• New Metrics I

    by Tyler Dellow • September 16, 2013 • Hockey • 17 Comments

    I’m playing around with some new metrics at the moment. They’re all related to the same sort of interest: pushing past Corsi% as a metric and getting into the stuff underneath it. As I’ve mentioned in a previous post, I’m calculating what I call an open play Corsi% – basically, I knock out the stuff after faceoffs and then the stuff I’m left with, theoretically, doesn’t have any faceoff effects. It’s just guys playing hockey.

    I’ve lifted the phrase “open play” from soccer, where they talk about “open play” and “set pieces.” Basically, a set piece is a corner kick or when a player gets fouled and a free kick is awarded. Open play is what happens when the game’s just flowing. There’s an obvious comparison with hockey and when there’s a faceoff: teams can game plan for faceoffs in each location and what they want to do a bit more easily.

    Two of the metrics I’m playing around with are the same concept, just for offence and defence. What I’m doing is breaking the game down into shifts by way of a proxy. For every time that a player’s on the ice for SF, I call that a shift. If there’s a shot within the next 60 seconds, I count that shot as part of the same shift. There’s going to be a bit of a fudge factor in this but not too much of one, I think – the new stats site Extra Skater ran a test of my methodology and found it was pretty much bang on with actual shifts. The downside to what I do is that I don’t count shifts on which there are no shots. The upside is that I don’t have to enter into the tangled thicket of the NHL’s shift data.

    Anyway, so I end up with a count of how many shifts a player was on for 1 shot attempt for (SAF), 2 SAF etc. Same for 1 shot attempt against (SAA), 2 SAA etc. (Side note: shot attempts are the same as Corsi events: goals, shots, blocked shots and missed shots.) What I’m interested in is this: once we know that a shot attempt has taken place, how good are players at preventing the second attempt or, if they’re on the offensive side of things, at generating that second, third shot attempt.

    What this gives me is a sort of breakdown of a player’s shifts and what percentage of those with at least one SAF were 1, 2, 3 etc. I can then create a number that answers the question “Given 100 shifts with at least one shot attempt, how many shot attempts for/against would we expect this player to be on the ice for?”

    Whenever I’m introducing something like this, I like to do it by way of percentiles, so that I can illustrate the range. I also like to break things down between forwards and defencemen – you never, know sometimes you find some some unexpected differences between the two groups. So first up, we’ll look at the range in SAF/100+ SAF for forwards for the 2007-13 seasons.

    Screen Shot 2013-09-16 at 12.29.31 AM

    A statement of the somewhat obvious: there seems to have been a gradual drift upwards in terms of the SAF/100+ SAF that forwards compiled. This may be a result of the NHL scorers getting better at crediting misses and blocks or teams beginning to shoot the puck more frequently or a combination or some unexplained third thing.

    If you think about what we’re doing here, we’re starting to break how a player achieves his Corsi% down into small, specific chunks. In effect, what this asks is, “To what extent have generating multiple shot attempts once you’ve got the first one contributed to your Corsi%?”

    Another statement of a potentially obvious fact: there doesn’t look to be a HUGE difference between the guys who are pretty good at this and the guys who are kind of bad. The difference between best and worst? Yeah, that seems pretty large. As far as the rest of it goes though, you end up pretty quickly with a sort of muddled middle without huge differences.

    To illustrate things, we’ll look at two players who’ve been on different arcs these past few years: Sidney Crosby and Alexander Ovechkin. Ovechkin was, as we all know, a Corsi% monster who kind of seemed to slow down and has never really come back. Crosby wasn’t really one until recently. What does it look like if we graph their SAF/100 Shifts with 1+ SAF for the six years for which this data is available?

    Screen Shot 2013-09-16 at 12.29.46 AM

    As it turns out, it’s pretty cool and it illustrates a lot. Ovechkin led NHL forwards in this stat in 2007-08 and 2008-09. He was a hair off the lead in 2009-10, finishing behind only Michael Grabner, who didn’t play a ton. He slipped in 2010-11, although he was still well above the 90th percentile. These last two years though…there’s just not a lot there. Hart Trophy last year or not, part of what made Ovechkin really cool to watch was the sort of maurading that went on, the sense of the Caps and Ovechkin dominating shifts. That hasn’t been around for a while.

    Contrast that with Crosby. He was nothing special in this early on in his career and now, the last two seasons, he’s posted the highest SAF/100 Shifts with 1+ SAF (I’m trying to use descriptive names but you can see the appeal of just calling something like this by the name of the creator) of any NHL F. People talk about big, heavy teams and big heavy players and those kinds of players are the ones I’d expect to score well in this metric, which is really an indirect way of asking how well you control the game offensively once you get established. Haunches aside, Crosby’s neither particularly big nor particularly heavy but the Pens are the best team in the NHL at generating multiple shot attempts on a shift when he’s on the ice. Different ways to skin a cat.

    What about defencemen?

    Screen Shot 2013-09-16 at 12.30.01 AM

    You’ll notice, if you contrast this with the forwards, that each percentile for the defencemen tends to be a few shots higher. My suspicion is that this number tends to be controlled more by the forwards who are on the ice and that, as defencemen will play with all four lines, they get a dose of the ability to generate multiple SAF after one SAF of each forward line.

    I suspect that fourth lines, in particular, lack the ability to generate multiple SAF after getting one, in conjunction with risk averse coaching staffs who don’t want them taking risks. I’ve put the list of forwards with fewer than 130 SAF/100 Shifts with 1+ SAF on Google Docs to illustrate what I’m talking about – lots of fourth line types, with the odd curiosity, like “What the hell was going on with the 2007-08 St. Louis Blues?” and “Why did Lennart Petrell get another contract in Edmonton?”

    I’m going to talk more about these metrics and do the SAA stuff and introduce the third metric I’m looking at as the week goes on. I think it’s important to touch briefly on why doing this sort of stuff matters to people who want to know more about hockey though. The better the information that we’re generating about what goes on on the ice, the better the answers we can come up with as why things change or where problems lie.

    Part of that will involve pushing past Corsi%, to try and figure out what makes it tick and whether some parts are more within the control of certain individuals than others. Eric Tulsky’s done some great stuff in this area, looking at zone entries. A fellow by the name of Pierce Cunneen is co-ordinating a project to track zone exits. The more people doing stuff like this, the closer we’ll get to figuring out things like the Ben Eager Conundrum (or the Patrick O’Sullivan Conundrum or the Eric Belanger Conundrum).

    A lot of this stuff, I think it’s hard to catch with the eye. Can you catch, with your eyes, that the Pens are suddenly generating an extra 20 shot attempts with Sidney Crosby on the ice on every 100 shifts with shot attempt? I’m not sure I could although I’ll bet that multi-SAF shifts with Sid on the ice will catch my eye this year. As you can see by the breakdown of his 2010-11 vs. 2011-12 seasons, the difference is pretty dramatic but I don’t recall anyone saying along the lines of “Gee, all of the sudden the Pens are way more likely to generate multiple SAF on a shift with Crosby.” It might be too small to catch just by watching but it’s real and it’s affecting his results.

    Screen Shot 2013-09-16 at 1.12.58 AM

    The flipside’s true with Ovechkin. I’ve heard all sorts of theorizing about why he hasn’t been the same Ovechkin of old but not a lot of breaking down of how his results have changed or theorizing based on anything beyond the fact that he isn’t producing results like old Ovechkin anymore.

    Screen Shot 2013-09-16 at 1.13.11 AM

    I haven’t heard anyone talking the shape of the change of Ovechkin’s performance. Even this year, when he won the Hart Trophy, he’s still a shadow of what he was at 5v5 once. I’ve heard tons about him changing wings (more than even some members of the PHWA) but it seems to me that a big shift in what Ovechkin does and how he does it has been missed. It generates all sorts of questions like “Why?” and “Can it be fixed?”

    If you were building profiles of your players with this degree of detail – and for all I know the Caps are – it’d give you something to work with your hockey technical people to try and isolate what’s changed and why. If you’re in the media and you’re responsible for covering a hockey team, understanding stuff like this so that you can ask intelligent questions seems a better use of your time to me than being the 30th person to tweet out line combos at practice. In training camp. When some of the names are guys who have a hard time getting into junior games.

    My point, I guess, is that while there’s a pretty strong case for doing this sort of stuff because it lets us understand the game better, it also makes for better stories about the game. I find this Ovechkin/Crosby business pretty fascinating – Washington’s been a revolving door of coaches lately but if I were doing an in-depth piece on Ovechkin – and it seems like lots of people do – I might like to track down one of Boudreau’s assistants and ask about this. Or talk to Dale Hunter. Or Adam Oates. We’re getting to the point now with the numbers that we’re starting to be able to paint pretty good pictures of what happened on the ice and ask more pointed questions about why. Those questions don’t all have to be directed at the data.

    Email Tyler Dellow at mc79hockey@gmail.com

    About Tyler Dellow

    17 Responses to New Metrics I

    1. rw9704039
      September 16, 2013 at

      “I’m trying to use descriptive names but you can see the appeal of just calling something like this by the name of the creator”

      Fine, we’ll name it after you. But for your hubris, it shall be known as DELLOWS.

    2. September 16, 2013 at

      Care to define “stuff after faceoffs” in more detail?

    3. Bruce Peter
      September 16, 2013 at

      I’d be curious to see Ovechkin and Green’s numbers on the ice together. Mike Green’s injury issues kicked in at the same time as Ovechkin’s numbers dropped. Similarly, Green’s production dropped off the map in 2010-11 in what were essentially three half-seasons of play for him. As we know, Green is an excellent offensive defenseman who was a primary puckmover for the Capitals at even strength, and liked to join the rush and create a numerical advantage in the offensive zone at 5 on 5.

      As for Crosby, tracking a guy like Letang might prove worthwhile, but I actually have eyeballed that his shifts are much more busy/dominant post-WC injury than pre. I didn’t have the data for it but I remember when he came back I was stunned by how dominant he was. It wasn’t THAT long out of the game, surely I would’ve remembered seeing him be that complete in his dominance. I’m not sure what it is, but it was definitely something I sensed.

      • Ken
        September 16, 2013 at

        Even past Green/Ovechkin, it would be interesting to see the Capitals as a whole (or as a whole, minus Ovechkin) over the same time period. When Boudreau came in, he installed a pretty open system that prioritized getting the puck up the ice quickly and pressing once it was there. 2010-11 was when Boudreau decided (was pressured?) to change to a more defensively-oriented system, and aligns with the beginning of Ovechkin’s demise. Boudreau’s gone near the start of 11-12 and he never bounces back.

        Could be concidence (especially since using FF as a proxy, no Ducks seem to have gained a massive Boudreau boost), but I’d be curious to see how it pans out.

    4. September 16, 2013 at

      Have you given any thought to breaking Corsi down further and identify who is responsible for each action offensively and defensively? Or is this already being done? The play by play data on NHL site shows that info so…

      Also, is it possible to start identifying who touched the puck prior to the player responsible for the event and who is touching the puck immediately after the event? The information that would provide could really take Corsi to the next level in terms of denting the ‘effect of line mates and opposition’ argument. (I think)

      • Jeremy Wright
        September 16, 2013 at

        Upon another read, you kind of already answered my question.

    5. Woodguy
      September 16, 2013 at

      Have you read what Mike is looking at in terms of the anatomy of a possession?

      http://www.boysonthebus.com/2013/09/13/an-exploration-into-how-possessions-begin-die/

      Interesting stuff there too.

      Are you able to estimate the shifts with no SAF or SAA?

      In terms of using that to evaluate players, I think its important as well as the ability to create (or stop) a SA has to be important as well.

      The key to getting a usable metric to evaluate a player is “who generates what SAF against X competitor levels”

      In other words, a player who produces a positive SA ration against the best is obviously more valuable than a player who produces the same ratio against not-the-best.

      A good example is Getzlaf and Perry in ANA.

      They produced very well when a checking line went against the opposition’s best, but in 11/12 there was a change in strategy, and Getzlaf/Perry was sent out power vs. power and their productivity dropped.

      The conundrum being “how do you delineate quality of competition?”

      BTN’s metric is ok, but it lacking.

      Some have tried to do it via TOI (I think you did this) and the results pass they eye test, but none of them quite feel good enough yet.

      Then we have team mate effects…….

      I like where you and Mike or going….into the cell and seeing what the organelles are.

      We still need a better way to judge the organelles.

    6. September 16, 2013 at

      Looking at those Ovechkin numbers, I thought of the quote in Friedman’s last 30 Thoughts, about what he expects to do differently this year: “Go to the net at a different level, I hope.”

      Maybe it’s a stretch to connect these dots, but it’s not hard to imagine Oates looking at similar stats and concluding that Ovechkin is still good at generating the first shot, but that his ability to create subsequent shots is declining. “Going to the net” is hockey cliche, but could be one prescription for addressing the decline.

    7. bob
      September 16, 2013 at

      Just because I’m a jerk: why is it ‘Shots Attempted For/Against’ instead of just ‘Shots For/Against’? Do your data discriminate between shots successfully attempted and shots unsuccessfully attempted?

      • Ralph
        September 16, 2013 at

        Uh, isn’t it that shots implies shots on goal, but he’s using Corsi events for here?

      • Tyler Dellow
        September 17, 2013 at

        Ralph has it right.

        • bob
          September 19, 2013 at

          Yeah, but it is redundant to say “attempted shots” instead of just “shots”.

          I mean, it makes sense in basketball to say “attempted field goals” because a field goal is what you are actually attempting.

          But it makes no sense to say “attempted SHOTS” for hockey because that isn’t what you are attempting. “Attempted goals”, perhaps, but really you are just tracking the number of shots taken, not the number of shots attempted. Or are we really tracking shots that were successfully executed vs total whiffs where the player falls down on the ice instead of actually shooting.

    8. Johnny
      September 17, 2013 at

      Just a quick reference to an older post at a different site — http://vhockey.blogspot.ca/2008/08/zone-time.html. Do you know what happened to the “zone time” statistic?

      Would it be useful to have zone times for each team and then see how each correlates with Corsi, Fenwick and your new open Corsi?

      Is the goal to try to measure team performance, individual performance, or both?

    9. September 18, 2013 at

      Interesting idea; will be interested to see how you develop this.

      As far as the Crosby and Ovechkin stuff, there’s a lot of context for these numbers that’s worth noting. Re: Ovechkin, the Caps were an awful possession team under Hunter, and aren’t much better under Oates, and 2010-11 was the year of Boudreau’s infamous switch to a slower style, so it’s not surprising that Ovi’s numbers have trended this way, considering what a high-Fenwick, high-event team the Caps were prior to 2011. The Crosby trend is harder to explain, though it’s worth remembering that his 2011-12 and 2013 numbers only reflect 58 regular-season games.

    10. Pingback: What is 'open play' hockey? - HockeyAnalysis.com

    11. September 22, 2013 at

      Fascinating stuff.

      Being awarded possession for a foul and getting a corner kick are a lot different then winning a face off. Winning a face off is a skill and can’t be ignored. However if you can identify the players who are better at gaining possession from a faceoff and then those who are best “the rest of the time” then as a coach you should be able to deploy them better.

      What metric do you use to filter out faceoff data? 10s? 15s?

    12. jerryseinfeld
      November 30, 2013 at

      NHL shifts last 45 seconds. Whats the point in analyzing shots 60 seconds later. Thats two shifts later. What does that have to do with said player?

    Leave a Reply

    Your email address will not be published. Required fields are marked *