.nhl.com/ice/statshome.htm or http://www.hockey-reference.com/ or http://www.hockeydb.com/ just due to institutional inertia (tune into any major league baseball game to see how RBIs are still a core element of batter statistics, even though it’s an atrocious metric, given that it’s NOT normalized to either the number or quality of opportunities).

    • James Mirtle says:

      The degree to which buildings miscount shots and shot attempts has been measured and it isn’t nearly as egregious as those other subjective stats. Because shots and shot attempts aren’t very subjective, probably.

  4. Tom says:

    As for the outshot question – if you look at season shot differentials overall, they correlate much more highly with being a good team. Your system accounts for outshooting a team 30-29 the same as Chicago outshooting Toronto 40-20 when clearly there’s a different quantitative value there.

    I realize this (and I am aware of the unreliability of SV% and SH%, too). Obviously good teams outshoot opponents by more and are outshot by less than poor teams. If this were not the case, we would not have the inconsistency.

    And the possession statistics account for that. Along with score effects, situational play, etc. It’s easy to say “this doesn’t work” when you ignore the depth of the analysis actually being done and dumb it down to just outshooting. You also don’t address the issue of postseason success, which is really what all teams should be aiming for.

    I’m not just saying this doesn’t work. I’m saying that if the narrative is correct, it should work. I could be misunderstanding the narrative about hockey being a puck possession game or perhaps there is a good explanation for why I cannot apply that narrative – and the data – to the performance of teams in games. Why was Toronto so good when they were outshot and so bad when they were not? Why have Vancouver and St. Louis been so much better when they get outshot? Why don’t team records say “Yes, getting more shots makes a difference.”

    I understand that shots are not as precise as Corsi or Fenwick Close, but I don’t want to digress from the main point here to discuss those measures. Shot differential is correlated with winning and that is probably good enough to say that there should be a pattern in the outshooting/outshot by data. Are you saying that the contradiction will go away if I used Fenwick Close or something instead of shots? The Leaf record will turn bad when they are outFenwicked and good when they Fenwick? If I use one of the more esoteric variations of shots, the team data will make sense?

    • James Mirtle says:

      Over a large enough sample size yes it will make a lot more sense than what you’re doing. You’re using bad data. Obviously there will be some anomalies.

      The other issue is that possession and shot differentials are not the entire game. For one, most of the time we’re just talking about even strength play. For another, these numbers ignore entirely the role of goaltending, for both teams, in a game.

      Doesn’t that answer your question right there? Some (rare) teams will win a disproportionate number of games based on their goaltenders being much, much better than the opposition’s in the games they play. But we have considerable evidence that that’s a tougher way to win consistently (and against good teams) than simply having the puck more than the other team.

      Here was a good piece along those lines today – essentially when you have both possession and goaltending going your way, you’re a dominant team. If you have just one of the elements, possession is more effective to have.

      http://puckprediction.com/2013/10/29/why-you-shouldnt-sacrifice-possession-for-shot-quality/

      • Tom says:

        Doesn’t that answer your question right there? Some (rare) teams will win a disproportionate number of games based on their goaltenders being much, much better than the opposition’s in the games they play. But we have considerable evidence that that’s a tougher way to win consistently (and against good teams) than simply having the puck more than the other team.

        It is a given that when teams are outshot and they score more goals, they will have a better save percentage in those games. I don’t think we have any evidence that a given team has the puck more than their opponent. There is no compelling logic in the idea that more attempted shots means more puck possession. I don’t even think it is an intuitive idea.

        Goals are what matters. More attempted shots (whether at even strength or not) at the global level does equate to more goals. It does so because more attempted shots should mean more shots. More shots should mean more good quality shots. More good shots should mean more goals. That’s a long chain of logic and at every step there is room for anomolies. If the team is good at blocking shots, the ratio of attempted shots to shots may be upset. Bad goaltending can blow the chain at the shot level. Good goaltending can blow the chain at the quality shot level. All of the anomolies wash out at the highest level.

        An assumption in your response is that the goaltender alone is responsible for the save percentage. This is clearly not true. It is clear that team can adopt tactics that trade off quantity of shots for quality. If this was not so, goalie save percentages would not vary according to the game score and Fenwick Close would not correlate better to wins than shots do. Apparently plenty of attempted shots are irrelevant. They do not lead to more goals.

        That in itself should raise some red flags. The entire edifice of the puck possession story requires the assumption that a shot is a shot and goals are scattered randomly among the attempted shots. Why shouldn’t strong puck possession numbers be rewarded in all circumstances? We have large identifiable chunks of time when “having the puck more than the other guy” produces no result. This shouldn’t happen if the key to winning hockey is having the puck.

        What if there is a different hockey assumption that fits the existing data better? Instead of tweaking the data, tweak the story:

        Hockey is not a game of puck possession. Teams get the same number of possessions per game and they very probably have the puck about the same length of time. Hockey is a game of puck position and the team that moves the puck the fastest will almost always get the most and the best opportunities to score. Unless the opposing goalie dramatically outplays your goaltender, you win. (I honestly can’t see how anyone who watches the games can disagree with that idea.)

        Once a team builds a lead, they are content to slow the play down because they have enough goals to win. They don’t make the risky pass because they want to avoid turnovers and because the clock is on their side. They may not generate good shots wandering up the ice instead of zipping the puck around, but they do it anyway because wandering is safer and because it takes more time. A team with the lead wants to play low event hockey. Corsi numbers swing against them because the opponents are pressing and throwing everything at the net. But scoring chances don’t follow because the game is slowed down. If the puck is not moving quickly the relationship between shots and good shots breaks down.

        I certainly think that description fits what we see when we watch hockey. If neither team can generate speedy puck movement we get a low event, low scoring chance game. If both teams are zipping the puck around we get a shootout, last goal wins. If one team moves the puck and the other does not we have a blowout. Most of the time, both teams have their moments and the game goes back and forth with “momentum swings” marking the point where the puck movement of one team slows and the other picks up. But at the end of the day, the team that moves the puck the best usually wins.

        Does the data support that vision of hockey? I think it does, although a key metric – scoring chances – would really help. We would expect that the teams that move the puck the fastest will get more shots as well as more good shots. The overall data should scan okay. The best teams tend to outshoot their opponents, but that’s not the reason they win. They win because they move the puck better and get the best chances to score.

        At the individual team game level the link between attempted shots and good scoring chances is much weaker and the shot numbers are usually close anyway. Teams frequently lose on the shot clock but win on the battle for scoring chances.

        If even strength shots when the score is tied is the best indicator of team quality, it is probably because that is when the relationship between shots and good shots and goals is strongest. That number is not a puck possession statistic though so it does not help advance the narrative. Maybe it can serve as a proxy for the team that moves the puck the fastest. I don’t know.

        But even then, I don’t think offense in hockey is not about taking lots of shots. Offense is about moving the puck. If you move the puck, you’ll get shots.

        • James Mirtle says:

          Teams get the same number of possessions per game and they very probably have the puck about the same length of time.

          Why do the Leafs score so low when people take a stopwatch and measure when they have the puck? Why does zone time correlate so highly with Corsi and Fenwick (which also correlate very highly with scoring chances, when they’ve been recorded)?

          I don’t understand why you’re so focused on the individual game level – these stats don’t purport to provide that information and frankly I don’t think that would be overly useful. Every hockey game is such a different entity that it only makes sense to evaluate a team over a larger sample size.

          The numbers are not perfect, but they provide some additional information as to the quality of a team’s play. That’s valuable, whether you want to call it possession or not.

          • Tom says:

            Why do the Leafs score so low when people take a stopwatch and measure when they have the puck? Why does zone time correlate so highly with Corsi and Fenwick (which also correlate very highly with scoring chances, when they’ve been recorded)?

            I assume the zone time question relates to the study Ferrari did on the season the NHL published zone times. It would be really interesting to see the NHL publish that data again. We were looking at it – and arguing about it – at the time it was published. The data has the same problem shots data had. When Ferrari looked at it globally zone time correlates to shots. We did not have the technology to easily do that, but we could look at the individual games. It was very surprising to find that in most games more time was spent in the winning team’s end. It was baffling. I think a careful study will show a global-individual game contradiction.

            I read the other study too, but it does not surprise me. If we have two teams alternating possessions and Team A really moves the puck and Team B has to struggle their way up the ice on every foray, which will possess the puck for more time?

            I don’t understand why you’re so focused on the individual game level – these stats don’t purport to provide that information and frankly I don’t think that would be overly useful.

            I’m trying to acquire your level of confidence in the data. It would confirm what the global data purports to show. The global data shows a correlation. Does that mean if a team gets more shots it will improve? Or does it mean if the team gets better, they will probably end up getting more shots? In other words, does the correlation equate to causation? Is it really a meaningful link?

            My point is that if getting more shots actually improved a team, it should be confirmed at the game level. When it is not, I think it implies that more shots at the global level are a result, not a cause.

            It is the difference between saying “Our offense is not good enough. We aren’t taking enough shots. Shoot more!” and “Our offense is not good enough. We have to figure out how to get better puck movement and if we do that, we will probably get more shots.”

          • James Mirtle says:

            I’m trying to acquire your level of confidence in the data.

            It took me years to get there – I followed this for a very long time before using it in my work. I’m now convinced it has some value even if it’s not quite to the extent some will argue it does. It all requires a lot of context.

            But being a good possession team, as defined by the stats we use now, is an important characteristic of being a good team. If you can consistently be in the position where your players feel like they have offensive opportunities significantly more than the other team, you’re doing something right.

  5. rickithebear says:

    Tom:
    1. # of 1st phase attacks; # of 2nd phase attacks (rebound, offensive deflections); etc
    2. # of zone entries to attack phases.
    3. # pucks directed at net per phase
    4. location of release
    5. type of release
    6. path of release (deflected, clear, obstructed)
    7. # that reach net. (blocked, deflected away)
    8. position of goalie.
    9. location of shot by elevation. 1-7 holes; pads; chest.
    10. goalies weakness
    11. result goal/no goal.
    This is procession analysis. they are all critical to the procession.

    there are 7 elements to a pocession before a shot can be registered.

    Corsi is a measure of volume. but precludes 1,2,4,5,6. all part of point of release
    7 is influemce of path. (gets there not)
    9, 10 are the goalie.

  6. Tom says:

    Perhaps the rise in possession stats reference in Toronto is higher due to the Leafs, but this was always going to catch on more at some point. I made the decision to start using them more after analytics guys predicted the eighth seeded LA Kings were likely candidates to win the Stanley Cup.

    How on earth could you ignore that?

    Believe it or not, I’m glad you made the decision even though I’m very sceptical every time I see a column based on numbers. The alternative is worse. The fact is that we don’t know very much about this game. We know lots of it is luck but nobody wants to talk about luck. Very often your job is to explain why one team or the other was favoured by randomness.

    I’d rather hear a case based on numbers even though I think the numbers are mostly bullshit than hear an explanation that is based notions like grit, chemistry or leadership. Those are arguments that I think are complete bullshit.

    That said, I don’t find this prediction all that amazing. Partly because I was pretty familiar with the team and I knew they were good enough to win in a year that it wasn’t hard to imagine any of the eight western playoff teams winning. More importantly, I don’t think they really predicted anything.

    It is always about probabilities. You might have given the Kings little chance to win If pressed y ou might say they had a 5% chance. If the quants made them their biggest favourite most probably didn’t – they still aren’t giving them a better than 20 or 25% chance. How do we credit them for making a guess and having it come through when even they knew they would be wrong 80% of the time? Predictions are a mug’s game. Most of the time you are wrong, and even when you are right, you got very lucky.

    • James Mirtle says:

      Because their analysis relies on said probabilities, not nothingness and BS. Hence their analysis is better, their forecasts more accurate and my time better spent reading said information.

Speak Your Mind

Tell us what you're thinking...
and oh, if you want a pic to show with your comment, go get a gravatar!