• Outshooting: Five Years On

    by  • October 29, 2013 • Hockey • 15 Comments

    Way back in May of 2008, I wrote what I assumed would be my definitive post about outshooting and its connection with wins. The money graph was this:

    The money paragraphs were probably these:

    You’ll appreciate that the key number is the number at the bottom of that chart: teams that outshoot their opposition in a given game have earned an average of 1.122 pts/g since 1987-88; teams that get outshot by their opposition have earned an average of 0.952 pts/g since 1987-88. Compare this year’s 1.15/1.07 outshoot/outshot split with the split for home road teams, which was 1.18/1.04. In last year’s NHL, you’d be about as successful picking games if you knew before hand which team was at home as you would be if you knew which team was going to outshoot the other.

    Two caveats about this post. First of all, this is as big picture as it gets. I’m not digging into the nitty gritty about game states here, which I think is incredibly important. Second, I haven’t looked at team specific cases, which will be topic of my next post on this subject. The most that I think can be drawn from this review is that, as a general principle, teams that outshoot their opposition in a given game have a better chance at success. That doesn’t establish that you can’t enjoy success despite getting habitually outshot. The real issue and (I think) the real reason why the Horcoff thread got so heated is that there’s a real question about whether or not you can build a team that gets habitually outshot, while enjoying tremendous success like the Oilers of the final quarter of 2007-08. It’s a question of relevance to persons with a vested interest in the performance of, say for example, the 2008-09 Edmonton Oilers.

    We’re five years on now and a discussion with Eric Tulsky and an email exchange with Tom Benjamin have prompted me to re-visit this. Let’s look at what’s happened in the intervening five years.

    Screen Shot 2013-10-29 at 12.51.52 AM

    Huh. Well that’s odd. OUTSHOOTING NO LONGER MATTERS. (Kidding – it does, as I’ll explain.) For a long time, outshooting the opposition was a pretty good indicator that you’d end up winning the game. For the past few real seasons, it hasn’t been. In 2007-08, teams averaged 96 points/82 games in which they outshot the opposition. By 2011-12, that fell to 91 points. In 2007-08, teams averaged 87 points/82 games in which they were outshot. By 2011-12, that had risen to 93 points.

    It gets weirder though. How do we reconcile this with something that we know to be true: teams that outshoot overall at 5v5 are more likely to be playoff teams and, the stronger they are at that, the more likely they are to proceed in the playoffs?

    Let’s drill down into two seasons for which detailed data is available: 2007-08 and 2011-12. In 2007-08, the historic pattern held and teams that outshot took more points. In 2011-12, it did not hold and teams that got outshot in a particular game took more points on average. We’ll start by looking at shots with the score tied and the record of the NHL in 2007-08 and 2011-12 when outshooting with the score tied versus being outshot.

    Screen Shot 2013-10-29 at 2.39.07 AM

    Let me explain this. What I’ve done is sort each game based on whether one team outshot the other with the score tied. So in 2007-08, the teams with the most shots with the score tied went 729-292-113 for 1571 regulation points.

    Some slight change but not very much. In 2007-08, teams averaged 114 points/82 games in which they outshot with the score tied. By 2011-12, that had fallen to 112 points. In 07-08, teams averaged 67 points/82 games in which they were outshot; in 2011-12, that rose to 72 points. The team that outshoots the opposition when the score is tied in a game takes way more points per game than the team that doesn’t. This is a kind of a critical point, because it’s the crux of the argument that outshooting matters.

    One of the ideas I had was that the volume of shots with the score tied had changed. For example, maybe the 2007-08 Red Wings ran up bigger advantages in the shot count at score tied than the 2011-12 Red Wings. The problem with this thinking, as you’ve likely sussed out, is that assuming goals occur randomly, the 07-08 Red Wings would likely have just played less time with the score tied. If you score on one of every ten shots, it doesn’t really matter how fast you get to ten shots. Anyway, the data doesn’t really bear out this idea. In 2007-08, 37.1% of shots taken were taken with the score tied. In 2011-12, 36.9% of shots were taken with the score tied. This isn’t it.

    Another idea – maybe the team outshoots when the score is tied in 2011-12 enjoys less of an advantage there than they did on average in 2007-08? We’ve heard lots about NHL parity; maybe the NHL’s parity means that that margin has narrowed?

    Screen Shot 2013-10-29 at 3.14.35 AM

    Hmm. Those look pretty much bang on. In 2007-08, the teams that took more shots with the score tied in a game had a total of 61.3% of the score tied shots. In 2011-12 they had a total of 60.5% of the score tied shots. A little bit of change but not very much.

    These posts have a tendency to turn into a wall of numbers, which makes them difficult to digest. As a result, I’m going to leave this one here. The point to take from this post is that outshooting with the score tied is a hugely valuable thing and that even though we saw teams do better when being outshot than outshooting in 2011-12, there’s been a negligible change to the results when the score is tied.

    Next, we’ll see if we can’t identify what’s happened that resulted in teams being more likely to win when being outshot.

    Email Tyler Dellow at tyler@mc79hockey.com


    15 Responses to Outshooting: Five Years On

    1. Jeremy Wright
      October 29, 2013 at

      Next, we’ll see if we can’t identify what’s happened that resulted in teams being more likely to win when being outshot.

      My initial guess is that it has something to do with the shootout.

    2. Ryan
      October 29, 2013 at

      Isn’t one of the biggest things here simply that the difference between, say, the Oilers getting outshot 48-18 by the Kings and outshooting Phoenix 30-28 is much greater than comparing them 1 to 1? Which is why, although the difference using this 1 win = 1 win view is negligible, the overall differential shows us what we expect to see – as shown in this chart which I think you initially showed me:


      Clearly score effects have a huge impact too, which I’d expect, but I think this is a main flaw in the way Tom was looking at the data.

    3. Tom Benjamin
      October 29, 2013 at

      Thanks Tyler.

      I do have some questions about the 5X5 shots when the score is tied, but I’ll leave those until I see where you are going with this.

      A methodology question though – why didn’t you choose to compare 2009-10 to 2007-08? The data might be more revealing because the results flip completely. In 2007-08, the outshooting team averages 1.17 PPG and the outshot team gets 1.06 PPG. In 2009-10 the numbers are exactly reversed. If there is an obvious reason for the change, isn’t it more likely to show up in the most extreme data?

    4. Tyler Dellow
      October 29, 2013 at

      Not trying to salt the results or anything – just picked 2011-12 because it was the most recent full year. Didn’t realize till after I’d done the work that it was the weaker of the three years with flipped results

    5. Saj
      October 29, 2013 at

      Great stuff, as usual.

      My guess is that score effects are playing a role, in that teams that are losing end up getting more shots, which doesn’t change the game outcome but changes who had the most shots.

      My question would be why that would have changed though in the last five years, if it indeed has. Are leading teams playing even more cautiously than before, resulting in more shots against? Are there more goals now, so that a bigger percentage of a 60-minute game is spent with a team losing (and therefore taking more shots)?

      I look forward to the next post.

    6. palefire
      October 30, 2013 at

      There are some weird effects that I’d be worried about when you restrict to score-tied shots, that make me suspect you haven’t done the right calculation here. This is sort of an elaboration of what Ryan says above. For instance there should be a bunch of games in your sample where one team outshoot another with the score tied exactly 1-0, representing games where a team scores on the first shot of the game and then their opponent trails the whole game. Obviously in that situation the team with the one shot gets 2 points, so it juices the numbers a bit for the ‘outshooting’ team.

      That’s an extreme example, of course, but it generalizes. The point is just that if the game doesn’t end in a shootout, then whoever has the last shot with the score tied wins the game. If team 1 has A shots with the score tied, and team 2 has B such shots, and if those shots are coming in a random order, you’d expect that team 1 had the last score-tied shot A/(A+B) of the time, and team 2 had it B/(A+B) of the time. If you throw out the games that went to a shootout and you average max(A,B)/(A+B) over all the remaining games, how well does that match the winning pct of the outshooting team?

      • palefire
        October 30, 2013 at

        (Of course when you restrict further to 5v5 score tied shots, it’s not literally true that the team with the last such shot wins, since the winning goal may not be scored at 5v5. But the basic point stands, the outshooting team has a built-in advantage in this calculation that may not really be measuring what you want.)

      • Tyler Dellow
        October 30, 2013 at

        There were 24 of those games in 2007-08 and 31 in 2011-12. I’m not entirely sure how that pollutes the analysis though – maybe those teams giving up the first shot should have not given it up and instead generated it themselves?

      • Tyler Dellow
        October 30, 2013 at

        That sounded glibber than I intended. But you take my point. A team that’s good at generating shots is more likely to generate the first shot in a game.

        • palefire
          October 30, 2013 at

          My point is that your calculation has a built-in, structural reason that “team that outshoots” should get more points than “team that’s outshot”. So even if one doesn’t believe that getting shots (or shot attempts) is paramount, one should still believe that “team that outshoots” is going to come out ahead in this calculation. For example, those 31 games where the shot count ended 1-0 already contribute around 0.04 ppg to the “outshoots” side, and were basically guaranteed to do so.

          I guess another way to put it is, I think you’re doing this calculation in the service of trying to produce evidence for a hypothesis about hockey. So, in isolation, what does this calculation tell us about hockey? And I think the answer is ‘very little’ because, for those structural reasons, this calculation was guaranteed to favor the “outshoots” team. Comparing the ppg for “outshoots” to 1/2 the average points awarded per game isn’t the right control.

          As I suggested before, I think an actually meaningful comparison would be: throw away all the OT games, and compare ppg for “outshoots” to the average of 2*A/(A+B) over all the games, where A is the number of shots for “outshoots” in each game and B is the number of shots for “outshot”. [That is, calculate 2*A/(A+B) for each game separately, and average those. For instance the games where A=1 and B=0 each contribute 2 points to that average, exactly like you expect them to.] If those match up pretty well, then you have good evidence that your propensity to win correlates with your propensity to get 5v5 score tied shots — your likelihood of having the “winning shot” matches up with your likelihood of having scored any particular shot. And if they don’t match, then you have a surprising finding that needs to be understood.

    7. TMS71
      October 30, 2013 at

      Always with the cliffhangers. I can’t wait, I can’t wait, I can’t wait!

    8. TMS71
      October 30, 2013 at

      Won’t it have to turn out that score effects have increased? Trailing teams are outshooting by more than they used to. But why? More defensive shell defenses? And does it work? Are come from behind regulation ties or wins more common?

    9. Phil Birnbaum
      October 30, 2013 at

      Good stuff! Just curious, was 2010-11 not available?

    10. October 31, 2013 at

      Interestingly, I published this analysis the same day as yours. I included all the data from the past 6 seasons, but my conclusion is the same as yours: score-tied shot differential relates strongly to season point totals. When I pulled the data together I also couldn’t help noticing that the tied shot differential (measured at season’s end) correlates very, very closely with Fenwick Close.

    11. Pingback: Islanders News: Battle of New York reactions | All New York Islanders

    Leave a Reply

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