Outshooting Wins Games

David Staples, who writes the Cult of Hockey blog for the Edmonton Journal, and who has done a lot of interesting work during his time in the blogosphere (perhaps most notably breaking the Jaromir Jagr to Edmonton talks) is a guy I generally have a lot of time for.

Today, though, I must admit that his latest item drove me a little batty.

Staples compared two items from last season: a ranking of shots plus/minus and a ranking of goals plus/minus. He focused in on how many spots certain teams were out in the rankings, saw that some teams were a fair bit out (notably Boston, Vancouver and Florida) and while admitting that it’s possible he’s missing a flaw in his study, he concluded that good goaltending can throw the two items wildly out of whack.

Things I Don’t Like About the Staples Study

I do have a few issues with the way Staples conducted his little experiment, which I’ll explain now:

1. He focused on the edges rather than the pattern. There were really two ways of presenting this data: looking at the edges or looking at the average. Staples looked at the extreme outliers rather than focusing on whether – on average – the out-shooting team generally scored more goals.

2. He didn’t reference shooting percentage and save percentage. The obvious, and incredibly easy way to quantify the impact of goaltenders/shooters would be to include references to these; I suspect Staples’ looked at them because he talks about good and weak goaltenders, but he doesn’t give us any idea of how much better than average these goaltenders were.

3. He used only a single season. This I think is self-explanatory; the NHL has data going back years and running it back as far as the lockout would have been nice.

A Different Vantage Point

With regard to point one, I went to nhl.com and retrieved the goals +/- and shots +/- data. The chart below is goals +/- per game vs. shots +/- per game. The black line indicates the trend:

In short: the data actually suggests that out-shooting does help a team out-score as well, as evidenced by the trend line.

Finally, I knew that Tyler Dellow of mc79hockey.com had done some work on this topic before and that his findings more closely match the chart I did above, and sure enough he’d done a studyon how the post-lockout Oilers had fared in goals +/- vs. shots +/-. To quote his findings:

If you treat the ES portion of a hockey game as a separate game, which I think is the only sensible way to look at whether or not outshooting your opponent matters, the Oilers, since the 2005-06 season, are 36-42-33 when outshooting their opposition at ES, 31-63-27 when getting outshot and 5-5-4 when the ES shots tied. Just to be crystal clear, this is their record in the ES portion of the game – it seems foolish to me to look at the game as a whole when talking about ES outshooting. So, for example, if Edmonton outshoots their opposition 20-15 at ES and outscores them 2-1 at ES, but give up 3 PP goals and score none, I treat that as a win for the Oilers.

Finally, with regard to point four I again turn to Dellow’s work. He did a massive big-picture post on team’s records when out-shooting or getting out-shot; all of the data from 1987-2008 is included. The full study is here; for the sake of brevity I’ll just post the records:

  • Outshooting Team: 9451 W – 7116 L – 1979 T – 612 OTL
  • Outshot Team: 7728 W – 8647 L – 1979 T – 804 OTL

That strikes me as significant.

In short, yes teams can win while being outshot – shot quality game-to-game and goaltender ability certainly do factor in, but the fact of the matter is that the majority of the time the team taking more shots is going to win the game. And for all the talk of different strategies, shooter ability, goaltender ability and the like, the most important point is simply this: not once in the past twenty years have the out-shot teams posted better records than out-shooting teams.

Comments (11)

  1. When you take a look at the teams that (for the season) are in the top 10 in shots per game, they usually remain in the top 10 for goals per game. More shots will equal more goals, usually. There are exceptions, (i.e. the New York Rangers) but they’re hardly the norm.

    The arugment against this is that crappier teams’ shots will be more from the outside and will therefore not have as many good scoring chances. I suppose that could be true, but wouldn’t a team be getting more shots BECAUSE it is a better offensive team? They would control the puck more,

    As for the outshooting/outshot records, they’re pretty much what I expected. Goaltending obviously plays a role, but when you accumulate data over 20 years, it all averages out.

    I think I just repeated exactly what Jonathan said.

    Anyway, that means that if you were a betting person, you should bet against my Colorado Avalanche, and bet on Toronto. I know you don’t want to, but both those teams will regress to the mean.

  2. I think you and Staples are both right and both wrong at the same time. Basically, while you would expect, on average, for outshooting teams to outscore, and outscoring teams to win more, it’s not the whole story.

    You almost always look to the mean, and that’s fine. I tend to look to the outliers, as Staples has done here. The problem is that you cannot tell whether or not any one team is an outlier until a significant way through the season. You cannot forecast whether shooting rates will go up or down (though you can forecast SH% and SV% rates). An argument that works today might not work tomorrow. Also, in a league where there are 30 teams, and you consider ~ 20% of the teams in the league to be anomolous/outliers/easy to ignore, your argument doesn’t have a lot of heft in the real world. A theoretical gestalt of outshooting teams making one team, completely ignoring SH%? Sure. A realistic team that has injuries, illnesses, coaching and personnel changes? Not so much.

    And the real kicker is that none of this means anything on a game-to-game basis.

    Plus, your math shows that the theoretical outshooting team wins 9% more often than the theoretical outshot team. That equates to 1 extra win in 10 games, so 8 extra wins in a season (assuming you’re not an outlier). 16 points. That’s also assuming that you *always* outshoot the other team, and the other team is *always* outshot. Let’s take an arbitrary number of 75% for a particularly good team (they outshoot 75% of the time), and 25% for a particularly bad team (who is outshot 25% of the time). That works out to an 8 point spread over a season. There was a ~25pt difference between first and 8th in the conf last year. Admitting that some teams are outliers, doesn’t that make the outshooting argument effectively moot?

  3. “and 25% for a particularly bad team (who is outshot 25% of the time). ”
    should be “and 25% for a particularly bad team (who outshoots 25% of the time). ”

    Don’t get me wrong, it says *something* but only in a very narrow view. It doesn’t mean that a team who only outshoots 25% of the time will be out of the playoffs or have an early playoff exit. It just means they’ll likely (though not definitely) have a worse record than a team who outshoots most of the time.

  4. I think you and Staples are both right and both wrong at the same time. Basically, while you would expect, on average, for outshooting teams to outscore, and outscoring teams to win more, it’s not the whole story.

    Well, no…that’s the point Jonathan is driving at. There’s no more to it than that.

    I think the problem you run into continually, Ender, is that you seem to expect single metrics to be holistic, ie; to be highly predictive of success all by themselves. We find a strong correlation to outshooting here, but you throw up your hands and declare it’s useless because, well…there are other factors that influence winning and therefore you can’t bet your mortgage on a single game based on outshooting.

    Of course, as you point out, all things aren’t always equal. But it’s a piece of the puzzle to be integrated with the other things we know about the game.

    That’s how I look at it, anyways.

    Thanks for the article Jon. I was going to write something similar but you beat me to it.

  5. @Kent

    Actually, I pretty reliably argue that single metrics *aren’t* holistic. It’s all well and good to say it’s a piece of the puzzle, but what puzzle, exactly?

    “[T]he majority of the time the team taking more shots is going to win the game.”

    This is what I argue against. Jonathan’s math (as with his SH% piece a few days back) does not support his conclusions. Maybe when it’s bridged across an entire league over multiple seasons this math points to a trend. That’s great. However, that does not mean that it will be that way for any one team in any one season. And especially when you can point to 5 or more teams that it does not follow for, you’re defining the margins to be 20% of the league.

    Whether or not that graph passes as “strong correlation” mathematically (which I don’t think it does), it still comes nowhere near passing as remotely causative.

    I think I get misrepresented as the guy who wants to tear everything down. I don’t. Hell, I posted on my site the other day modifying Willis’s argument about SH% and SV% to show what the data *actually* shows.

    I’m fine with people using math. I’m fine with people making unsubstantiated statements. When people try to use math to argue unsubstantiated statements, I say so (now and again).

    To shift your words a bit, I think the problem I run into continually is that people treat metrics as holistic whether they believe them to be or not, and some get really annoyed when I point out that 1+1=2 and not “pancreas.” Some don’t. The most vocal do.

  6. //”[T]he majority of the time the team taking more shots is going to win the game.”

    This is what I argue against. Jonathan’s math (as with his SH% piece a few days back) does not support his conclusions. Maybe when it’s bridged across an entire league over multiple seasons this math points to a trend.//

    Ender, that’s nuts. In every season in the past twenty years, teams that shoot more win more games.

    Therefore, teams that shoot more have a better chance of winning.

    That’s exactly what the math supports. Exactly. I’m not arguing that shooting more wins every team every game by itself, but it’s clearly connected.

  7. Um, I think you have a different definition than me for the word “exactly.” What the data supports is that a gestalt team of outshooters wins 9% more than a gestalt team of undershooters.

    When you jump to individual teams, things get much fuzzier, and you don’t seem to be respecting that. Just because “teams that shoot more win more games” doesn’t mean that a *team* that shoots more will win more games. It doesn’t mean that “the team taking more shots [generally] wins the game” either. It means that one abstract construct would win more games than another abstract construct. And I know how you hate looking at the outliers and would rather look at the mean, but how about those outliers? I mean, I’d wager that if you take away the top 3-5 teams in the league and the bottom 3-5 teams in the league (regular season point-wise) everything would even out. Possibly less – it’s always possible that the worst are much worse than the best are better.

    Now if that evens out (which I’m pretty sure that it will) or even if it doesn’t and my maximum 16pt spread holds, how much does this general theory mean when you jump from the construct to a real team? If the former is true, you now have a theory that only applies to the best and worst 3-5 teams in the league. If the latter is true, then outshooting might mean something, but it’s not going to mean enough to miss the playoffs, and once the playoffs hit, all stats are effectively out the window. Either way, your jump from “teams” to “team” does _not_ hold because you haven’t made *any* link from “teams” to “team.”

    Unless we have different definitions of “exactly.”

  8. Ender:

    //It doesn’t mean that “the team taking more shots [generally] wins the game” either.//

    Ender, I’m going to rephrase. In the last twenty years there have been just shy of 20,000 regular season games played. The team with more shots won 55% of those games. So, in general, the team taking more shots is the winner; if it had no impact on winning, there’s no way we wouldn’t be much closer to 50/50 here. Which means that outshooting positively effects winning, on the whole.

    That’s it. That’s all I’m saying here. Outshooting has, generally, a positive effect on winning, so for the majority of teams outshooting will gain them points. I don’t see how you can possibly deny that.

    //I’d wager that if you take away the top 3-5 teams in the league and the bottom 3-5 teams in the league (regular season point-wise) everything would even out. Possibly less – it’s always possible that the worst are much worse than the best are better.//

    Since I know you’re a stickler for having everything nailed down, do you have any math to indicate that, or is it gut feeling?

    //the latter is true, then outshooting might mean something, but it’s not going to mean enough to miss the playoffs, and once the playoffs hit, all stats are effectively out the window.//

    You’re mistaken on both points here, Ender.

    In the first case, it depends what the margin of error in the playoff chase is. With four points between 6th and 10th in the West last year, I’d say that’s manifestly untrue.

    In the second case, stats aren’t out the window in the playoffs. A statistical model may not be able to predict something as small as a seven-game series with a high degree of accuracy, but it does give us (based on the 82 preceeding games) a very good idea of the strengths and weaknesses of each team and more often than not the better team wins.

  9. Willis:

    //I don’t see how you can possibly deny that.//

    I’ve rephrased myself a number of times now, trying to show you exactly that – a gestalt team is not a real team and as such you have not made _any_ connections between this fictitious team and a real team. Your math does not support it.

    //Since I know you’re a stickler for having everything nailed down, do you have any math to indicate that, or is it gut feeling?//

    Just a guess, which is pretty obvious given that I said “Now if that evens out [...] or even if it doesn’t.” It could be wrong. Actually, let’s go check some math.

    Last season 11 teams had a better record when they were outshot. 1/3 of the league. Of those 7 were playoff teams. 1/3 of the league who excelled while being outshot and nearly half of the playoff teams were teams who were outshot. The president’s Trophy winner had a better record when they were being outshot.

    But I’m sure that you’ll argue that they outshot more often than they were outshot, and you’d be correct. However, making the connection between outshooting and wins in the case of one team doesn’t really seem to correlate strongly here. Just because the “trend” for every team who ever outshoots tossed in a big pile says something, it does not determine the success for any one team.

    //You’re mistaken on both points here, Ender. //

    I’m pretty sure I’m not. Sure the second was an exaggeration. However, when SV% and SH% change drastically in the postseason, illness and injury have much more effect, and lucky bounces affect the play much more, you really need to rethink who you’re defining as the “better” team. If there is even the remotest of parity between the two teams it could go either way and regular season record doesn’t really make a good predictor. I get that you want to say “make the best team possible and you should win” and that’s fine, but really it’s a matter of the better you make your team for the regular season, the higher seeding you will get and the worse opponent you will likely get – bad enough that the parity isn’t there to overcome.

    Regarding the first (and given the outshooting numbers I cited above) outshooting cannot be used to tie to playoff positioning. Only wins can. Given the numbers I posted above outshooting cannot be tied to winning. While at the low end of the playoff race the points get very close, we’re not looking at that. Your data is looking specifically at how much outshooting affects wins. If the difference between 1st and 8th is more than 16 points then outshooting itself cannot be said to affect it. *Possibly* some combination of outshooting and a number of other stats would, but you haven’t shown that. Your data doesn’t even *hint* that.

  10. Isn’t this a correlation vs causation problem?

    There is definitely a correlation between outshooting and winning and being outshot and losing – Jonathon’s data clearly demonstrates that.

    But whether outshooting actually causes winning and being outshot losing isn’t as clear.

  11. Hey, now that Derek just linked to this post, I’ve just read it, and I have to agree with your points here.

    When it comes to team play, I do think out-shooting is associated with out-scoring, though that’s only because there’s likely a strong link between out-shooting and out-chancing.

    The team that gets the most shots is also likely going to get the most scoring chances over time, and the team that gets the most scoring chances over time is going to score more over time.

    My main issue with all of this stuff isn’t rating teams with Corsi or outshooting or outchancing. I think there’s real value in that.

    My main issue is with applying a team based number to an individual player, to say that since Lubomir Visnovsky is on the ice for 200 more shots at the opposition net than at the Oilers net, that proves he’s a good player, a star, an out-scorer.

    I don’t think you can make that kind of leap. It’s fraught with peril. And, in fact, it doesn’t match up with my own observation of a player like Visnovsky this year.

Leave a Reply

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