• (Pyrrhic) Faceoff Wins In Edmonton

    by  • April 27, 2014 • Hockey • 28 Comments

    I’ve been rolling around this problem with the offensive zone faceoff losses and the 2013-14 Oilers in my brain for the past few days, trying to work it out. The problem is probably best summed up in a graph from this post:

    In a nutshell, the problem is “Why did they become terrible following OZ faceoff losses after December 7?” The change is so stark that there has to be a reason for it. I had a sneaking suspicion that it might have something to do with the Oilers trying to win more offensive zone faceoffs, so I decided to put together a graph with two things: their 5v5 offensive zone faceoff winning percentage over the season and their Corsi% following OZ faceoff losses.

    Huh. So the OZ- Corsi% went down as the OZ faceoff winning percentage went up. Is it possible that those two things are related?

    The way in which I’ve been looking at and talking about post-OZ faceoffs has been a bit artificial. I’ve been suggesting that tactics play a big role in the outcome when you lose a draw. The artificial part is that when a coach is choosing the plays and tactics that his players are going to employ, he doesn’t know whether the faceoff will be won or lost.

    It’s not unlike blackjack, in the coach wants to make the decision that has the best expected value in the face of unknowns. In blackjack, the player knows what his cards are but he doesn’t know what the dealer has or what the next card out of the shoe will be. He has to make his decisions with an eye towards maximizing his value based on the probability of various unknown outcomes.

    A hockey coach needs to maximize his value by choosing the tactics that maximize the expected goal difference. Goals are rare in hockey, so we use Corsi% and, for the time being, trust that the goals for/against will take of themselves. There’s basically a two step process here that determines the value that you’re getting following faceoffs. The first, obviously, is whether you win the bloody thing. Once you’ve won or lost the faceoff, it’s then a question of how good you are at generating Corsi% out of that situation.

    The twist, which doesn’t really exist in blackjack, is that you can alter the likelihood of winning the faceoff by the decisions that you make about how your team approaches the faceoff. What’s more, how you approach the faceoff can then impact on the Corsi% that you’re going to generate post-faceoff.

    Basically, it’s an equation for the coach to maximize:1+((Probability of losing the faceoff)*(Corsi% post-faceoff loss)) = faceoff Corsi%. There’s a weighting issue to take into account – there are going to be fewer Corsi events in period X following an OZ faceoff loss than there will be in the same period following an OZ faceoff win – but don’t worry about it. This is the technical side of what the coach has to do.

    I hate when these things get too abstract, so let’s talk about the Oilers here. I’ve been using December 7 as my sort of date when it all changed since I got talking about the Oilers and OZ faceoff losses. Essentially, my working theory has been that something changed in terms of how they approached the faceoffs, resulting in their Corsi% falling apart. I might not have the date exactly right – it could be a bit before or after that – but I think it’s somewhere around there.

    Here’s what the pre and post December 7 OZ faceoff numbers look like for the Oilers.

    Boy, there are a lot of names in there. You know what? I can summarize this. Let’s do it this way:

    That kind of highlights what happened, eh? Gagner and Nugent-Hopkins were helpless on 5v5 offensive zone faceoffs right up until December 7. After December 7 (Game 31), they were as good as the rest of the team put together was at offensive zone faceoffs was. Seems a belter of a coincidence that they both got a lot better at the same time. I’ve graphed their OZ faceoff winning percentage as the season went along, which you can see here – they improve in lockstep:

    It seems strange to me that they would both learn how to win faceoffs at exactly the same time. When I happened across this, I was reminded of a Craig MacTavish press conference at the end of last season. MacT was asked about Gagner’s trouble on the dot and he said this:

    Apparently from the analytics people, every 40 faceoffs that you lose in the defensive zone, one will end up in your net in the next 30 seconds, so it’s valuable but there are ways you can help support a centreman who has trouble winning faceoffs.

    Winning faceoffs is just part of the equation though. We care about faceoffs because possession of the puck means that you’re more likely to generate Corsi% (and, by extension, goals) than the other guy. We know that, in general, winning faceoffs means that you get a Corsi% boost overall. We would not, however, want to employ a strategy that resulted in us winning more faceoffs if it resulted in a net reduction of our Corsi%.

    Is it possible that the Oilers employed a strategy to try and prop up Gagner and RNH’s abysmal OZ faceoff numbers that a) increased their winning percentage on offensive zone faceoffs and b) resulted in an overall net reduction in the Corsi% that the Oilers were generating following offensive zone faceoffs? Amazingly, I kind of suspect that the answer to this is a) yes, it’s possible and b) yes, it happened.

    When I measure Corsi% post-faceoff, I have different lengths of time for which I measure it, depending on the outcome of the faceoff. Different outcomes and locations of the draw result in effects for different lengths of time on average. Win an offensive zone faceoff at 5v5, the effect is visible for 37 seconds in the data. For offensive zone losses, it’s visible for 21 seconds.

    If I want to look at how a team did after offensive zone faceoffs overall, I need a standardized length of time. To do this, I just counted all events within 37 seconds of an offensive zone faceoff, regardless of whether it was won or lost. The question that I am asking becomes this: did the Oilers do markedly worse following offensive zone faceoffs post December 7 despite a significant increase in their team OZ faceoff percentage because of the improvement from Gagner and RNH?

    Well that’s disheartening. Like Jesus said, what shall it profit a centre if he wins 100% of his faceoffs but his Corsi% goes down?

    I can split this out a little bit more and break it down by wins and losses. I’m going to do the losses first, because I have a bit more of a nuanced point to make about the wins.

    We already knew that the Oilers got really bad following offensive zone losses post December 7, so there’s nothing new here. Obviously, this is a big part of the reason for their overall decline in Corsi% following offensive zone faceoffs – even though they reduced their lost faceoffs from 56.1% to 50.9%, they dropped 18 points of Corsi following losses.

    What’s going on here is a little more subtle. There’s a decline in Corsi%, enough to make the Oilers a pretty ineffectual team when they won an offensive zone faceoff at 5v5 over the last 51 games of the year – that 68.6% figure stinks when you’re starting with the puck in the opposition’s end. The interesting thing is the why – the rate at which they allow shot attempts following offensive zone faceoff wins actually declined, from 0.323 shot attempts per OZ faceoff win to 0.292. The rate at which they generated shots following OZ faceoff wins plummeted, from 0.848 to 0.646.

    Two final factbombs for this piece, which is already too long but I sort of had one of those moments where you see the whole horse, so I figured I wouldn’t parcel it out and I’m prone to parentheticals which up the word count so here we are. The Oilers basically had three guys who did all of the offensive zone faceoff work: Gagner, RNH and Boyd Gordon. Gordon’s kind of the opposite of Gagner and RNH in that he basically has no offensive skill to speak of but he can win faceoffs without a lot of help.

    If you were a hockey coach who was designing something to help out your guys on offensive zone faceoffs, you wouldn’t necessarily involve Gordon – he doesn’t really need the help. Sure enough, we saw in the stats for offensive zone faceoff wins pre and post December 7 that Gordon’s number just basically stayed the same while Gagner and RNH saw dramatic improvements.

    If that’s the case and we’re suspicious that there was a change to faceoff tactics that hurt the Oilers Corsi% following offensive zone faceoffs, we wouldn’t necessarily think that it would show up with Gordon on the ice. You don’t need to fix things that aren’t broken. We might therefore suspect that Gordon’s Corsi% following offensive zone faceoffs would be pretty much unchanged pre and post December 7, while Gagner and RNH saw big declines.

    Yep. The drop for Gordon is insignificant – it’s one Corsi event going the other way. The drop for Gagner and RNH is massive. All of the decline is focused around them, which is consistent with my theory that the Oilers were doing something to try and help them out and it backfired badly.

    Final question: does any of this have any significant effect? I’m inclined to think that the answer to this is yes. The Oilers generated 1784 shot attempts this year and allowed 2323. If we assume that the Oilers could have carried on exactly as they had prior to December 7, with a lower faceoff winning percentage but a higher Corsi% off offensive zone faceoffs, that gives us a 44.8% Corsi% for the year instead of 43.4%. Some back of the envelope math tells us suggests that’s worth 4 or 5 goal difference. That’s a conservative estimate as well – more shot attempts likely has some knock-on effects.

    To sum it all up, the data shows us that the Oilers became far less successful in terms of Corsi% following 5v5 offensive zone faceoffs as the season wore on, despite winning more of those faceoffs. The improvement in winning faceoffs came from two players: RNH and Sam Gagner. The other centre, who didn’t have a big improvement in his faceoff winning percentage, saw his Corsi% following offensive zone faceoffs stay essentially the same. Gagner and RNH cratered. It seems reasonable to suspect that the Oilers changed their approach to offensive zone faceoffs in a successful effort to improve the numbers when Gagner and RNH were on the ice and that this succeeded in improving the statistic which shows up on the gamesheet each night, while undermining a far more important one that doesn’t.

    Email Tyler Dellow at tyler@mc79hockey.com
    1. Probability of winning the faceoff)*(Corsi% post-faceoff win []

    About

    28 Responses to (Pyrrhic) Faceoff Wins In Edmonton

    1. Zach
      April 27, 2014 at

      Very good and interesting piece. For Gagner and RNH to have a post-December 7 Corsi% of only 55% following OZ wins is startling.

      Perhaps the Gagner/RNH OZ strategy requires the wingers to immediately lend help to them? And because of this, although the face-off wins increase, perhaps the opposition is virtually unimpeded towards the Oilers’ defence which limits the amount of shots being generated from the increased face off wins.

      Your piece clearly shows all face-off wins are not equal. It seems that telling RNH/Ganger to ‘just do the best you can/ try be only moderately below average at face offs’ would be a better strategy than imploring this Corsi-sucking strategy to increase their win%.

      Is it possible that a DZ win (with the ‘get Gagner/RNH more wins’ strategy) is more valuable than an OZ win with the same strategy? If that were the case, then there would be times to employ said strategy, while just letting things buck in the OZ.

      • Tyler Dellow
        April 27, 2014 at

        Perhaps the Gagner/RNH OZ strategy requires the wingers to immediately lend help to them?

        Stop spoiling my next post.

        • ubermiguel
          April 27, 2014 at

          Read to the end of the article thinking “so what was the strategy change?! I must know!”. Awaiting part 2.

    2. verv
      April 27, 2014 at

      well this is pretty fucking incredible. i dont have a valuable thing to offer but good work both on figuring this out and presenting it in a good manner

    3. D green
      April 27, 2014 at

      You’re working too hard – incidentally good win for blackpool yesterday …..

      • Tyler Dellow
        April 27, 2014 at

        Pretty crucial. Ever more hopeful that they can survive, what with four points out of Brighton and Wigan.

    4. D green
      April 27, 2014 at

      You’re working too hard – incidentally good win for blackpool yesterday …..

    5. Muji
      April 27, 2014 at

      Very interesting. Intriguing findings. Malcolm Gladwell like!

      I don’t think it’s conclusive; consider this scenario:

      Gagner/RNH: Yes! I won the faceoff!
      The rest of the team: What the hell? Why’s the puck on side? He won?
      Gagner/RNH: Yea, I did. I’m getting better at these faceoffs guys. I’m only below average now. I’m no longer LOL bad.
      The rest of the team: Cool. That’s great! Hmm. Maybe we should do something with this puc..
      Opponent: Thanks for the puck guys! Why were you just standing there? Is this like the first faceoff that Gagner/RNH has won or something?
      *Opponent skates the puck out of their zone and into the Oilers zone*

    6. Woodguy
      April 27, 2014 at

      Wow.

      4-5 goals? That’s a two wins isn’t it?

      That’s worth over $1MM and you give it away for free.

      Mind you, keep publishing this type of stuff and someone will pay for it soon enough.

      Exceptional work sir.

      Also,

      As much as I like Eakins and his verbal of being a possession team, that’s two significant changes he made to their tactics that has decreased their shot attempts ratio.

      The increase in chip and chase that you showed earlier this year with Hall and now this.

      Speaking of Hall, have you figured out this one item’s impact on his shot attempt ratio?

      According to Extraskater he was on the ice for 1178 face offs, 33% of which were OZ, so 389 OZ FO.

      Would be interesting to see its impact directly on Hall.

      This is yet another example of coaching out of fear and trying to be conservative rather than coaching to be aggressive and having it have the complete opposite effect.

      This is much more subtle of a result than the chip and chase increase and I highly doubt that anyone in Oilers management or coaching staff actually knows that this result exists.

      The fact that the Oilers have a worse shot attempt ratio after a OZ FO win than many teams do after a OZ FO loss is really disheartening.

      Doing something (coaching a strategy to win more FO) because “more=good” without actually knowing if the results are improving is pretty damning stuff.

      I can’t believe that the Oilers internal tracking just shuts down when someone checks off the “face off won box”

      Actually I can believe it.

      Now I’m sad.

      • Tyler Dellow
        April 27, 2014 at

        As much as I like Eakins and his verbal of being a possession team, that’s two significant changes he made to their tactics that has decreased their shot attempts ratio.

        The increase in chip and chase that you showed earlier this year with Hall and now this.

        Just to be clear, I’m still not sure about Hall. There were pieces of the puzzle that I was missing. This is a little more clear cut and I think and there’s video to come.

        This is yet another example of coaching out of fear and trying to be conservative rather than coaching to be aggressive and having it have the complete opposite effect.

        Definitely don’t agree with this. I think this is an example of faceoff percentage being a stat that you see on the gamesheet every night and Corsi% post-OZ faceoffs not being one.

        Keep in mind, the effect adds up but we’re talking about subtle effects here. On average, it’s 1.5 fewer SAF and 0.6 more SAA per night. You think you could a) spot that and b) tie it back to something like this? I think it’s asking a lot of any coaching staff in terms of processing information and tying it back to the source to expect them to catch that.

        • Woodguy
          April 27, 2014 at

          Definitely don’t agree with this. I think this is an example of faceoff percentage being a stat that you see on the gamesheet every night and Corsi% post-OZ faceoffs not being one.

          Keep in mind, the effect adds up but we’re talking about subtle effects here. On average, it’s 1.5 fewer SAF and 0.6 more SAA per night. You think you could a) spot that and b) tie it back to something like this? I think it’s asking a lot of any coaching staff in terms of processing information and tying it back to the source to expect them to catch that.

          I think you are letting the Oiler coaching staff too easy.

          In your Boudrea threads you pointed out that all the teams in the Pacific division, who would have seen ANA often, picked up on their OZ loss strategy, and they all improved their own.

          If opposing coaches are picking up on something that a team is doing, asking the coaching staff to track and pick up on trends that their own team is creating is not being unfair, or asking too much.

          Its more like the job description.

          Implementing any strategy in any business and not tracking it to check for success is just lack of good management.

          The Oilers track scoring chances for sure, it should have shown up there fairly quickly, no?

      • Steve Burtch
        April 27, 2014 at

        4-5 goals is slightly less than one marginal win in the standings. It’s around 5.8 iirc.

        Still interesting. Assuming the weaker FO guys required help from the wings which resulted in uncovered players on the opposition thus damaging their chances of puck retrieval post FO win. Win the battle lose the war.

        I know working out the formula you described might not be overly simplistic buy any suggestions as to baseline rates could be informative. If you could implicitly differentiate and optimize that would seem quite useful.

      • Matt
        April 27, 2014 at

        I think it’s a perfectly reasonable hypothesis that if you win more faceoffs in the offensive zone, you will get more shots, give up less shots and win more games. It is just further proof that the human eye isn’t astute enough to discern the exact effect a particular strategy has… proof that analytics are NEEDED!

    7. godot10
      April 27, 2014 at

      “In the department of economy, an act, a habit, an institution, a law, gives birth not only to an effect, but to a series of effects. Of these effects, the first only is immediate; it manifests itself simultaneously with its cause – it is seen. The others unfold in succession – they are not seen: it is well for us, if they are foreseen. Between a good and a bad economist this constitutes the whole difference: the one takes account only of the visible effect; the other takes account of both the effects which are seen and those which it is necessary to foresee. Now this difference is enormous, for it almost always happens that when the immediate consequence is favourable, the ultimate consequences are fatal, and the converse. Hence it follows that the bad economist pursues a small present good, which will be followed by a great evil to come, while the true economist pursues a great good to come, at the risk of a small present evil”

      – Frederic Bastiat

      Ditto between a good coach and a bad coach this constitutes the whole difference.

      The Oilers analytics people give MacT and Eakins some good nugget on faceoffs, and Eakins and MacT take a completely wrong course of action.

      This is a replay of them destroying Hall’s Corsi, until you pointed it out. Again trying to fix what wasn’t broken, and breaking it in the process.

      Eakins is a bad coach.

    8. Dan M
      April 27, 2014 at

      Would the Oilers’ schedule have any impact on this?

      You’ve shown the Ducks to be very impressive after offensive OZ FO losses and all of the Oiler games against the Ducks came after Dec. 7.

      If you take the Blues, Kings, Sharks Blackhawks, Devils, Rangers and Bruins,all good possession teams by both Corsi and Fenwick measures plus the Ducks for their stats after OZ FO losses, the Oilers played those teams 5 times pre Dec. 7 (16.67% of their 30 games) and 21 times post Dec 7 (40.38% of their 52 games), would that not have some impact?

      I wasn’t able to watch enough games to know if the strategy changed or if it comes down to playing better opponents more often later in the season, but I’d like to see if there are set plays being used by the better teams after a faceoff loss and see if it could be copied by the Oilers so they could improve.

      • Tyler Dellow
        April 27, 2014 at

        Would the Oilers’ schedule have any impact on this?

        You’ve shown the Ducks to be very impressive after offensive OZ FO losses and all of the Oiler games against the Ducks came after Dec. 7.

        Remember – an OZ FO loss for the Ducks is a DZ FO win for the Oilers, so it’s not what I’m talking about here. As far as schedule goes, it could be that but I’m going to throw up some video soon that blows that up, I think.

    9. Tyler Dellow
      April 27, 2014 at

      The Oilers analytics people give MacT and Eakins some good nugget on faceoffs, and Eakins and MacT take a completely wrong course of action.

      We have no idea what process unfolded. You don’t need an analytics guy to tell you that RNH/Gagner stink at faceoffs. You don’t even need someone who has seen a hockey game before or can see.

      Eakins is a bad coach.

      I don’t believe this. Guy’s got an impressive track record in the AHL. If you listen to him talk, he’s obviously a bright guy.

      Look, we saw this with Renney to a degree – his rebuilding Oilers took a steep turn north in his second year. Then he was inexplicably fired. Then Krueger’s team went backwards. No summer for Ralph. Then Eakins’ team didn’t really take much of a step forward – I don’t think they stepped backwards really but it was kind of stasis.

      It wouldn’t surprise me if there’s something to be said for a coach having a summer after his first season with a terrible team to figure out what worked and what didn’t. Keep in mind – it’s easy for me to sit here after the fact with the data and blow a day doing this. I suspect I’ve got access to better data than the Oilers do, or at least I’ve manipulated it more effectively.

      I’ve said this before (and raised the ire of sacamano) but I think that coaching has to be very difficult in some ways because there’s a ton of disorganized information in a hockey game and you’ve got to suss it out, organize it and figure out cures, all while playing a game every 2.5 days, running practices, answering questions about why your team doesn’t hit more and dealing with 23 people.

      I’d still bet on them taking a step forward next year. Eventually, like one of those economists who predicts 9 of the next 6 recessions, I’ll have to be right. ;)

      • GCW
        April 28, 2014 at

        Isn’t that one of the reasons teams now have analytics departments? To help parse through all the data, and provide insights to the coach who doesn’t have time, nor likely the skills, to do it himself? If the Oilers “analytics guys” aren’t tracking stuff like this and feeding it back to the coach, shame on MacT and the analytics guys for not giving Eakins the support a modern coach requires. If they are and Eakins isn’t using it then Eakins is a bad coach.

    10. Jeremy Wright
      April 27, 2014 at

      I’m curious. If you carry through the pre December 7th face off numbers and don’t end up with the uptick in face off wins, what is the net effect on SAF. Wouldn’t there be more SAF just because those are wins, regardless of what happens to the CF%?

      • Jeremy Wright
        April 27, 2014 at

        Nevermind. I’m a little slow to catch on, you’ve already addressed that I think.

    11. David S
      April 27, 2014 at

      The Oilers were completely in the ditch by December and out of the running for a playoff spot. As fans we want to believe that our team is doing everything possible to try to win every game. But if you change your assumption to “doing everything possible to win a high draft pick”, then much of this makes sense.

      I think it may be incorrect to assume the team doesn’t have access to the same stats you’re tracking, nor have the people who can make sense of said stats. They have some pretty smart guys compiling and parsing the information constantly (doesn’t Parkatti do some work for them?). It’d be reasonable to think that if they noticed the downward shift you’re describing as a result of a failed faceoff strategy, they would have informed Eakins and he would have shifted tactics. That they didn’t is the interesting thing in all this.

    12. Steven
      April 27, 2014 at

      What would the data look like if instead of a cumulative % of corsi and FO% vs game number vs some sort of rolling average, or even just the raw numbers? Is the noise to much to plot it that way? I fear as the season goes on, to much signal is washed out by the cumulative average.

      How does RNH and Gagner’s FO% in the neutral zone and D-zone compare with the O-zone? Does it increase the same, or do these only appear to be O-zone tactics?

      Lastly, where do you find your game by game data? I have a few things I want to look into, but can never find the data I am looking for.

    13. Richard
      April 28, 2014 at

      Two thoughts: 1) Player usage 2) Opposition strategy

      1) Did Eakins use different players in the OZ based on the fact that RNH/Gagner sucked at faceoffs, or were lines and player usage consistent?

      2) Did opposing teams simply key in on the Oilers OZ FO win strategy? If it were predictable enough, could opposing teams turn a potential liability into a repeatable opportunity?

    14. Cam Charron
      April 28, 2014 at

      Great work Tyler. I’m excited for the follow-up.

      I guess my next question would be whether there was a significant change in RNH/Gagner’s FO stats in other zones, or if this was something the Oilers were doing to try and maximize possessions in the OZ.

    15. Hockeyfan
      April 28, 2014 at

      Here’s a stat: 30, 30, 29, 23, 28.
      I don’t know whether that’s a RelFenwick or a Corgi Sledgehammer or a Fibonacci Sequence. I’d have to look at the spreadsheets. Which I’ll do after I watch more actual hockey.

      • Simon Lamarche
        April 28, 2014 at

        This is parody right?

    16. Nick
      April 28, 2014 at

      I like the analysis, and while I understand the concept that Corsi should be superior to GF/GA in terms of measurement of these things (especially due to the rarity of GF/GA) I have some questions.

      1) Where and how do you get your data?
      2) GF/GA should probably be included for this data set. My suspicion is that the wingers come in to help RNH/Gagner, which leaves them badly out of position in case of a clear faceoff loss. The increase in wins is due to the Oilers winning more scrums (because the wingers are coming in to help), everything else might go clearly out of the zone (or the wingers being out of position helps the opposing team move it out easier). HOWEVER, I wonder if it isn’t still preferable to change a guy from 33% to say 50% in FW%. This would allow you to run more set plays on wins, which may result in more goals. Whether that played out or not, it’s hard to say because sample size will have been small. Basically is it preferable to have more “random” Corsi events for or fewer “set” (i.e. set plays) Corsi events for?

      • Tyler Dellow
        April 28, 2014 at

        1) Where and how do you get your data?

        I get my data using the fantastic tool put together by Red Line Station. I then do a fair amount of work to turn it into data that I can use in something like this with Excel.

        As for your second point, it’s possible but I’m a big skeptic. Everyone’s got a system that lets them beat their Corsi% until they get punched in the face.

    Leave a Reply

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