• Big Oilers Data VI: Sam Gagner

    by Tyler Dellow • May 7, 2013 • Hockey • 10 Comments

    This is part of a series looking for reasons for the Oilers Corsi% collapse in 2012-13 by examining things on a shift-by-shift basis. Part 1 can be found here. Part 2 can be found here. Part 3 can be found here. Part 4 can be found here. Part 5 can be found here.

    There’s been a lot of tough slogging through fairly abstract stuff to date in this series. This is a new way of looking at things, there aren’t a lot of well established benchmarks and it’s kind of been a firehose of data, in terms of looking at forwards or defencemen as groups. I want to start sort of narrowing the focus on this and start talking about players more specifically and in some more depth. As I alluded to in Big Oilers Data V, I’m going to start by looking at Sam Gagner.

    Gagner was one of the big storylines for the Oilers in terms of their collapse in Corsi% this year. The arc of his career has been pretty flat from his rookie season though 2011-12 in terms of Corsi%: 46.1%, 49.4%, 48.2%, 48.1% and 48.9%. Then, this year, 43.1%. Trainwreck. (An aside, but my recollection is that he was a horror show for the first half of his rookie year and then improved.)

    Here’s a summary/refresher of what happened on Sam Gagner’s shifts in 2011-12 v. 2012-13.

    Terminology refresher:

    1+SAA%: Percentage of all shifts with at least one Shot Attempt Against
    Multi-SAA%: Percentage of shifts with at least one Shot Attempt Against that saw multiple Shot Attempts Against
    1+SAF%: Percentage of all shifts with at least one Shot Attempt For
    Multi-SAF%: Percentage of shifts with at least one Shot Attempt For that saw multiple Shot Attempts For

    So, in Gagner’s case, we know that there was a big uptick in his 1+SAA% and a downtick in his Multi-SAF%, with smaller negative changes in his Multi-SAA% and 1+SAF%. I pointed out in the previous post in this series that Gagner’s usage, at least in terms of the faceoffs that he took, also changed this year. Again, a summary of that data:

    So, as discussed, there are slightly fewer neutral zone faceoffs and slightly more defensive zone faceoffs in Gagner’s mix. There’s also been a significant collapse in his neutral and defensive zone faceoff winning percentages. The question that this gives rise to is an important one: to what extent is the different context in which Gagner played this year (and, to be fair, for which he was partly responsible with the decline in his faceoff winning percentage) responsible for the decline in his Corsi%?

    It’s not an entirely straightforward question to answer but it’s one that I think is kind of important. Part of what I’m trying to do with this project is isolate, as much as possible, the areas in which things went wrong for the Oilers this year. A hockey game is a big thing, that ranges over a lot of space and a season is 82 (or, every so often, 48) times bigger than that. There’s so much happening that, frankly, I kind of think it’s hard for the human brain to isolate things through watching alone. I read a quote from Juergen Klopp, Borussia Dortmund’s manager, recently about how he learns from and breaks down soccer games that I thought was interesting:

    For me the best analysis is to watch the game again. I know it’s very old fashioned. Tape in, forward and rewind, forward and rewind…a thousand times…spent 5 or 6 hours on a 90 minute game. I haven’t been able to do it any faster. But to be clear: this was my education, no book or seminars or anything from renowned trainers. 10 games a week and I usually started before breakfast.

    For what it’s worth, in addition to being one of the coolest looking men in Germany (see photo below), I think Klopp’s a pretty bright guy; I don’t think he’s slow or anything. If he says that that’s how long it takes to do it properly, I believe him.

    A soccer game lasts longer than a hockey game but when you factor in the amount of time that the ball’s in play, they’re pretty similar. Hockey also features line changes and is a much faster game; there’s more stuff happening per minute that you’d have to track if you wanted to break it down properly. There’s also more games per week during the season – soccer teams basically never play more often than twice weekly. All of which is to say that I tend towards the opinion that finding ways to use data to provide you with insights into what’s happening is even more critical in hockey than it is in a sport like soccer because of the sheer size of the firehose of information that a single hockey game, let alone 82 of them, produces.

    Given the change in Gagner’s usage and, particularly, in his faceoff winning percentage, I thought it would be worth exploring what Gagner did this year and last year in situations that I define based on whether or not there was a faceoff during his shift and, if so, where on the ice it took place. Apples to apples. We’ll start by looking at shifts in 2011-12 and 2012-13 during which there was no faceoff when he was on the ice.

    In 2011-12, 662 of Gagner’s 1436 shifts (46.1%) didn’t feature a faceoff. This year, it was 412/919 (44.8%), a pleasingly similar figure. When you drill down far enough, hockey’s hockey and doesn’t change a ton from one season to the next. The game is what it is and faceoffs happen when they happen and don’t when they don’t. How many SAA did the Oilers give up on Gagner’s shifts in 2011-12 and 2012-13?

    Well. That looks awfully similar, year over year. The percentage of shifts with zero SAA and one SAA are about as close to identical as you can get. He had slightly fewer shifts without faceoffs this year on which there were multiple SAA (15% to 15.7% in 2011-12) but, curiously, since you’d figure that the same skill set in preventing one SAA from becoming two SAA is involved, slightly more shifts on which there were more than two SAA (5.6% to 3.8% in 2011-12). I’m not entirely sure if I should take anything from that – as I said, the skill involved seems to me to be the same as preventing one SAA shifts from turning into multi-SAA shifts and he was virtually unchanged in that respect.

    What about SAF?

    Again, it looks awfully similar, although there’s a more pronounced inability to create multi-SAF shifts that’s consistent going from one SAF to two, from two to three, etc. I’m more comfortable thinking that there was something real happening that prevented multi-SAF shifts than I am thinking that there was something going on that resulted in more shifts with 3+ SAA because of the consistency in the drop from shifts with 1 SAF to 2 SAF, something that doesn’t show up in the data for SAA.

    I’ve summarized this data and expressed it Corsi% style in the table to the left. Recall that I mentioned above that Gagner’s Corsi% fell from 48.9% to 43.1% this year. You can see that his decline in shifts without a faceoff was a lot tighter than his overall decline. If he’d had eleven more Corsi+ events and eleven fewer Corsi- events, it would have been exactly the same. Fun math trivia: if Gagner’s had the same multi-SAF shift ratios as he did last year, he would have had 11.15 more Corsi+ events this year. This, presumably, would have reduced the Corsi- events that he experienced, as you can’t give up SAA while you’re creating SAF; the best defence is a good offence and all that.

    What do I conclude from this? Outside of this issue with creating multi-SAF shifts, I don’t think that the collapse in Gagner’s Corsi% this year is really attributable to what was happening during shifts that didn’t involve a faceoff. The numbers are incredibly similar to last year for him and to the extent that they’re down, it appears to be tied up with this multi-SAF shift issue. Shifts involving faceoffs? Well, that’s a different story and one that I’ll come to in my next post.

    (A word on the data quality here: my spreadsheet captures about 99.7% of faceoffs. The effort required to make that 100% would be substantial and, while I may do it at some point, I can’t see how three faceoffs in a thousand would change the conclusions here. If anyone disagrees, feel free to let me know.)

    Email Tyler Dellow at mc79hockey@gmail.com

    About Tyler Dellow

    10 Responses to Big Oilers Data VI: Sam Gagner

    1. Lloyd B.
      May 7, 2013 at

      Tyler If the entire world operated using 99.7% of the information available I can only imagine where we would be as a society. Probably some sort of advanced civilization. If it matters, you are going down a very interesting path here. I am coming to a few thoughts as to what may be going on but need this additional information you are gathering on Gagner before I comment. As a co-worker once told me, get the facts before the facts get you! You are certainly getting the facts. Thanks for your efforts.

      • Tyler Dellow
        May 8, 2013 at

        Thanks for the kind words Lloyd – appreciate the feedback.

      • blacquejacque
        May 8, 2013 at

        If the world operated using 99.7% of the information available, we’d be lynching bankers, politicians, and the ultra-rich. :)

    2. Woodguy
      May 7, 2013 at

      *waiting expectantly for face off win/loss SAA/SAF and WOWY.*

    3. Saj
      May 8, 2013 at

      Great stuff!

      I don’t think the remaining 0.3% of faceoffs matters, but just out of curiosity why aren’t those in your sheet? Are they just randomly missing or are they special in some way?

      • Tyler Dellow
        May 8, 2013 at

        I kind of screwed up converting everything. I usually work with minutes and tenths/hundredths of a minute. For work like this, it’s better to work with seconds, which I switch to part way through. I had a little bit of trouble with converting back to seconds – a couple of cases where I was getting, say, a shift that started at 2033.2 seconds, which is impossible given how hte NHL records their data. Rounding it works to solve that but, when I have the master sheet count the faceoffs, if there was a player on the ice at 2033 seconds for a faceoff, that won’t be captured. Very, very minor issue, I think and one that doesn’t affect the substance of what I’m doing.

    4. Bruce McCurdy
      May 8, 2013 at

      I’ve finally had a chance to catch up on this series, and I first must say it is outstanding work. My hat is off to you, Tyler, both for volume of number-crunching but also finding new ways to take a sip from that firehose.

      Further to that exchange we had on Twitter awhile back, I like that you’ve taken the trouble to record neutral zone faceoffs, whereas many ZoneStart type stats choose to ignore them. I especially like that you have further parsed shifts without any faceoff, hockey “on the fly” which is forgotten altogether in the ZoneStart type of analysis (which is nonetheless great stuff, it simply has its own limitations).

      Your previous data in Part V showed that 5v5 there are more neutral zone faceoffs than in either end zone. As a first approximation I wonder if those shifts that start with/include a neutral zone faceoff might be expected to have similar outcomes as shifts with no faceoffs at all. I also wonder whether wins and losses of neutral zone draws makes a fig of difference. So I’m hoping you’ll do what Woodguy suggests, but subdivide by faceoff wins and losses in all three zones.

      It’ll be a small sample prone to random variance for a single player like Sam Gagner, but I wonder if a big data analysis of all players in those situations might help establish a coefficient of the value of forcing a faceoff in a given zone, and the value of winning that faceoff.

      Would also be very interesting to differentiate between faceoffs that occur at beginning of shifts (true “zone starts”) and those that occur *during* shifts, with the former being an initial condition but the latter being to some an extent an outcome of a player’s own performance. If your goalie is being forced to freeze the puck a lot while you’re out there, getting “stuck” with a bunch of d-zone faceoffs is a cost of doing business. The faceoff doesn’t necessarily *precede* the shot, eh.

      The other interesting component is of course faceoffs at the end of shift. As I’m reading this they wouldn’t show up in the data (at least as presented here) cuz the player is by definition off the ice when that faceoff actually occurs. But it is an outcome of that shift, as Vic Ferrari’s pioneer work demonstrated. ZoneFinish attempts to capture this, though imperfectly in my view. Lots of possibilities for research in this whole area; kudos to you for what you’ve bitten off here.

    5. Hawerchuk
      May 11, 2013 at
    6. Pingback: Sam Gagner: The more things change… | Edmonton Journal

    Leave a Reply

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