I’ve been getting a lot of comments on my look at the Toronto Maple Leafs this morning. In it, I argued that the Leafs’ save percentage and shooting percentage are unsustainably low, with both being worse than the expansion Atlanta Thrashers, the low water mark for NHL teams over the last 10-15 years.

Still, the Maple Leafs aren’t the only team whose results have been skewed by the percentages over the first few games of the season, and I was curious who else in the league was experiencing either abnormally high or low marks. With that in mind, I’ve calculated shooting percentage and save percentage for every team this season, and I have the best and worst marks from last season marked in bold. One caveat: these numbers are based on the NHL’s per game numbers, rather than the total numbers (which they don’t provide) so there’s bound to be some slight rounding error.

Shooting Percentage

  1. Atlanta: 15.4%
  2. Edmonton: 15.2%
  3. Colorado: 15.0%
  4. Calgary: 14.5%
  5. Los Angeles: 14.2%
  6. NY Rangers: 12.8%
  7. Washington: 12.1%
  8. Philadelphia: 11.6%
  9. Pittsburgh: 10.9%
  10. Last Year’s Best (Pittsburgh): 10.9%

  11. San Jose: 10.6%
  12. St. Louis: 10.3%
  13. Dallas: 10.1%
  14. Columbus: 9.9%
  15. Chicago: 9.6%
  16. Anaheim: 9.6%
  17. Boston: 9.5%
  18. Detroit: 9.2%
  19. Vancouver: 8.9%
  20. Montreal: 8.8%
  21. Carolina: 8.5%
  22. Tampa Bay: 8.2%
  23. Ottawa: 7.7%
  24. New Jersey: 7.6%
  25. Last Years Worst (NY Rangers): 7.5%

  26. NY Islanders: 7.5%
  27. Minnesota: 7.5%
  28. Toronto: 7.1%
  29. Florida: 7.0%
  30. Phoenix: 6.9%
  31. Buffalo: 6.5%
  32. Nashville: 4.7%

Save Percentage

  1. Phoenix: .950
  2. Buffalo: .945
  3. Colorado: .940
  4. Columbus: .933
  5. NY Rangers: .928
  6. Last Year’s Best (Boston): .925

  7. Anaheim: .920
  8. Atlanta: .914
  9. Nashville: .913
  10. Edmonton: .912
  11. NY Islanders: .910
  12. Pittsburgh: .909
  13. Tampa Bay: .906
  14. Philadelphia: .905
  15. Ottawa: .901
  16. San Jose: .901
  17. New Jersey: .901
  18. Los Angeles: .898
  19. Carolina: .897
  20. Florida: .896
  21. St. Louis: .896
  22. Calgary: .894
  23. Washington: .893
  24. Montreal: .890
  25. Dallas: .888
  26. Last Year’s Worst (Toronto): .885

  27. Vancouver: .876
  28. Boston: .874
  29. Detroit: .873
  30. Chicago: .872
  31. Minnesota: .864
  32. Toronto: .841

What It All Means

View the bolded numbers as a rough framework of what kind of variance we can expect in shooting and save percentage. Teams on either end of the bolded numbers are going to see their shooting/save percentages move to the midpoint (or, to use the fancy math wording of commenter Thrillhouse from this morning, ‘regress to the mean’).

I mentioned it in the post linked above, but Toronto simply can’ continue to get batted around like this, unless they’ve decided to employ me as their starting and backup goaltender (they haven’t, by the way). They will rebound, and when a bunch of stories come out crediting Ron Wilson’s approach to their long break coming up, or Jonas Gustavsson’s lucky underwear, or even the motivational speaking of Ryan Hollweg, just remember that this was going to happen anyway.

Similarly, the Minnesota Wild (currently ranked 15th in the West) are going to improve. Their 7.5 shooting percentage will probably get better, and despite the fact that Jacques Lemaire has skipped town there’s absolutely no chance that their goaltenders finish the year with a combined .864 save percentage.

On the other end of the spectrum, the Colorado Avalanche and New York Rangers are the two teams combining incredible shooting percentage with incredible save percentage; that will not continue, and with any luck John Tortorella will freak out on Sean Avery. I’m really hoping that the latter’s movie plans go ahead and that gets the blame, just for the sake of absurdity. That said, I imagine the Rangers will still be good (they should have a plus save percentage anyway), although I can’t say the same for Colorado.

Does anything else on this list stand out?

Comments (27)

  1. Rats! And just when I was thinking that maybe Pat Quinn really was a far superior coach to MacT . . . It turns out he really just has a better relationship with Lady Luck.

  2. David Staples:

    I like most of what Quinn has done so far, but there’s no denying that some of the Oilers’ results have come about because of that lovely SH%.

  3. Any idea exactly how much SVPCT and shot % regress to the mean? [Story idea!]

  4. While watching the Montreal game the sportscasters noted that the shot clock was way off the number of actual shots taken by the Oilers given deflections and blocks. I wouldn’t raise this but for the fact that at 5 games it’s a small sample size and easily skewed.

  5. If you factor in the 4 own goals I’m hoping averages out…

  6. The joys of small sample spaces.

    The Nashville game aside, the Oilers had a 12.3% shooting percentage with 23% of their scoring on the PP. The first is definitely closer to the mean while the latter is about where it should be given their currently detroit-esque PP% (they’d be in the top 5 last year).

    Last year, EDM had a 9.5% shooting percentage with a 17% PP%. Shouldn’t that about even out? Don’t get me wrong, I hear what you’re saying about percentages and whatnot, but given the sample space you’re taking and comparing it to last year instead of how the team in question did last year is kind of apples to oranges, isn’t it? Or will shooting percentage stay about the same regardless of PP%?

  7. (I guess I should mention that I’m ignoring the Nashville game specifically because a blowout this early in the season = 30% of total goals)

  8. Ender:

    Although the Nashville game also represents one-third of the Oilers wins.

    As for comparing it to last year, I’m really only doing that to give us an idea of where we should expect things to fall. Using Edmonton or Atlanta, for example, we don’t expect them to continue being 150% as efficient as last year’s Stanley Cup champions. Thus we know that they need to shoot more if they want to keep up their current pace.

    I’m not trying to make any grand statements here, just showing where we should expect to see more or less pucks to go in going forward, and reveal who possibly hasn’t been as good/bad as we might think otherwise.

  9. Willis:

    Like I said, I get what you’re saying about the percentages and regressing to the mean. A lot of rookies are getting their feet wet everywhere so defense is likely to be shoddy in a lot of cities right now, along with goaltending and offense. At the same time, luck is always more apparent in a small sample space and though Edmonton has had a ton of bad luck already this season, they’ve also had some good luck as well. They just both get amplified within the small sample space.

    That said, there’s no reason why last years numbers need to mean anything. Goal scoring could be higher or lower across the board this year, and a lot of teams are going to get hammered with crappy schedules because of the olympics.

    However, I’m sure you’ll argue that at least at the extreme peaks and valleys of your list there has to be some change, and I’m sure there will be. That said, if the Edmonton SH% is skewed 3% by one game in the 5 they’ve played, I’m sure a lot of the other teams’ numbers are horribly skewed as well.

    So I’m not sure that you’re really accomplishing “showing where we should expect to see more or less pucks to go in going forward, and reveal who possibly hasn’t been as good/bad as we might think otherwise.” You’re really only saying which teams should expect their percentages to go up or down. Percentages going up or down don’t equate to better/worse or more games lost/won, because as you said yourself “we know that they need to shoot more if they want to keep up their current pace.”

    At least do it against Corsi rather than shots that hit the goalie. You could just as easily argue that a higher percentage of Oiler shots that get through to the goalie go in – all the bad shots just get blocked anyway. It doesn’t really say anything about the shooters, goalies, or defense.

    I mean, you’re trying to quantify luck here Jonathan.

  10. Ender:

    “Percentages going up or down don’t equate to better/worse or more games lost/won” is the bit where we disagree.

    Let’s say Team X takes 100 shots in 4 games and score 15 goals. We know Team X is talented, and we expect that talent gives them a mean SH% of 10%. If they want to make up the five less goals over their next four games, they need to take 50 more shots. It simply isn’t likely to happen.

    I grant that we don’t know what the mean is for these teams, but we do know with a reasonable level of certainty that every team here has a mean somewhere between 7.5% and 11%, possibly with some displacement to allow for league scoring trends going up or down as a whole.

    Can a team like Nashville be expected to shoot half as much as they are now? Obviously not, particularly since we know that the gap in shots for per game last year between the best team was only 31.6%. We’d expect fluctuation on the same team to be less than that, if we believee that shots for over the course of a season is a product of personnel and systems play, which doesn’t change.

    In short, absolutely the percentages shifting changes who wins and who loses games because we know that variable shifts far more than shot count does.

  11. Let’s alter your argument a bit, to show you where I’m coming from. Let’s say Team X attempts 150 shots in 4 games and 100 get through, scoring 15 goals. We know that Team X is talented, and we assume they should be at about 10% SH%. If they want to make up the five less goals over their next 4 games, they take up to 30% more shots, bringing their total up to 195 shots over the next 4 games, and 150 of those get through to the goalie. It’s easily reasonable to happen.

    The issue is that we don’t know the baseline for any team. We know that from game to game just about every team will have between 20 and 40 shots count as shots. They’ll also have at least half their total on top of that go a bit wide or get blocked.

    Now, knowing that % of Shots that get Through changes from game to game, knowing that SH% also changes from game to game and knowing that SV%/GAA is independent of SH%, we cannot make any direct connections between SH% and winning games. Sure, it helps, but it doesn’t mean much in the grand scheme of things. In a game where you have a high SH%, you’re also likely to have more SA than usual because the other team is playing to the score (unless you’re outright dominating).

    Like I said, your argument *might* mean something with Corsi, and likely would mean something if you took the corsi shot % and added it to the SV% (like adding PP% and PK%) but your math, as it currently sits, doesn’t say what you seem think it does. It *might* be correlated, but it’s not a very strong correlation at all. Like I said, it seems to say more about the luck differential on each team right now than how good or bad any team is.

  12. As a sidenote, even teams within the bolds have to be skewed. Edmonton, for example, has scored a disproportionate number of own goals and had a number of goals due to bad bounces. Nearly half of their GA, IIRC. Yet they’re on the high end of the bold for SV%. Given what’s already happened, by your argument about outliers, shouldn’t Edmonton’s SV% be bound to improve, drastically?

  13. Ender:

    There’s a bit more wiggle room for shot totals to change drastically when you look at it that way, but it still seems vastly easier for 5 shots in 100 to go in the net than for a team to add 50 new shots in a short period of time. I haven’t tried to prove that since it seems obvious to me intuitively, but perhaps I will because of your objections.

    As for Edmonton’s SV%, I don’t see how you get that from my argument. What constitutes a disproportionate number of goals due to bad bounces? I don’t see any proof that Edmonton’s ratio of bad bounces is any worse than the majority of NHL teams.

  14. Just so we’re clear, I’m saying it’s feasible to add 50 new SOG in a short period of time partly because of less missed shots and partly because of shooting more. It’s not just one or the other. (You may already know that, but just in case I wasn’t clear).

    Regarding the SV%, of the 14 GA (in regulation), at least two were weird bounces and 3 were own-goal deflections. Combined, that’s more than 30% of the GA, which, to me, is likely larger than the norm, though I haven’t looked at other teams too deeply.

  15. Also, for the sake of clarity, I’d expect shots/game totals to come to a mean over the course of a season – I just don’t think that 5 games is enough to judge how much teams can potentially shoot the puck.

  16. Ender:

    It might interest you to know that over sixteen five game stretches last season, the Oilers never deviated more than 20 shots from their average.

    On the other hand, they deviated more than 5% from their average SH%.

  17. WIllis:

    As in 4 shots per game? Above and below? What was the average number of shots/game? Also, if that’s the case, wouldn’t that mean that if Edmonton had a SH% of 9.5 last year and an abysmal PP%, wouldn’t that mean that with a good PP we would have had a better SH% than Pittsburgh?

    Either way, I wrote out a more complete addendum over at SNN if you’re interested.

  18. I should add that that is interesting info. I’m just not sure how relevant it is with this sample space. Without a baseline being set, I mean.

    If the last 5 games are the least they will shoot (from the norm) adding another 20 shots drops their SH% to 13%. If the Nashville game is dropped as I did above, and adding 16 shots instead, that drops the SH% to 10.6%, under the bolded line. Even within your criteria, Edmonton could still very well be within acceptable parameters and not an outlier at all.

  19. Ender: Perhaps, but that requires you to drop the Nashville game and knock their winning percentage from 60% to 50%, which I’d argue is relevant.

    Rejecting 20% of the sample and then inflating shot totals by more than 15% is fine in theory and all but basically what you’re doing there is looking at the data, deciding you didn’t like it, and assuming that Edmonton was at the lowest ebb in the shot count.

    It’s possible, but it’s really unlikely and the smart bet is that it isn’t the case. That’s really what this is about – we can’t predict things with absolute certainty, but we have a decent idea of what’s probable and your hypothetical isn’t.

  20. Ender’s writeup that he refers to above can be found here:

    http://stillnoname.com/2009/10/an-addendum-to-something-i-didnt-write/

  21. The issue with Nashville is that it’s very rare for a team to get 6 goals in 20 shots. Like less than once per season rare. As such, we can immediately see that it skews the results drastically. Using something that we know to be anomalous in setting a theoretical baseline isn’t so much “not throwing out data” as knowingly including data that will not reflect the baseline if/when we have one.

    Regarding adding the shots, I’m working within your framework here. You said there was up to that much divergence so I thought I’d take a look. Again, without a baseline set, the values we have are within that curve, so we can’t say something about that value with this n.

    It’s possible, but it’s really unlikely and the smart bet is that it isn’t the case. That’s really what this is about – we can’t predict things with absolute certainty, but we have a decent idea of what’s probable and your hypothetical isn’t.

    Now who’s cherry-picking? Without a baseline you can’t say how likely or unlikely it is – it’s still within the theoretical curve taking shot deviation into account. The “smart bet” isn’t even an educated guess. It’s just a guess. You don’t have the data to support the idea (historically or otherwise) that the baseline is elsewhere. Again, given last years SH%, coupled with this years’ PP%, it’s just as reasonable to say that it is sustainable.

    The only thing that’s probable in all of this is that numbers are likely to go up and down. That in no way is a direct linkage to games won/lost. My hypothetical is just as likely, given these two stats, as anything you’re saying because it fits within your framework.

    Your argument likely works due to history – your Atlanta argument with regards to Toronto is a good one. However, what you’re arguing does not follow from the numbers above, period.

    Mine does.

  22. Ender:

    //Again, without a baseline set, the values we have are within that curve, so we can’t say something about that value with this n.//

    It’s true we don’t have a baseline. But probability indicates that we’re closer to the centre of it over the first five games than we are to the bottom; probability always indicates that.

    So yes, maybe this is the one five-game chunk of the season where the Oilers come in at the very bottom. But that’s not a logical assumption to make.

  23. //It’s true we don’t have a baseline. But probability indicates that we’re closer to the centre of it over the first five games than we are to the bottom; probability always indicates that.//

    Um, probability never indicates that. That’s like saying the first time you toss a die you’d expect it to be 3 or 4. That’s not how probability works. It’s only with a sufficiently large n that we can see where the mean actually is.

    //So yes, maybe this is the one five-game chunk of the season where the Oilers come in at the very bottom. But that’s not a logical assumption to make.//

    I wasn’t making any assumptions. I’m saying that within a probabilistic system, without a baseline, it’s reasonable to *say* that Edmonton’s on the low end of their shooting (coaching changes aside – which should actually reinforce that). It’s also reasonable to say that their on the high end, or somewhere in between. Basically, within the numbers you’ve provided, we can say that Edmonton can shoot a lot more, a lot less, or about the same. As long as the goal total and variance in shots taken can be made to fit within your bolded lines, you can’t say *anything* about the sucess of that team moving forwards based upon said data.

    I mean, nobody can stop you from making blanket statements, but if you want to understand the game better, you need to understand what the data gives you and what it doesn’t.

  24. Scratch the first reply there. I misunderstood you. I thought you meant they were *at* the center, which isn’t likely. Regardless, we’re talking about 4 low-shooting games and 1 mid-shooting with 1 high-shooting game in our now 6-game set (on a scale of 20-40 SOG). Proportionally that would seem to be on the lower end of what’s possible for any team, doesn’t it?

  25. Ender:

    With 125 shots, I’d suggest it’s near the low end just based on last year’s results, although those indicate that the Oilers’ expected shots/game is south of 30.

  26. Which is what I’m saying. If you take out that Nashville game which is going to be an anomalous data point regardless, Edmonton’s numbers work out to being easily under the bolded line. I’m sure other teams’ values have also been skewed. That’s why I think another reference point is *necessary* to say anything meaningful about this data (like I did on SNN).

  27. Hey Willis I think that the Oilers high shooting percentage is due to the fact that Quinn wants them to take more high percentage shots and they’ve been doing so. A lot of the goals they’ve scored have been very nice. It may fall a bit but there is quite a lot of skill in the line up.

Leave a Reply

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