Thursday, July 31st, 2014

Tyler and Dave

36

Tyler Dellow is one of my favourite bloggers. I’ve read him for years. If pressed to explain why when I don’t hold much truck with hockey analytics, I say something like this, “He’s got an interesting point of view and he writes well. I don’t think the data he uses has anywhere near the validity that he thinks, but at least it is different. Like most fans, I’m beyond tired of the usual narratives most of the mainstream media attach to the players and the teams. I take different bullshit over the same old bullshit.”

In a recent post Tyler is all over some comments Dave Nonis made about the use the Leafs make of hockey analytics. (They don’t make much of them at all. Mirtle has more.) Tyler begins with:

For reasons I’ll never understand, they never really have a coherent advocate of hockey analytics on these panels – at most, you get some sort of a nebbish character who’s intimidated by hockey people. At worst, you get a guy who knows less than nothing about the topic on the panel, saying things that mean nothing. The result tends to be pretty useless to a listener, I think – you don’t get a discussion about the insights that can be drawn from and flaws of hockey analytics, you get silly criticisms leavened with ignorance.

No doubt, but to be fair to the nebbish character it is difficult to concisely state the chain of logic that leads to the insight without stumbling over problematic premises. And to be fair to Dave Nonis, there is no single big flaw in that quant’s logic. It is an accumulation of smaller problems so Nonis can’t make a coherent case either. The result is the quant throws out something that is incoherent to most fans and the Dave Nonis character seizes on one of the small problems and blows it all out of proportion.

Nobody is well served. The nebbish character does not put Tyler’s case very well, and Dave Nonis does a bad job explaining my position, which goes something like:

Corsi is not much different than the plus minus, a statistic that has been around for a long time. It uses attempted shots as a proxy for goals. The problem with the number is that credit for the goal (or attempted shot) is arbitrarily spread among all the players on the ice and the individual player statistic is hopelessly polluted by the quality of the team and the player’s role on the team. There is very little real evidence that Corsi actually measures individual player quality any more than the plus minus does. The narrative quants attach to the number – that it indicates whether or not the player drives “puck possession” – is nonsense.

Even if Dave Nonis thinks there might be something to examining Corsi numbers at the team level, there is very little he can do with the data. The only way to improve the shot differential is to become a better team and he’s already trying to do that. He’d surely prefer more shots in every game, but in the meantime he can take some comfort from the fact that for some inexplicable reason the Leafs have been a far better team when they are outshot for the past four seasons.

Most of the advanced stats are about shots, which are an outcome, not part of a process. To say “A team needs to get more shots (or give up fewer) to win more games” isn’t really different than saying “A team needs to score more goals (or give up fewer) to win more games.” It is usually true, but it isn’t helpful. No matter how the game’s shot statistics are parsed, none of them describe or measure the process that produces the shot or the goal. They do not describe hockey.

Hockey is a team game. On offense, players have to coordinate their play to exit the defensive zone with the puck, move quickly through the neutral zone, enter the offensive zone, and move the puck around the zone until somebody has a chance for a shot. A team that does those things well will get shots and goals. Defense is precisely the opposite. Players coordinate to mount a good forecheck to prevent easy exit from the zone, to harass and slow the play through the neutral zone so the defense can hold the blueline and force a dump-in. A team that does those things well will be hard to score on.

The Edmonton Oilers are not a lousy hockey team because their Corsi numbers suck. They are a lousy hockey team because it is easy to move the puck against them, and they do not get out of their own end very well. Their transition game isn’t good enough in either direction. When they do generate speed through the neutral zone or gain control in the offensive end, the young stars on the team can make some hay, but the Oilers don’t do either nearly often enough.

That is an opinion. It is unsupported by data, because there is no data that measures the parts of the game that either lead to a shot or prevent one.

According to Mirtle:

So while other teams like the Minnesota Wild have made enormous strides towards improving their even strength play at least partly because of their use of analytics and a buy-in from coach Mike Yeo, Toronto isn’t close to attempting a similar, data-driven shift in ideology any time soon.

I don’t think there is any evidence that the strides made by the Wild have anything to do with their use of analytics. Yeo is pretty vague about it, sounding pretty much like the Canucks on the subject. (Very important to leave no stone unturned, they say. We look at everything.) Nonis would probably be better off with his critics if he pretended he thought it was important stuff.

Me, I think the Wild improvement is mostly due to stronger goaltending and better players playing better. Yeo’s willingness to accept more risk defensively might have something to do with it, too, but teams make that kind of choice all the time. If there is an advanced statistic that informs that kind of reward vs risk assessment I have not seen it.

A Billy Beane approach is paying off in Minnesota? I call bullshit. Different bullshit, to be sure, but bullshit all the same.

Be Sociable, Share!

Comments

36 Responses to “Tyler and Dave”
  1. James Mirtle says:

    I don’t know where to start with this one.

    Tom, all stats are an outcome, in some sense. But shots are part of a process in that creating them is part of the process of creating goals. Teams that shoot more, score more. Teams that attempt more, score more. Teams that “Corsi” more, score more and prevent more. It gives you a “why” there by eliminating some of the variance involved in shooting percentage, etc.

    So the Wild becoming a top differential team is important. As important as getting better goaltending.

    But the advanced stats are also not just about shots or outshooting in individual games, which is the most common criticism. They’re about quantifying zone time using shot attempts. It doesn’t really matter if you want to call that possession or something else. What matters is that it’s worthwhile measuring this aspect of a team’s play as teams that do this over a full season are very good teams (100+ points). Teams that don’t are, in general, very poor (80 points).

    I’ll let Tyler address the rest of this, but the idea plus-minus is as useful as a stat like Corsi is completely absurd. Even just looking at the league leaders (and trailers) will show that.

    But you’re using bad data to try and condemn good when you talk about the Leafs record when outshooting. Extrapolate it over a larger sample size, add the attempt data, adjust for score effects, and then you have something of use. No one uses these numbers the way you’re presenting them here. You can’t discredit something by using a dumbed down, simplified version of it.

    • Tom says:

      But shots are part of a process in that creating them is part of the process of creating goals.

      I disagree. The process of creating a shot is exactly the same as the process of creating a goal. The only difference is that one shot went in and the other did not. From the offensive player’s perspective attempting to score a goal is no different than scoring one.

      Teams that “Corsi” more, score more and prevent more.

      Yes. The same thing is true for shots and goals.

      So the Wild becoming a top differential team is important. As important as getting better goaltending.

      The difference is that nobody sets out to become a top differential team. Aside from getting better generally, there is nothing anybody can do to become a top differential team. While stronger goaltending doesn’t change a shot differential it can change a goal differential.

      That differential is a result. No coach tells his team to go out there and improve their shot differential next period. He tells them they have to pay more attention to their gaps, or that he wants the defense to pinch more aggressively, or to make sure the dumpin can’t be handled by the goalie. The Wild are playing better.

      But the advanced stats are also not just about shots or outshooting in individual games, which is the most common criticism. They’re about quantifying zone time using shot attempts. It doesn’t really matter if you want to call that possession or something else.

      Well, I call it bullshit, but even if it is true that shot attempts quantify zone time, what does it tell us? I think it is a meaningless narrative to make it sound like they are talking about something besides shots. No matter what the narrative says, it is still just measuring shots.

      What matters is that it’s worthwhile measuring this aspect of a team’s play as teams that do this over a full season are very good teams (100+ points). Teams that don’t are, in general, very poor (80 points).

      This tells us what that we didn’t already know? The same teams have a very good (or bad) goal differential.

      I’ll let Tyler address the rest of this, but the idea plus-minus is as useful as a stat like Corsi is completely absurd. Even just looking at the league leaders (and trailers) will show that.

      I don’t think the plus minus is very useful at all. I don’t, because it has the same flaws as Corsi. I don’t think either are very meaningful or useful. A 53% Corsi% for Phil Kessel does not mean the same thing as a 53% Corsi% for David Booth or Chris Tanev. A 53% Corsi% in Toronto is not the same thing as a 53% Corsi% in Vancouver. I cannot look at that number and say, “Oh boy. Let’s trade for him. If we add him our Corsi will improve and we will get better.” The reverse is probably more likely to happen. The player’s Corsi will drop when he goes to a bad team.

      Extrapolate it over a larger sample size, add the attempt data, adjust for score effects, and then you have something of use.

      Of use to do what? To think that we have figured out a way to objectively evaluate individual hockey players? None of these numbers can possibly do that. If they cannot do that, what do they do?

      I agree that I am simplifying, but my criticism is fundamental. It applies to plus minus and it applies to all the proxy variations of plus minus. No matter how you fiddle with the number, you cannot separate the individual from the team.

      No one uses these numbers the way you’re presenting them here.

      How are they used? How is Dave Nonis supposed to use them?

      • James Mirtle says:

        Well goal differential includes goaltending in it. Goaltending is quite volatile, as we all know – one year a guy can post a .930, the next he can be .910. Isn’t it worthwhile to have stats that are independent of goaltending to quantify a team’s play? Especially when goaltending varies to a large extent when certain players are on the ice (i.e. some players are there when every save is made).

        Most of the possession stats are also calculated on the individual level as relative to their teammates – so by definition they’re separating players from their team (to some extent). Teams have a wide variety of different Corsi numbers on them – if it’s all team based, shouldn’t they all be grouped together?

        How could Nonis use them? For starters, he’d see that a player like Holzer was getting filled in on his blueline and demote him to the minors after less than the 22 games it took them. He’d value Grabovski a little more and Bozak a little less. He’d realize McClement is a great penalty killer but probably overused at even strength. He’d learn that playing Orr and McLaren on the same line resulted in the worst possible outcome of any other duo on the team but that lines with only one of them were much less vulnerable. He’d see that players like Bolland were succeeding remarkably despite playing in tough roles and realize that acquiring Ryan O’Byrne at the deadline was idiotic given he was no longer an NHL level player.

        And on and on. I could give an example about every single player on the team if you want.

        Yes, the numbers are influenced to some extent by teammates. That’s why when you talk about them, you talk about who they’re often on the ice with or against as a way of providing context. Every statistic available requires context. Otherwise, you can argue goals aren’t a valuable measure of a player because Tlusty was tied for fifth in them last season. Or Kunitz was a top 10 player in scoring. Teammates affect those numbers, too. Does that make them invalid as measures of player performance?

        I think your problem is you come to these statistics from the wrong perspective. Rather than look at the data first and draw conclusions later, you come with a narrow view of what hockey “is” and what defines success in the game. I was very skeptical about some of these stats in the beginning, but the longer I’ve followed them, the more they’ve forecasted both team and player play correctly. You say you don’t care about making predictions; what exactly is a GM’s job if not to make bets on players based on such predictions? And this data, while not foolproof and with some limitations, helps to do so.

        • Tom says:

          I think your problem is you come to these statistics from the wrong perspective. Rather than look at the data first and draw conclusions later, you come with a narrow view of what hockey “is” and what defines success in the game.

          Science can’t work this way. You begin with a hypothesis and then you find data to confirm or deny it. Tyler does his best work when he finds data that shows a position taken by a coach or pundit or analyst is ridiculous. I think there is lots we can learn about hockey – maybe even hockey teams – if we learn to ask the right questions and look for the right data. What we can’t do is evaluate the quality of individual players with any kind of precision. The data just is not there.

          It is a mistake to start with the data and try to suss a narrative and some conclusions out of it. Why? Because humans can always find patterns in data and we are great storytellers. Once we believe in something we will seek out and find other examples. We notice every good prediction and our beliefs are reinforced. Bad ones are explained away with context or the need for more time or a larger sample. None of us can avoid these kinds of traps when we start with the data.

          Nonis knew Holzer was getting filled in. The Leafs have some system of grading every player every game, don’t they? Nonis and his staff do their best to evaluate the talent. If I ran the Leafs. my video coach would grade out every shift. I liked Grabovski better than Bozak too, but Randy Carlyle clearly disagrees with me and whatever the numbers say. Neither one of us can prove he is wrong.

          I can remember when Tyler was pounding on Joffrey Lupul and he was right to pound on him. Lupul was a pretty awful player in Edmonton. He was so bad I figured Burke was forced to take his contract in the deal with Anaheim. His numbers said he was worthless.

          Was acquiring Lupul a mistake?

          • Roberto says:

            What we can’t do is evaluate the quality of individual players with any kind of precision. The data just is not there.

            What science most definitely does not do is negate questions or declare that some things are not open to questioning, which is what you have done there.

            To say the statistics and analyses are incomplete, flawed, or even useless is one thing. To declare that even the inquiry is a useless endeavor, based on nothing but specious argument and self-serving definitions is quite another.

          • Tom says:

            What science most definitely does not do is negate questions or declare that some things are not open to questioning, which is what you have done there.

            I don’t think I have done that at all. I am saying that you cannot evaluate players with any precision with this data. It does not work.

            Sometimes you have to take a different approach or find other data to answer a question instead of pretending a flawed approach has meaning because it is the only data available. Corsi is not convincing to me or to Dave Nonis.

            The idea that some of the quants are looking at zone entries and zone exits is a good thing. It may not lead anywhere – I don’t think so – but at least it is different. I wish the league counted touches. I’d guess that the team with the most touches wins as consistently or more consistently as the team with the most shots. I’d like to see a touches plus minus, DZ touches, NZ touches, OZ touches. Touches per minute. Or maybe the data would not show anything.

            To say the statistics and analyses are incomplete, flawed, or even useless is one thing. To declare that even the inquiry is a useless endeavor, based on nothing but specious argument and self-serving definitions is quite another.

            I don’t think inquiry is ever useless. I do think it is wrong to actually use the results of that inquiry if they are are incomplete, flawed or useless.

            I don’t think it is fair to declare my argument to be specious without actually addressing the argument. I’m trying to make my criticisms of the approach very clear. If I’m wrong, show me where. I don’t think it is fair to simply call my definition of the game self serving either. I’ve tried to describe the game as clearly as I can as well. If my self serving definitions are wrong, show me where.

  2. PopsTwitTar says:

    “Hockey is a team game. On offense, players have to coordinate their play to exit the defensive zone with the puck, move quickly through the neutral zone, enter the offensive zone, and move the puck around the zone until somebody has a chance for a shot. A team that does those things well will get shots and goals. Defense is precisely the opposite. Players coordinate to mount a good forecheck to prevent easy exit from the zone, to harass and slow the play through the neutral zone so the defense can hold the blueline and force a dump-in. A team that does those things well will be hard to score on.”

    Right. And teams that do all of those things tend to have good team Corsi/Fenwick stats.

    Sure, there are not perfect ways to connect Corsi/Fenwick to the actions of each individual player – but those are being developed and considered. For example, we see more people now tracking zone exists and entries. Do you believe that the players who do well in those stats are going to have *bad* Corsi/Fenwick? Do you doubt that in a few years as we track those specific individual plays more closely, we wont be closer to “describ[ing] or measur[ing] the process that produces the shot or the goal”? We wont be closer to “describ[ing] hockey”?

    • Tom says:

      Right. And teams that do all of those things tend to have good team Corsi/Fenwick stats.

      Probably so.

      Sure, there are not perfect ways to connect Corsi/Fenwick to the actions of each individual player – but those are being developed and considered.

      There aren’t any imperfect ways either, are they? Goal differentials and proxies for goal differentials clearly reflect quality of team. If that was all there was to it, nobody would care. The huge leap of the plus minus – and all the Corsi variations – is the assumption that you can fairly award five credits and five debits for every shot to every player on the ice.

      The unstated assumption is that the connection between Corsi and goals that exists at the team level is valid at the individual level. That assumption is very problematic. I don’t see any evidence to support it.

      For example, we see more people now tracking zone exists and entries. Do you believe that the players who do well in those stats are going to have *bad* Corsi/Fenwick?

      Players who do well on these stats? These are not usually individual activities. At least two or three players are usually involved and if the others aren’t in the right place too, it probably breaks down. All of the things I describe in that paragraph are team activities. A Sidney Crosby probably doesn’t contribute very much more than the average Penguin in those elements. The big difference between Crosby and the next guy shows up after the Pens gain the offensive zone. I would expect the top centres on the best teams would lead the categories.

      Do you doubt that in a few years as we track those specific individual plays more closely, we wont be closer to “describ[ing] or measur[ing] the process that produces the shot or the goal”? We wont be closer to “describ[ing] hockey”?

      I do doubt it, but I’ve got an open mind. When I look at the process of producing (or preventing shots) I don’t see what to count. That said, I’m perfectly willing to concede we may eventually find a measure that better describes the quality of an individual player. That would be great.

      Until then, I’m with Dave Nonis. I don’t see how it is useful.

      • Nanodummy says:

        Uh, Zone Entries can be quite a solo act, and there seems to be a correlation between those players able to enter the zone with possession and players generating more shots, versus those who dump and chase.

        And you agree that shots=goals.

        So what we have here is a stat that is much more isolated (at most three players might be involved in a zone entry on one team, but more likely it’s one or two) that leads to a result: more shots and likely more goals.

        So you can start tracking players raw possession entries vs. failed entries vs. dump-ins and get a read on that kind of player without having to consult the tape every single time. One check of the tape by an intern paid to count this stuff is enough. And that data can recontextualize the game where a player fluked out two goals, appearing great, but actually did many of the small things wrong. This is why data like this is valuable.

  3. beingbobbyorr says:

    Tyler Dellow is one of my favourite bloggers. . . . If pressed to explain why when I don’t hold much truck with hockey analytics . . .

    They’re few and far between, but the essays he writes, sans MoneyPuck, can be a master class in what great sports journalism should be: question accepted wisdom, gather evidence, posit alternatives, and do it with style and humor. I’m thinking of the pieces he wrote on Colin-Campbell-gate and the NHL awards voting process (titled “A trip to the sausage factory” IIRC).

  4. beingbobbyorr says:

    . . . comments Dave Nonis made about the use the Leafs make of hockey analytics. (They don’t make much of them at all.) . . . I don’t think there is any evidence that the strides made by the Wild have anything to do with their use of analytics.

    I place no faith in any public statements about MoneyPuck made by NHL/franchise employees (GMs, coaches, Directors of Player Personnel, scouts, etc.,). I assume that the truthful answer — were they motivated to give it — would be “we’re studying the hell out of this stuff with our own versions of Bill James, Paul DePodesta, etc., and — whatever (if anything) we’ve found so far — it’s not in our interest to disclose a damn thing (yet)”.

    There’s also something fishy in the Nonis quote “The last six, seven years, we’ve had a significant dollar amount in our budget for analytics and most of those years we didn’t use it.” If there is a budget to do X, that implies that either (a) the GM thinks X is important and ASKED for money to do it, or (b) somebody in the executive suite (who controls the piggy bank) thinks that X is important and PUSHED money across the table to hockey ops with the intent they pursue X (even if it leads nowhere; i.e., Gillis’ No Stone Unturned). Nonis all but says that (a) is not the case, leaving us with (b), to which Nonis-supporters can only say: we hope he knows what he’s doing, because being a Refusnik with respect to your employer’s philosophy usually gets one shown the door sooner rather than later.

    The people who should have an interest in accelerating MoneyPuck’s lexicon into the hockey discourse are player agents. It gives them more data (regardless of it’s validity) with which to bedazzle GMs at the negotiating table (or confuse fans in the court of public opinion).

  5. beingbobbyorr says:

    Corsi . . . . uses attempted shots as a proxy for goals.

    We don’t need a proxy for goals (we already have the direct measurement: the score). Isn’t the correct way to put this more like

    “Corsi . . . . uses attempted shots as a proxy for measuring WHO on our team is helping tilt the ice in the correct direction and therefore LIKELY to be a KEY CONTRIBUTOR TO our scoring more / being scored upon less, because goals (+/-) are too-small-a-sample-size and small sample sizes are more pollute-able by noise/randomness.”

    • Tom says:

      I think you are mixing up two different ideas here. The entire tilting the ice idea is very problematic. Set it aside for now. Otherwise, I agree. One of the big problems with the plus minus has always been sample size. Another big problem is that arbitrarily divides up the credit or blame for the goal. As a result a player on a good team usually has better results than the same player on a bad team. The plus minus was attached to the individual, but it was obvious to everyone that it did not tell us very much about the quality of the player.

      At some point (I think it was Klein and Reif back in the 80′s) somebody realized shots correlated pretty well with goals. When Corsi came along, somebody else discovered that Corsi correlated even better. The larger the sample size, the better the correlation. The sample size problem may not have been entirely eliminated, but it is certainly lessened. If goals are scattered more or less randomly among Corsi events, the attempted shot serves as a large sample size proxy for goals whether we call it tilting the ice, puck possession or anything else. (And if it does not serve as a good proxy for goals, what is the point in tilting the ice or puck possession?)

      Unfortunately, when you look a little closer, you realize that goals are not scattered more or less randomly among Corsi events. A shot taken by Steven Stamkos is much more likely to go in than a shot taken by Tom Sestito or Chris Tanev. Shots taken against a team that has a big lead are less likely to go in. Teams that collapse to the net give up more low percentage point shots than teams that pressure the points. And so on. These wash out in the big picture, but they do not wash out of games and they do not wash out of individuals.

      If goals are not scattered randomly among Corsi events, it all breaks apart. When you roll up the data into large samples, the correlation works because the Stamkos brilliance and the Sestito ineptitude washes out. But when you bring it back to the game level or individual player level correlations are not very meaningful. A big enough sample and all shots are equal. In the real world, all shots are not equal.

      To me, that blows a huge hole in Corsi even before one considers whether it is reasonable to spread the credit for the shot, the team quality or role within the team.

      It also raised questions among the quants. One solution was to throw out gobs of ice time where the correlation between shots and goals deteriorates. I don’t think that helps. If the correlation with goals does not hold up in all game states, we have to question the number in all game states not just the ones it does not hold up in.

      Quants will also acknowledge there is something called finishing ability when an objection like Stamkos vs almost anyone else comes up without acknowledging the problem that creates when considering the Corsi number produced by an average player. The shots generated when he is on the ice are probably not the same value as shots generated when an above average player is on the ice. But they are the same value in the statistic.

      The puck possession, tilting the ice narrative doesn’t change anything for me, but your mileage may vary. Instead of having to explain why shots don’t pay off at the same rate when a team is trailing in the third period, they should have to explain why tilting the ice doesn’t pay off when a team is trailing in the third period.

  6. beingbobbyorr says:

    The problem with the number is that credit for the goal (or attempted shot) is arbitrarily spread among all the players on the ice . . .

    That’s true for individual goals (or attempted shots), which is why we try to aggregate a crapload of such events (seek large sample sizes, ergo shots attempted is more useful than goals scored), so that the signals have a better chance of rising above the noise.

    . . . the individual player statistic is hopelessly polluted by the quality of the team and the player’s role on the team.

    In an ideal world, coaches would methodically vary their line combos/D-pairings in the same way engineers use DOE (Design of Experiments) to determine sensitivity to interactions among variables. Sadly, coaches don’t work in a laboratory; they have to win games now, and so, lacking time, their gut prevails. But, to the extent that line combos/D-pairings do get re-arranged periodically because of inevitable injuries and/or the need to ‘get so-and-so going’ (change something in his environment/routine), the results of such brief experiments are part of MoneyPuck’s attempt to refine QoT, QoC, and zonestarts in the hopes of teasing out these effects.

  7. beingbobbyorr says:

    The narrative quants attach to the number (Corsi) – that it indicates whether or not the player drives “puck possession” – is nonsense.

    Actually, we’re talking about 4 different things:

    1) Score = goals for vs. goals against
    2) Corsi = shots for vs. shots against
    3) Ice Tilting = percentage of time the puck spends in O-zone vs. N-zone vs. D-zone
    4) Possession = percentage of time the puck spends on our team’s sticks vs. on the other team’s sticks

    How any of #2, #3, #4 correlate to #1 should not need to be a matter of opinion. Surely this has been (or could be) a project of the analytics crowd (granted item #4 is problematic due to viewer/building judgment variances), but, once done (adequate sample size of games, across random teams, across multiple seasons, etc.,), then #2, #3, or #4 gives us an order-of-magnitude larger sample size from which to then hopefully be able to ferret out individual contributions. Of course, if there is no correlation (of any of #2, #3, or #4 with respect to #1), then Corsi, Ice Tilting, and Possession go out the window and we have to start looking somewhere else.

  8. beingbobbyorr says:

    To say “A team needs to get more shots (or give up fewer) to win more games” isn’t really different than saying “A team needs to score more goals (or give up fewer) to win more games.”

    Yes, they are different. The later is merely the definition of winning games. The former may be one of multiple methods of/clues to achieving that end (another being the as-yet-undetermined thing that TML has been doing the last 4 years; some posit that they’re doing it with “shot quality”, as opposed to “shot volume (Corsi)”).

  9. beingbobbyorr says:

    It is usually true, but it isn’t helpful. . . . If there is an advanced statistic that informs that kind of reward vs risk assessment I have not seen it.

    The point of MoneyPuck should not be to prove preconceived notions (“Corsi rocks!”) or throw up our hands when one or more cases of a preconceived notion fails (“But Corsi MUST correlate with/drive winning, you damn 2009-13 TML!”), but rather to find out WHAT drives “consistently scoring more goals than the other team”. There could be many answers to that. Corsi might be important. Or it could be a dead end. Or something in between. But exploring it — even if it is a dead end; hell especially if it’s a dead end — still has value in trying to turn the inexplicable (TML’s last four seasons) into the explained.

    It’s helpful if we look at today’s still-nascent Advanced Stats the same way scientists look at their experiments: as one step (toward more future steps as they hopefully asymptotically approach reality/truth), rather than ladening them with the expectation to be the Holy Grail itself (this early in the game). I suspect that part of Tom’s problem with MoneyPuck is that, to date, they’re just a hodge-podge of disjointed metrics that have not (yet) been synthesized into any kind of predictive tool (singular). But, just like scoring a goal in hockey, it’s a process, and the MoneyPuck project is currently at the regroup-and-break-out stage, nowhere near ready for that game-winning one-timer.

    If this assumption — that we’re in the Jurassic era of MoneyPuck — is true, then I would caution MP enthusiasts against attempting to predict team performances until the community can first make meaningful statements about individual players (measuring past contributions, then predicting future performances).

  10. Tom says:

    I suspect that part of Tom’s problem with MoneyPuck is that, to date, they’re just a hodge-podge of disjointed metrics that have not (yet) been synthesized into any kind of predictive tool (singular).

    Not quite. I realize that is asking too much. We just don’t have the data. It works in baseball because we know the value of a single – in terms of runs and wins – and we know who is responsible for that single. A single has pretty much the same value no matter who hits it.

    As a result, if you know how many hits a team had, we can accurately predict how many runs they will score. And therefore an individual player’s batting record can logically be converted into runs. His offensive contribution is very well measured.

    In hockey we have shots and goals and they do correlate when the sample is large enough. That’s a start. I might know the value of a shot just like the baseball analyst knows the value of a single. If all shots are equal with goals scattered randomly among them, the logic works. Knowing a team’s shot total will tell me how many goals they scored. If that is the case, maybe it is reasonable to split the credit for the shot among the skaters and trust that the shot leads to goals and to wins.

    The problem is that goals are not distributed randomly among the shots and so even before we try to divide up the credit for the shot, anomolies at the team level start to pop up. One would expect an outlier or two, but I hardly started to look at it and the edifice started to crumble. It is not just the Toronto Maple Leafs. In the past three years the entire league has played better – quite a bit better – when they are outshot. Vancouver and St. Louis are two other teams that win a lot more when they are outshot even though, unlike the Leafs, they get more shots than their opponents in the aggregate.

    If the team level data made sense, it might be reasonable to try to extend the logic to the individual player. But I do not think it is reasonable to extend it to the individual without resolving the anomolies. If we cannot trust the relationship between shots and goals and wins at the team level, how do we trust it at the individual player level when we split up the team data?

    I would caution MP enthusiasts against attempting to predict team performances until the community can first make meaningful statements about individual players.

    I think this is backwards. First, we make the team data make sense. If we get consistent results at the team level, we can then try to break the team performance into individual contributions. But if the team performance does not make sense logically, you cannot divide that performance up and expect it to be meaningful. At least I don’t think so.

  11. Excellent blog here! Also your website loads up fast!
    What web host are you using? Can I get your affiliate
    link to your host? I wish my web site loaded up as quickly as yours lol

  12. Excellent web site you’ve got here.. It’s difficult to find high quality
    writing like yours nowadays. I truly appreciate individuals like you!
    Take care!!

  13. Hurrah, that’s what I was exploring for, what a stuff!
    present here at this web site, thanks admin of this website.

  14. David Blair says:

    At this time I am ready to do my breakfast, afterward having my breakfast coming yet
    again to read further news.

  15. rhinoplasty says:

    An fascinating discussion is value comment. I believe that you must write extra on this subject, it won’t be a taboo topic but typically persons are not enough to speak on such topics. To the next. Cheers

  16. Wow! This can be one particular of the most helpful blogs We’ve ever arrive across on this subject. Basically Excellent. I’m also an expert in this topic therefore I can understand your effort.

  17. Hi there i am kavin, its my first time to commenting anyplace, when i read this post
    i thought i could also create comment due to this sensible
    paragraph.

  18. playstation says:

    If you like open world, adventure and role playing games then the Move may not be a good choice for you.
    The P90 is a specialised weapon that provides you with good mobility during combat,
    thanks to its shape. Are all available for you to buy and download to
    your PS3.

  19. It’s great that you are getting ideas from this article as well as from our
    dialogue made here.

  20. Link exchange is nothing else but it is simply
    placing the other person’s blog link on your page at proper
    place and other person will also do similar in support of you.

  21. Hi there! Do you know if they make any plugins to protect against hackers?
    I’m kinda paranoid about losing everything I’ve worked hard
    on. Any tips?

  22. It’s in fact very difficult in this busy life to listen news on Television, thus I simply use web for that purpose,
    and take the latest information.

  23. review says:

    We are a group of volunteers and opening
    a new scheme in our community. Your website offered us with
    valuable info to work on. You have done an impressive job and our whole community will be grateful to you.

  24. Howdy! This is kind of off topic but I need some advice from an established blog.
    Is it very hard to set up your own blog? I’m not very techincal but I can figure
    things out pretty quick. I’m thinking about creating
    my own but I’m not sure where to begin. Do you have any tips or suggestions?
    Appreciate it

  25. great points altogether, you simply received a new reader.
    What may you recommend about your put up that you made some
    days ago? Any sure?

  26. Can I simply just say what a comfort to find someone
    that really knows what they are discussing on the net.
    You actually know how to bring an issue to light and
    make it important. A lot more people have to check this out and understand
    this side of your story. I was surprised that you aren’t more
    popular since you most certainly have the gift.

  27. Advice says:

    Wow! In the end I got a website from where I
    be able to actually take helpful data regarding my study and knowledge.

  28. Wonderful beat ! I wish to apprentice at the same time as you amend your web site, how could i subscribe for a blog site? The account helped me a acceptable deal. I have been a little bit acquainted of this your broadcast offered brilliant transparent concept

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!