• After the DZ Draw Is Won, 2013-14

    by  • May 3, 2014 • Hockey • 7 Comments

    I’m going to try and throw up some of the situational micro-data that I’ve generated for this year over the next couple of weeks. I posted a little table on Twitter today that got a little buzz and it’s a pain to talk about it on Twitter, so I decided to post it here.

    This table shows the rate at which shot attempts occurred per defensive zone faceoff win last season. The window I’m looking at is 21 seconds long. So, for every defensive zone faceoff that Winnipeg wins, the two teams combine for 0.410 shot attempts over the next 21 seconds. 29 of the 30 teams are pretty close to each other. The puck has to be somewhere, of course, and if it’s somewhere dangerous, people will be attempting to get a shot at the goalie.

    Then there’s New Jersey. Chicago, who sit 29th in terms of shot attempt volume following a defensive zone faceoff win are closer to first place Winnipeg than they are to 30th place New Jersey. I’ve written about this before but New Jersey is playing a completely different kind of hockey than the rest of the NHL. It’s almost certainly incredibly boring hockey. (One of the lousy side effects of a league with so much revenue sharing is that it almost encourages boring hockey by curbing the price that teams pay for playing it.)

    Just as a little sanity check, I took a look at the road data for New Jersey. It’s not just a scorer issue – the Devils only see 0.229 shot attempts per defensive zone faceoff win when they’re at home. It’s just something about Devils hockey. Some day soon, someone will figure out exactly why this is, how the Devils are different and show us in a very cool post. How do they do it? Is this something other teams should be doing? These are all excellent questions that would be great to have answers for and, to be honest, they strike me as the sorts of things teams should be interested in too.

    What about the specific results of defensive zone faceoff wins this year? In other words, the Corsi% and GF% post DZ faceoff win? I’ve put that in the table on the left.

    So you can see that there’s a Corsi% range from about 52.8% to 34.4% post DZ faceoff win this year. The league average was a 42.9% CorsI% and a 40.8 GF%. LA being at the top shouldn’t be that surprising – the Kings are Corsi kings in basically every 5v5 situation. Other things that jump out at me – I’m kind of surprised at where CHI and STL rank – they were league averageish last year, so this is a big drop off. I’m not really surprised that Edmonton and Toronto did terribly.

    A word needs to be said about the sample sizes we’re talking about here and the utility of Corsi% in this scenario. The sample sizes are obviously tiny. For the average team, there are 221 Corsi events involved here. It’s a very specific circumstance: we’re basically measuring what happens in the next 21 seconds when a team starts with the puck in its own end of the ice but there’s going to be some noise.

    That said, we have a reasonably good idea after half a season what the next half a season with this stat will look like. I put together a scatterplot to illustrate this.

    The correlation from the first half to the second was 0.46, which isn’t bad given how small the samples are. What’s more, when you look at some of the teams that had big swings, you can kind of guess what happened. Calgary went from 34.4% in the first half (well below average) to 45.9% in the second. Do you think that maybe the table below tells you why? Calgary’s second half had a lot more Giordano and a lot less of their bad defencemen and when you take that into account, suddenly it looks a lot less random.

    I suspect that part of the reason the correlation’s low (although, like I say, it’s not terrible) has to do with tactics changing and injured players exiting/returning to the lineup. It’s worth keeping in mind if you’re working with this stuff though. It’s a record of what happened; your job as an analyst is to figure out the context, being players coming in/out of the lineup and the tactics so as to understand what it’s telling you.

    This wouldn’t be a post about some statistic for me without a brief diversion into Oilers related weirdness. Here’s Edmonton’s Corsi% post-DZ FOW as the year progresses:

    Early season sample size spikes aside, it peaks around Game 40, then, well:

    That’s a horrible number post-game 40. File this away with the OZ faceoff losses where there’s a mid-season change that didn’t seem to have a positive impact on things.

    One last point that I think I should make about these kinds of micro-stat. I’m very comfortable using Corsi% in the big picture, believing that a team needs to post a big Corsi% to be a Cup contender, because I know that Corsi and scoring chances are pretty tightly correlated. When you start slicing things as small as I am here, that may not be true. Personally, I suspect that it still is true, but my certainty of that is somewhat lower than it is for overall Corsi%. This hasn’t been validated the same way it has with overall Corsi%.

    A graph might help express the issue here. What I’ve done is sorted the thirty teams last year by their DZ FOW Corsi% and then done line graphs of that and the GF% that each team had. That produces this:

    You can see that the Corsi% forms a nice line downwards, which makes sense given that the list is sorted by Corsi% but that the GF% numbers are all over the map. This isn’t that surprising – there are very few goals scored in a hockey game and I’ve sliced this data pretty fine. If you have a sense that the GF% line seems to be trending down, you’re right. If I split it into three groups: best, average, worst, this is what I get in terms of Corsi% and GF%.

    There’s our familiar Corsi% -> GF% relationship showing up again. I wanted to try this with a bit bigger sample so I took the six full seasons since 2007-08 (discarding 2012-13), sorted every team post-DZ FOW by Corsi% and then created ten “teams” by aggregating the data – Team One had the best 18 DZ FOW Corsi%, Team Two had the second best 18 DZ FOW Corsi% etc. That produces this:

    You can see that it’s pretty clear but there’s still some spikes. That said, we’re still not talking about a ton of shot attempts. Team 8, with their killer 51.9% GF% despite a 41.6% Corsi% had a total of 3647 shot attempts. That’s maybe 45 games worth or so from an overall Corsi% perspective. The most likely reason for the big GF% spike is just randomness but it is there and worth acknowledging, given that we haven’t tied scoring chances to Corsi%.

    All of that said, I’m inclined to believe that Corsi% does tell the story here. I think that this kind of underlines a point I’ve made before though – if teams are too focused on goals in doing their analysis, they’re going to make a lot of mistakes. In the short term – and we’re talking entire seasons as a short term – there simply aren’t enough goals to point you in the right direction. Even if you make the right decisions on everything, you’re going to have aspects of your systems that look like they aren’t working if you’re focused on goals, just due to chance.

    Focus on getting your team spins of the wheel and denying them to the other team. The better the job you do at that, the more likely that it is that the goals take of themselves.

    Email Tyler Dellow at tyler@mc79hockey.com


    7 Responses to After the DZ Draw Is Won, 2013-14

    1. May 3, 2014 at

      I think you’re on to something. While the Devils Corsi and Fenwick percentages have been quite favorable in recent seasons, they’ve also been doing it with a lot of relatively low event play at 5-on-5. However, I’m not sure what the next step would be since I’m unfamiliar with looking at specific in-game situations that don’t necessarily result in an event. What would be the best way to start looking at DZ faceoffs? Is formation or location of the puck crucial?

      • Tyler Dellow
        May 3, 2014 at

        If it was me and I was interested in this, I’d want to break down two teams – say the Devils and a league average team, like Philly.

        I’d then track all of the exits/entries and timestamp them. Once I had that, I’d be looking for differences between the Flyers/Devils, in terms of time spent in each zone and shot attempts generated per entry type. That would narrow things signficantly.

        • DK
          May 3, 2014 at

          Another note: the Devils have been notoriously bad at faceoffs as a whole since the 2005 lockout. Perhaps they are being coached to win faceoffs in a specific way or have their wingers take a different position off the draw?

        • May 3, 2014 at

          The Devils and Flyers did play each other a number of times (including one really depressing game early in the season), so that would work. I can’t promise how far I’ll get, but it’s a start. Thanks.

    2. May 5, 2014 at

      Random question – why the 21 second window?

    3. Zack
      May 5, 2014 at

      Is the data easily parsed by centre? The only thing I can think of as a Blackhawks fan is that Quennville is a little unusual in that, this year, he was mostly matching zones rather than lines. The 4th line (mostly Bollig – Kruger – Smith) took almost all the d-zone draws when they were available. He wasn’t nearly as strict last year, probably in part because Kruger went from a 46% to a 56% faceoff man over the summer. This also pushed every other full-timer on the team to over 60% ozone starts.

      • Zack
        May 5, 2014 at

        Kruger’s share of d-zone draws only went for 42% to 44%, so that doesn’t explain it, unless last season he was deployed more often as a second centre or something. Bollig’s share went from 14 to 38%, but I can’t believe he’s that much of a drag, even as an up-jumped facepuncher.

    Leave a Reply

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