Friday, November 21st, 2014

The Numbers and Pat Quinn

12

For the most part I haven’t argued very much with those who like to look at hockey through a statistical lens even though I don’t buy much of it insofar as it applies to individual players. I don’t think it hurts anyone to play around with the numbers. Some of the best hockey bloggers are in the club. One can enjoy reading Tyler Dellow or Matt Fenwick or James Mirtle while disagreeing with their approach to player evaluation.

But I can’t let this Andy Grabia post slip by unchallenged because he implies he knows more about what contributes to winning hockey than I do because, after all, I pay very little attention to the things he considers important. The post also implies that I disagree with him because I believe in hockey fairy tales.

For what it is worth, I believe (and ironically, I learned it from Pat Quinn) that scoring chances are what really matters. I’m usually happy with my team’s performance if, at even strength, we get the most the scoring chances. All of Pat Quinn’s decisions are designed to help create more scoring chances or to help prevent opposing chances.

Andy’s ire was triggered by Pat Quinn’s unorthodox decision to place JF Jacques – a physical non-scoring winger – on a line with Horcoff and Hemsky. Criticising that decision is fair ball, but I wish Andy made an actual hockey argument to support the criticism. After serving up with a quote from a baseball book (sigh), Andy writes:

What I care about is the broader point of Posnanski’s post, and Epstein’s interview, which is this: far too many professional sports teams, sport reporters, and sports fans focus on aspects of their particular game that are either secondary or irrelevant to the game’s primary purpose, which is winning. They elevate things that, while perhaps fun and entertaining, haven’t really been proven to exist or matter. Coaches who talk about players needing more “crust” in their game, for example. Fans who brag about a player being the team’s best body-checker, for another. Reporters who mythologize a player for being “clutch” in “crunch time.” Me, I’d rather people–managers, coaches, players, reporters, fans–worried more about figuring out what things actually determine success in a sport. Do I know exactly what those things are? Hell, no. But I do have some guesses.

I don’t believe any of these myths, but they will persist no matter what. We can demonstrate statistically that the first goal of a game is no more important than any other, but that’s not going to stop people from believing it is particularly important to get a lead. As long as there are clutch goals – and there are – people will conflate that to clutch ability. Reporters will mythologise and teams will happily sell myths no matter how often they are debunked.

I’m that way because I’m guessing (I’ve been convinced, actually) that some of these [new measurements] are more important in determining the outcome of a game than the things we’ve traditionally looked at and used to determine outcomes (and excellence).

None of the new measures embraced by Andy determine the outcome of a game. At best, they measure a few of the things that happened during games in the past. The outcome of any game is actually determined by how fast the players skate and how well they pass, shoot and check. (That’s why I like the players who skate fast, pass well, shoot hard and check diligently.)

You want to win? Have guys on your team who do these things well. Forget about how “big” his heart is, those “huge” goals he scores, and whether or not he can impale an opponent on his stick like a trident.

Is there a single coach in the league who actually makes personnel decisions based on the fairy tales that surround the game? They will choose a goal scorer over a big hearted clutch goon with leadership skills every time. They always have and they always will. They do not need scoring rates or Corsi numbers to tell them to do so.

When I see a near historically bad NHL player getting first-line minutes because he’s big and hits people, a little bit of my sanity dies. Why? Because the bottom-line is the guy isn’t helping his team win when he is placed in that position. His skill set does not match up with what is required to win hockey games. And the coach isn’t helping his team win, either. In fact, he’s hamstringing their chances of victory, based on some antiquated and unproven model of success. And there goes my sanity, right out the window!

If Jacques busts for the net every time his line gets the puck in the offensive end – a job that fits his skill set to a tee – he can create scoring chances for other Oilers without touching the puck. He can turn a routine point shot into a scoring chance. That helps his team win. I may not be able to demonstrate it statistically but it is very difficult to score off an offensive zone cycle without getting a player to the front of the net. That is a hockey truism that is not a myth.

Ideally the player in front has enough skill to cash a few of those chances – a deflection, a rebound off the screened shot – but if you don’t have a player with the size and skill required for the role, size alone is better than skill alone. It is better to get a plugger in front setting up a screen and drawing a crowd than nobody. Quinn doesn’t have a Ryan Smyth and he doesn’t have many forwards big enough and strong enough for that role, period. He’s picking Jacques from among the options. That seems reasonable to me.

This isn’t the first time Quinn has had this problem. One year he frequently played Gino Odjick with Bure. It worked because Gino went to the front of the net and Pavel did everything else. Gino scored 16, Bure scored 60 and the team went on to the Stanley Cup Final.

Pat Quinn has his own model for success but it is hardly antiquated or unproven. His model has been remarkably consistent and remarkably successful. He’s improved every team that hired him and I’ll be very surprised if he doesn’t improve the Oilers.

Be Sociable, Share!

Comments

12 Responses to “The Numbers and Pat Quinn”
  1. Magicpie says:

    I’m that way because I’m guessing (I’ve been convinced, actually) that some of these [new measurements] are more important in determining the outcome of a game than the things we’ve traditionally looked at and used to determine outcomes (and excellence).

    None of the new measures embraced by Andy determine the outcome of a game. At best, they measure a few of the things that happened during games in the past. The outcome of any game is actually determined by how fast the players skate and how well they pass, shoot and check. (That’s why I like the players who skate fast, pass well, shoot hard and check diligently.)

    Yeah, but there’s two problems with just watching games and evaluating players’ skills as a method of player evaluation.

    1) You can’t really evaluate players in any quantifiable way (even in terms of individual skills) just by watching games. For example, how do you evaluate a skill like “shooting”? Is it accuracy? speed? quickness of the release? Is it something more subtle like knowing the right spot to aim for? Are all these sub-skills equally valuable or is one more important–can knowing the goalie’s weaknesses and aiming for the right spot make up for having a weak, inaccurate shot with a slow release? (You may argue that you don’t really need to go through each of these individual sub-skills to get an idea of how good a shooter someone is, but I don’t see how you’d be able to. If someone has a fast shot with a quick release can you say for sure they’re a good shooter? What if they can’t hit the side of a barn?)

    Are you going to be able to evaluate all of these individual sub-skills, and how they fit together, to get an idea of, say, how good a shooter each player on the Canucks is, even if you watch every game this season? And are you going to be able to evaluate the X other skills a player needs to have, each with their own sub-skills, at the same time in any quantifiable way? I think that’s asking too much. The fact is there’s simply too much stuff going on in a hockey game, and the way all of these skills interact is too complicated to allow you to evaluate a player’s skills in quantifiable way just by watching games.

    2) You can’t quantify how the discrete skills you mention above work together to help a player be successful. Are all these skills equally important, or do some matter more than others? If you have to choose between a guy who’s a great skater but has just an OK shot, or a guy who’s a great shooter but is just an OK skater, all else being equal, which one will score you more goals? (and how many more?) You can’t really answer this question just by watching games and you need to answer it to be able to fully evaluate the effect players have on your team.

    I’m not saying that the methods used by the stats guys solve either of these problems. They don’t. But at least they try to study things in a simple, quantifiable way that gets around these problems. For instance, going back to the example of evaluating a player’s shooting, instead of studying video each player in the league and evaluating him on a 10 point scale on the 12 different sub-skills that make a good shooter, a stats-based approach starts with the ultimate object of shooting–scoring goals–and work backwards. It judges players’ shooting simply by how many goals they score on how many shots (i.e. shooting percentage). It’s not a perfect method, but it does provide some objective, quantifiable way of judging given player’s shooting skill, without having to worry about whether the player is successful through picking the right spot to shoot, having a really accurate shot, or some combination of the two, and without having to understand how all the different sub-skills involved in shooting interact to affect the ultimate effectiveness of a player’s shooting.

    The same principle essentially underlies most of the new stats mentioned by Andy. Instead of looking at individual skills and working forwards to try to get an idea of a player’s value, they start with the ultimate purposes of hockey, scoring and preventing goals, and work backwards. Scoring rates and adjusted plus minus and most of the other new stats are all basically ways of looking at how many extra goals a team scores/allows because of an individual player, without having to figure out whether a player is effective because he’s a fast skater, or because he’s a good hitter, or he has a good shot, (or quantify these skills in any way) which is good, because as I mentioned, that shit is frigging complicated and I don’t know of anyone who’s come even close to figuring it out. Even better, these methods results in direct quantifiable figures that allow you to put a specific number of players’ contributions.

    These new stats aren’t perfect, by any stretch,and I don’t think a lot of stats guys would argue that they are. Nowhere in his post does Andy say that the methods he endorses are unquestionably the best way to look at things. In fact the last paragraph makes the opposite point. But at least they accept the fact that you can’t study this stuff just based on your visual observations about players, and make some attempt to study things systematically in a scientific manner, and put actual figures and testable hypotheses out there. And that seems a lot less arrogant than saying that you know everything you need to know about hockey because you know that fast players are better than slow players.

  2. Tom says:

    The fact is there’s simply too much stuff going on in a hockey game, and the way all of these skills interact is too complicated to allow you to evaluate a player’s skills in quantifiable way just by watching games.

    Why is it important to quantify it? I agree that I cannot do that. If I could it would be great, but I can’t. Besides, I’m only interested in knowing whether the guy has a shot that can generate (or finish) scoring chances for me. One of the reasons Alex Edler appeals to me as a player is that he has an excellent shot. It goes on the plus side of his ledger. Why do you need more information than that?

    If you have to choose between a guy who’s a great skater but has just an OK shot, or a guy who’s a great shooter but is just an OK skater, all else being equal, which one will score you more goals? (and how many more?) You can’t really answer this question just by watching games and you need to answer it to be able to fully evaluate the effect players have on your team.

    Every player is a package of skills. Hypothetically speaking – all other things are never equal – the shooter will probably score more goals, but he may not be the better player. If the players are of more or less the same quality with the different skills, then the other players on the team dictate the choice and the degree of success each player would have. A team that doesn’t score enough would probably get more value by adding a shooter. A team that already scores a lot would probably get more out of the speed.

    a stats-based approach starts with the ultimate object of shooting–scoring goals–and work backwards

    Why? Work back to what? Why? We’ve already got goals. What are we trying to learn?

    Scoring rates and adjusted plus minus and most of the other new stats are all basically ways of looking at how many extra goals a team scores/allows because of an individual player

    If these statistics actually did that, they would be very useful. They aren’t even close. And to be a science, it has to be better than close. It has to work. Like in baseball where I can determine how many extra runs a player creates with a formula that works at both the team and the league level. It can make predictions. That is a testable hypothesis that passes the test.

    Are the hockey hypotheses testable? Do they make predictions? If I give you the scoring rates of every player in the league and the adjusted plus minus, can you tell me how many goals were scored in the league? If I give you those statistics for every player on a team, can you tell me how many goals the team scored and gave up?

    If you can’t – and nobody can – these are nothing but untestable hypotheses and numbers that may or may not mean anything. And simply because they have numbers in them doesn’t make them better than an untestable hypothesis about, say, leadership.

    Nowhere in his post does Andy say that the methods he endorses are unquestionably the best way to look at things.

    No, he doesn’t. He doesn’t even tell us which Edmonton players he thinks should be playing with Hemsky and Horcoff. He does seem to be saying a) That Jacques clearly does not score enough to do well there, b) Quinn put Jacques there anyway because Quinn is a traditionalist who overvalues physical play, and c) this is driving Andy insane.

    I’m saying that a) There is significant value in getting a player to the front of the net that will not show up in any of Andy’s statistics. b) Like me, Quinn believes getting somebody to the front of the net is a critical part of the offense. This is something big, strong players can do and small skilled ones cannot. Jacques may not be a great choice, but Quinn doesn’t have any great choices. c) Andy does not have to go insane unless he is prepared to argue that Quinn does not have a good reason for his choice.

    But at least they accept the fact that you can’t study this stuff just based on your visual observations about players, and make some attempt to study things systematically in a scientific manner, and put actual figures and testable hypotheses out there.

    Study what stuff? I can’t study anything based on my visual observation of a player. I accept that fact. I have no objection to anyone making an attempt to study things systematically in a scientific manner. I don’t believe anything can be learned by it, but I’m quite willing to be proved wrong. I did make the attempt, and could not make it work.

    I do object to pretending that we are talking science before we see testable hypotheses. I do object to pretending an untestable statistic is better than an untestable anything else.

    • Roberto says:

      This is something big, strong players can do and small skilled ones cannot. Jacques may not be a great choice, but Quinn doesn’t have any great choices.

      …except Quinn does have 6’4″ 240lb Dustin Penner at his disposal, who just so happens to have scored 5G and 4A in the Oilers’ first 8 games, and most surely has better hands than M. Jacques…

  3. James Mirtle says:

    As they say, “I didn’t know you felt that way about me.”

    I like what the numbers add to my understanding — they’re complimentary to watching a lot of games. Because it’s impossible to get a read on the entire league and all 700+ individuals, data like that which Gabe Desjardins puts together is invaluable when trying to keep tabs on more than just one team.

    It’s also a concrete way to say “This is why a player like Nick Schultz (or Paul Martin or whatever) is so useful and why he plays so many minutes a game.”

    I think they’d be more useful for a GM than a coach, though.

    • Tom says:

      I like what the numbers add to my understanding — they’re complimentary to watching a lot of games. Because it’s impossible to get a read on the entire league and all 700+ individuals, data like that which Gabe Desjardins puts together is invaluable when trying to keep tabs on more than just one team.

      It’s also a concrete way to say “This is why a player like Nick Schultz (or Paul Martin or whatever) is so useful and why he plays so many minutes a game.”

      I agree that the league is too big to track unless doing so is a full time job. I really only pay attention to the West, aside from the standings. I don’t have a problem using the statistics this way, as long as we accept that the reliability is dubious and we don’t try to decide Schultz is better than Martin (or vice versa) because of an unreliable number.

      To me the most telling point in your comment is backwards. The fact that the coach gives Schultz and Martin so much ice time is the best indication of their quality. Guys who get the job done, get the ice time. All of the statistical evaluations of the Martin-Schultz type of player tend to reflect the team. The shutdown d-man on a good team will have good numbers, the shutdown d-man on a bad team will look bad statistically.

      I think they’d be more useful for a GM than a coach, though.

      If they were reliable, they’d be most useful for the media and the fans. Teams have the technology and the resources to know every player in the league.

  4. Magicpie says:

    Every player is a package of skills. Hypothetically speaking – all other things are never equal – the shooter will probably score more goals

    See, but how do you know this. How do you know the fast skater won’t score more goals because his speed lets him get shots off from more dangerous places than the shooter, for instance? What do you have to tell you that she shooter will score more goals besides sort of “just knowing”?

    Why? Work back to what? Why? We’ve already got goals. What are we trying to learn?

    By work backward I meant that instead of studying all of a player’s skills, ranking them, and trying to figure out how they lead to goals, they start with goals created/allowed judge a player based on that end product instead of trying to figure out how the aforementioned skills work together to create that end product.

    If these statistics actually did that, they would be very useful. They aren’t even close.

    Well no, that’s exactly what they do. If a stat says that last season the Wild scored 3 extra goals per 60 minutes of ice time when Gaborik was on the ice instead of Minor Leaguer X, then last season the Wild scored 3 goals per 60 minutes of ice time when Gaborik was on the ice instead of Minor Leaguer X. You’re right in the sense that this doesn’t prove that Gaborik necessarily increases a team’s goalscoring by that amount, but I don’t think that something like this is untestable. If the stat says that Gaborik increases a team’s scoring by that amount in that situation 5 seasons in a row, then there’s probably something to it. If Gaborik goes to the Rangers and their scoring increases by 3 goals a game while he’s on the ice and the Wild’s drops by that amount there’s probably something to it. Smarter people than me can probably think of better ways to test it. Any number you put out there is going to be testable to some extent, and I don’t really see what basis you have for saying that it can’t.

    • Tom says:

      See, but how do you know this. How do you know the fast skater won’t score more goals because his speed lets him get shots off from more dangerous places than the shooter, for instance? What do you have to tell you that she shooter will score more goals besides sort of “just knowing”?

      I’m missing something here. One of the two players will score more goals than the other. Ceteris parabus, I’ll take the guy who scores the most goals. Why do I care whether he scores them with speed or a shot? Why do we want this information? What kind of team decision does it drive?

      Shooting is a very important skill because every goal scorer has a good shot. I can think of lots of players who parlayed not much more than a shot into a lot of goals and a long career. The speedster with the okay shot has an upside limit. He can be good but probably not great. I’m fairly confident that in making that prediction simply because I can’t think of a single goal scorer who didn’t shoot the puck.

      I’m also fairly confident that a player with a good shot will score more on a team with a shortage of players who can shoot than he will on a team with lots of good shooters. Assuming his coach understands the game, he will get more opportunities if he has the second best shot on the team than if he has the fifth best shot.

      Well no, that’s exactly what they do. If a stat says that last season the Wild scored 3 extra goals per 60 minutes of ice time when Gaborik was on the ice instead of Minor Leaguer X

      Except the stats say no such thing. Somebody has invented a formula that produces a statistic that says this, but there is no validation of the formula. That’s what I mean by testing it. In baseball, the formula has been validated in the sense that it will make accurate predictions.

      I can point out the flaws in this particular in the idea without actually seeing it, but it isn’t easy to make sense of it. Why 60 minutes? Is this supposed to be saying that if Gaborik played the entire game the Wild would score three more goals than if a minor leaguer played the entire game in his place? The comparison doesn’t make sense in another way, too. Gaborik’s numbers are based on the ice time he got which includes as many offensive shifts as is practical. If he is out of the lineup, Minor Leaguer X does not get his ice time. The second best Wild is given the most opportunity to score, everybody shifts up, and the minor leaguer gets the ice time of the 12th forward.

      Even if it did make sense, what is the value in knowing this piece of information?

  5. Magicpie says:

    I’m missing something here. One of the two players will score more goals than the other. Ceteris parabus, I’ll take the guy who scores the most goals.

    Me too, but what I was asking is how do you know, all else equal, the good shooter will score more goals than the the good skater.

    Except the stats say no such thing. Somebody has invented a formula that produces a statistic that says this, but there is no validation of the formula.

    Maybe you’re referring to something different, but these stats aren’t created out of some invented formula. In the example I gave, you’d get that stat by simply looking at last year’s box scores and time played, and seeing how many goals per 60 minutes Gaborik scored while he was on the ice at even strength and how many minor leaguer X scored at even strength. You don’t need any kind of invented formula to do this, it’s basically no different than counting goals and assists.

    Why 60 minutes? Is this supposed to be saying that if Gaborik played the entire game the Wild would score three more goals than if a minor leaguer played the entire game in his place? The comparison doesn’t make sense in another way, too. Gaborik’s numbers are based on the ice time he got which includes as many offensive shifts as is practical. If he is out of the lineup, Minor Leaguer X does not get his ice time. The second best Wild is given the most opportunity to score, everybody shifts up, and the minor leaguer gets the ice time of the 12th forward.

    Even if it did make sense, what is the value in knowing this piece of information?

    The per 60 minutes doesn’t matter, it could be per 20 mins, per 10 mins, whatever. What matters is that the stat is per a specific unit of time so you take differences in ice time out of the equation. With regards to your second point, the comparison doesn’t have to be with Minor Leaguer X, it can be Gaborik and Brunette, Gaborik and Bouchard, whoever.

    The value of knowing this is that you’d know how many more goals your team scored last year by playing Gaborik instead of Bouchard, Brunette, or Minor Leaguer X, and this should to some extent predict how many extra goals your team would score this year by playing Gaborik instead of those players (the extent to which the stat predicts this is questionable, yes, the same way a players points totals from last year don’t necessarily predict the point total’s he’ll get this year, but the stat does put a testable number out there).

    • joe cheesy says:

      So, if it’s testable can you give examples? Walk us through the hypothesis, the test, and the result? I’m dying to see something concrete.

  6. Magicpie says:

    arg..mistyped that should say how many goals Gaborik’s TEAM scored while he was on the ice at even strength

  7. Tom says:

    The per 60 minutes doesn’t matter, it could be per 20 mins, per 10 mins, whatever. What matters is that the stat is per a specific unit of time so you take differences in ice time out of the equation. With regards to your second point, the comparison doesn’t have to be with Minor Leaguer X, it can be Gaborik and Brunette, Gaborik and Bouchard, whoever.

    Well, you can’t take the ice time out of the equation because Gaborik is going to be used differently. The quantity of ice time is not the quality of ice time. If you are saying the Wild will score more when their best players are on the ice – that’s who Gaborik will play with – when the coach is giving that group the best offensive opportunities with matchups and offensive zone faceoffs, I think you are simply stating the obvious.

    A replacement level player does not get those opportunities and plays a lot less. You can’t take his results playing with the worst of the Wild, equalize the ice time and pretend you are comparing apples to apples.

    The value of knowing this is that you’d know how many more goals your team scored last year by playing Gaborik instead of Bouchard, Brunette, or Minor Leaguer X, and this should to some extent predict how many extra goals your team would score this year by playing Gaborik

    It “should to some extent” predict… I don’t believe it. What good is “it should to some extent predict”… Something either predicts or it does not. Gaborik is not in the Minnesota lineup this year. Can you predict how many goals they will score this year? How? Can you predict how many extra goals the Rangers will score this year? I think it is reasonable to predict that Minnesota will score less and the Rangers will score more because Gaborik is a hell of a player, but quantify it?

    If Havlat scored 2.5 goals more than a replacement player in Chicago is that because he is worse than Gaborik or does it mean that Chicago has better players to play when Havlat is off the ice? If Havlat did score 3.1 goals more per 60 minutes and Gaborik 3.0, would you choose Havlat over Gaborik?

Trackbacks

Check out what others are saying about this post...


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!