I’m in a hurry, unfortunately, and don’t have time to comment at great length on this at the moment. Nevertheless, I wanted to throw this out there. I’ve been digging into the SF/SA numbers at ES on a game by game basis, pretty much entirely because of a comment I saw over at Lowetide’s that just stunned me in terms of (IMO) missing the point on discussions of outshooting. I’ve got a longer post in the works on that but for now, two very quick points:
1. 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.
There is an idea floating around out there that MacTavish’s coaching style involves building the sort of team that weathers the storm and then sticks the knife in on the counter attack, letting them get outshot with impunity. I’ve never bought it and I have serious doubts that there’s any team that will do better over the long run in games where they get outshot. Since 2005-06, 26 teams have done better in the ES portions of games in which they outshot their opponents than they have in the ES portions of games in which they were outshot. The four teams that have done worse are Vancouver, Montreal, St. Louis and Chicago.
2. This part stunned me. I’m going to go back and recheck my math but I’m pretty sure I’ve done this properly. I actually sat on it for a bit so that I could take a second look and I’m going to check it a third time but here it is:
I assume that the part that surprised me will be obvious to people. I’m going to check my numbers a third time, just to be sure, but wow. I’ve got some theories as to why that I want to check out but there seems to me to be a pretty solid argument there that playing for teams that outshoot their opposition on a regular basis is a drag on your save percentage.
* * *
I’m sad to see that Mike Winters has decided to shut it down. Can’t blame him, as running a frequently updated blog takes a lot of time. Their site was absolutely one of my favourites, just for the writing alone, and the ideas were good too. Hopefully his cartoon site continues and he’s able to offer us the odd bit of Oilers commentary through that format:
Excellent work. The evidence of the significance of outshooting at es mounts.
Dawgbone, a goalie himself, showed one time that a goalie’s save percentage is much better in games where he faces a lot of rubber, and much worse in games where he goes long stretches without seeing a shot.
I don’t know that this makes a hell of a lot of sense to me intuitively, but there was a compelling amount of evidence there. And I went through the games that I could remember, and by and large they were NOT games where one teams was just throwing a lot of rubber on the net. There looked to be the right amount of total shots in the game, just that the ice was severely sloped in one direction.
I mean the team that dominated usually won the game, but the save% of the guy who only faced a dozen or so shots was usually bad. And the guy that got peppered just grew better and better.
Since then, I’ve read some stuff around the internet that discredits the theory, but I think that they are looking at it wrong. They are checking see if the save percentage of team A is better than team B if team A allows an extra 5 shots on goal per game.
That wasn’t his point as I understood it, and I’ll happily be corrected:
His point was that goalies who are getting bombarded are sharper, and more likely to stop the next shot on goal. And goalies who haven’t seen a shot in 10 minutes of play are a hell of a lot more likely to let in the next shot than they would be if they had been busy.
Vic - I’ve looked at that before and didn’t find a hell of a lot of evidence in support of his position, although I had to cut it off at 60 seconds for lack of data. I should go back and do it again now that I’ve got 160,000 ES shots to look at. Personally, I have a hard time with the idea.
My idea is that the composition of the shots changes when a team gets outshot. Implicit in that is the idea that a team gets outshot will, on balance, take fewer shots than they do when they outshoot. The D are less likely to be holding the line and getting shots from far out, the third and fourth lines aren’t going to be clubbing the puck in the general direction of the net as frequently, that sort of thing. It’s an easy enough thing to check out and I’ll make a point of running it for the Oilers.
The reason that this stuff is tricky is that there are a lot of confounding (I think that’s the right word) things going on. Also, as you’ve alluded to, people aren’t measuring the same stuff. Even DB and I aren’t measuring the same thing - he puts a game where the Oilers outshoot 10-8 at evens in one catgory as a low shot game, I put it in another as a game in which they outshoot. That’s, of course, assuming that he is just focusing on ES, which seems doubtful, because there seem to be about 8 people on the internet who have this data in a useful form.
Yeah, mudcrutch, in my opinion the thing that you have to do to model dawgbone’s thinking is write a script to find the games where the total number of shots, PP and SH included, is reasonable (that’s the vast majority of games of course) and then grab the ones of those where one team had way more than their share of the total shots.
Now teams that are dominating at evens almost always end up with more PP opportunities as well. So overall the total save% of the guy who is getting peppered may work out closer to the same than is fair, because he’s seen more PP shots.
But both the EVsave% and PPsave% for the guy facing a tonne of rubber is going to be better than expected, methinks. I’m sure DB will correct if I’m misinterpreting him, but I believe that was his thinking.
He whacked up some numbers on HF boards to drive his point home, iirc. A google turns up nothing, though.
At the time it didn’t sound right to me, and I wrote a quick little script to churn those out (for 02/03 or 03/04 I think, this was a while ago) and it was a fairly obvious pattern. Hardly a comprehensive study or anything, but once it got to the point where it seemed like only a fool would bet against the theory … I conceded the argument and carried on with my day.
Now it’s possible that my rough check may just have revealed a coincidence, but his statement seemed to have veracity. Go figure.
MC,
it looks like Mike has joined the gang at BoA.
Nice post, I look forward to the in-depth version.
What does this info do to the career of Grant Fuhr, and more to the heart of the matter of the author, Kelly Hrudey?
mc79:
To add. There may well be validity in your second paragraph. I mean if a team is penned in their own end for most of the game, they won’t get many scoring chances or many shots. But they are bound to get a bounce that sends them out on an odd man rush eventually. And the team that’s owning the play is going to see it’s depth forwards get shots and chances a lot, just because they’ll be starting in favourable circumstances a lot.
There aren’t many games, especially post lockout, where the shots and the scoring chances are really lopsided, though. It would be a fairly small chunk of games to look at.
And to keep things in perspective we’re looking at a chance in shot quality that results in about 1 extra goal-for per 100 hundred shots if you are being outplayed.
And 1 extra goal-against per 100 shots if you have the territorial advantage.
So it’s a really tiny thing to try and measure, and very tough to separate the effect of the goalie standing around do buggerall for 10 minutes of real time … and the quality of the chance being 1% better when it does come. We’re painting with fine brushes here.
Maybe that’s why the effect is only intuitively noticeable in games where one team absolutely dominated by faceoff zones and on the shot clock (i.e. territorially), I’m sure that the percentages are big enough to be noticeable then.
As well, you can’t refute DB’s thinking with just EV shots. Because a goalie will show “4 minutes without an EV shot” if he’s been in net while his team killed of a double minor. But really he has been ’staying sharp’, for lack of a better word, during that time.
Lowetide:
Well, as you know, 80s hockey in the regular season was often played with a lot of indifference. I did some cash in hand work for a guy back then, copying all of the stats from The Hockey News into a data file. And he largely ignored everything that happened before February.
Even with Gretzky’s style of game, though, the 80s Oilers dominated the shot clock in the playoffs. I guess we’ll never know for sure, the Oilers did have long stretches of territorial dominance, especially in the playoffs when they were trailing, and Sather would start using essentially two forward lines. I remember because it wasn’t a good thing if you were cheering against them.
Still, it never seemed like it was an eternity between chances on goal against the Oilers. I don’t remember Fuhr standing around looking bored a lot Dryden-style.
Some games against DET nowadays, or N.J or DAL a decade ago … it just seemed like you might never get another chance to score in the game. It never felt quite that way even with the great Oiler teams. Granted sports memory is notoriously distorted, but unless the NHL decides to publish this stuff electronically, we’ll never know.
Tyler
Sorry for the endless stream of consciousness, but I have a thought on a way to test your idea that when a team outshoots … their 3rd and 4th liners are getting more of the shots.
Write a script to:
1. Scan through the NHL.com ES sheets and pick off the games where one team badly outshot the other, say the 10% of games (120 per year) where that happened the most.
2. Using those 120 games, scan through the ES sheets agains and list out the forwards, on each team, that got the 1st, 2nd, 3rd and 4th line ice time. grab their jersey numbers.
3. Run that against your EV shots data, see if the depth forwards were taking more shots than normal.
I posted something on IOF a while ago, and without looking it up, at evens the on-ice shooting%s for lines 1 thru 4 were somewhere in the vicinity of 9%, 8%, 7%, 6%. Not that extreme, but that general pattern iirc.
So you could get a pretty damn good idea of the “expected EVsave%” just based on that.
Just a thought.
Totally agree that the rope-a-dope theory doesn’t hold water.
On the s%/sv% discrepancy, how about this: defending leads late in games. To start off, if you have a lead late in a game chances are you’ve had a better shot% for the game than the other guys. If the game has been close in shots, the trailing team has a good chance of surpassing the winning team in shots as they try to tie it up while the winners play it safe. Is there any evidence of trailing teams outshooting leaders late in games?
Vic:
“And to keep things in perspective we’re looking at a chance in shot quality that results in about 1 extra goal-for per 100 hundred shots if you are being outplayed.
And 1 extra goal-against per 100 shots if you have the territorial advantage.
So it’s a really tiny thing to try and measure…
Detroit’s mammoth territorial advantage over the sample means they allowed 73 shots for every 100 they took. Going by the league average 8.3 s%, that adds up to a difference of just 2.2 goals per 100 Detroit shots. The 1 goal per 100 shots looks pretty significant in that light.
Vic wrote:
“Now teams that are dominating at evens almost always end up with more PP opportunities as well. So overall the total save% of the guy who is getting peppered may work out closer to the same than is fair, because he’s seen more PP shots.”
This is also what I immediately thought about after reading the post, and it seems to be supported by the evidence.
For example, there is basically no correlation between shots against and save percentage at the team level for each individual season between 1997-98 and 2007-08 (see link).
http://1.bp.blogspot.com/_FgIrL1Fnd-Y/SQS2B_c3iKI/AAAAAAAAAAc/LRK220QkGgM/s1600-h/New+Bitmap+Image.bmp
So what seems to be happening is that getting outshot at even strength tends to increase save percentage, but this effect is balanced by the fact that getting outshot leads to being shorthanded more often, which in turn leads to more high quality shots against, thus lowering save percentage.
With three years worth of data, the EV s% still have a wider spread than the net shot differential.
S% ranges from 7.5% (PHX) to 9.6% (BUF). Average number of standard deviations from the mean: 0.775. Is three seasons worth of data enough to make these S% numbers significant? There must be some error range, but I don’t have the stats chops to figure that out on my own.
Shot differential ranges from -679 (PIT) to +1642 (DET). Average number of STDEV from the mean: 0.750.
That surprised me. I would have guessed that with this much data the S% would regress closer to the league-wide mean, and that three years worth of shot differential would make the gap between the good and the bad more obvious.
The S% isn’t a perfect proxy for shot quality, but it looks like shot quality is a factor at EV. Not as big a factor as the shot differential, but significant.
I basically went with 5 random goaltenders (well, Brodeur and Luongo were planned… the other 3 were random). I took 3 years of game data and essentially took every game were they faced 30 or more shots and put them in one category and the games where they faced 30 or less I put in another.
I took out all the games where they faced less than 30 shots but didn’t play the whole game (no reason in particular other than I felt those few samples might skew the results one way).
Basically what I found out was that all 5 of the goaltenders had a significantly higher sv% in games they faced more than 30 shots than they did in games where they faced less than 30 shots.
I don’t recall the data, but overall (between 3 goalies and 5 years) it averaged out to about a .008 difference.
I didn’t flesh it out much farther than that, it was more of a point of interest than anything I was trying to prove (see the small sample size).
Maybe someone could check and see if the goaltenders who faced the greater ammount of shots saw a higher ratio of shots from defensemen.
Or some sort of disparity in shot distance.
If you figured out the average ammount of shots directed at the net for an NHL game and then seperated goalie data into whether they faced in excess of that ammount or not, maybe some trends would become apparent with 3 or 4 season’s worth of game.
Here’s my uncomprehensive example to sort of illustrate what I’m getting at.
Team Corsi:
Det +150
Cgy +135
T.B -61
NYI -101
Shot Attmepts by Defensemen
Det 120 9.23/G
Cgy 135 9.00/G
T.B 67 5.15/G
NYI 94 6.71/G
So for example, over the course of the season opponents of Detroit could see 300+ more shots directed on goal from No Man’s Land than the opponents of Tampa Bay.
Only 5 goals have resulted from the collective 411 attempts by the 4 teams. 269 shots made it on net for a .018% shooting percentage, or to better illustrate the opposing goaltenders have a save percentage in excess of .981%.
I would get more shots from their defenders than a poor possession team because at even strength it usually takes some sort of attack zone possession to generate a meaningful shot from a blueliner.
That last paragraph should start: “I would think strong possession teams would get more shots and a greater ration of D-man shots than…”
If I’m way out to lunch someone should chime in.
Slipper -
I think that you might well be right, but it’s a bitch of a thing to check properly (because I’m obsessive). I’ll get to it at some point because I’m pretty curious about it myself.
Funny thing - this topic came up back in the day and it was pretty much the same crew discussing it. Plus Lowetide trying to wreck it all with baseball references.
slipper
That makes a bunch of sense to me.
Not to horn in on the nostalgia then, but,
I find Jeff’s theory at #10 to be really compelling. I’d like to co-sign onto his hypothesis.
Trailing teams take bad shots because they’re desperate for a lucky break. Leading teams don’t. Note that believing this doesn’t require any particular opinion on what happens to territorial play, though you’ll certainly see a leading team do a worse job supporting the puck deep when they want a FWD back.
But mainly I like it because the scenario sounds familiar. (I’m saying I like it as a hypothesis, not a conclusion.) It’s not an uncommon thing: the shot clock tells you that Local Team X mounted 3rd period pressure when behind, but your eyes say they never got anything going. Might be enough to account for a non-trivial difference in overall shot quality.
So, slipper’s theory is that a good possession team will take more low % point shots while a poor possession team will surrender more point shots against.
Easy enough to test. If the theory has merit, you’d expect:
a negative correlation btw SF and s% for strong possession teams,
a positive correlation btw SA and sv% for poor possession teams.
For the top 10 teams in shot differential, correlation btw shots for and s%: -0.25.
For the bottom 10 teams in shot differential, correlation btw shots against and sv%: 0.70.
It looks like there is something there.
Matt - To clarify my suggestion, I don’t think it is the extra poor quality shots taken by a trailing team that drive the difference in s%/sv%. The discrepancy looks too big for that.
I think those extra poor quality shots are simply more likely to move the game into the ‘outshooting’ category for the losing team. Since it’s a game they’re already losing, we can guess that their s% was probably a lot lower than that of their opponents (who would conversely slip into the ‘outshot’ category).
As a test, one could check to see if the losing team is more likely to outshoot the winning team in close games. Close in terms of EV shots, and close on the scoreboard when you include special teams goals. If the ‘outshot by’ category is selecting some close wins because of the late game flurry effect, it could add up to a big s% difference.
JeffJ
Yeah, even simpler, we could just knock out the third periods. Unless it’s a three goal lead, not many leading teams start ‘playing to the score’ before the 2nd intermission. That should catch most of it.
And not many trailing teams start the desperate flinging of pucks toward the net before the third either.
Related to this: if you are using one of the possession metrics to predict future results, it works a bit better if you actually exaggerate it a bit with an exponent.
So Corsi^1.3 predicts a shade better than Corsi alone.
Maybe that’s because we’re using the ability to have have possession as a measure of their repeatable quality as an EV team, and the ‘late flurry’ thing has made their outshooting (or zone) numbers less impressive than they should be.
Does that make sense?
And post 21 was inspired, jeffj.
Now we’re getting somewhere.
If someone was willing to write out the longest url in the history of the internet, we could see the percentage of EV shots by dmen that each goalie and each team faced. Just using the shotsopp.php app at TOI.
Scratch the notion about shotsopp working. It uses who was on the ice when shots were taken, it doesn’t look at the player who took the shot.
I think I’ll just wait for mudcrutch to get around to it.
Wouldn’t this whole issue be moot if we just had a decent shot quality model?
(Though perhaps that’s a little like saying “Wouldn’t the Kings have been better last season if they just had better defense, offense, and goaltending?”)
Jeff, re: #22 — yes, I caught that, I think. Your proposed test makes plenty of sense. So does Vic’s idea of eliminating the 3rd period: if 2/3 of the discrepancy disappears, then bingo.
Gut only, but like I say, it feels like I see games all the time where a team was outscored, outplayed & outshot in the first two periods, then outshoots the leading team 10-4 in the 3rd without actually “outplaying” them in any reasonable sense.
Although per #21 there appears to be something there with Slipper’s theory, I’m a bit skeptical and still lean towards your initial theory. DET and SJS are *still the same team* whether they are outshooting or being outshot, so any good explanation for Tyler’s results needs to account for the difference in those teams’ Shoot% for those two states.
Reggie Smith was a great player.
Re: Slipper #16
Not sure if you’re using 08-09 YTD data, but looks like you are from the totals…
What I couldn’t tell from the data set was whether DET etc actually had a greater percentage of total shots from their Dmen.
So I’ve grabbed the Team shots per game totals from NHL and there’s quite the range…
DET 36.5 s/g or 25.3% fr D
CGY 29.4 s/g or 30.6% fr D
TBL 29.9 s/g or 17.2% fr D
NYI 31.2 s/g or 21.5% fr D
Do these numbers indicate a correlation between D shots and out-shooting? They look like it.
Or do they indicate teams with stronger D corps/poorer forward grps?
Eg.
NSH 27.6 SF/g; 27.1 SA/g
CHI 29.8 SF/g; 29.8 SA/g
NSH gets 8.8 S/G fr their Dmen or 31.8% of their total shots. They don’t put up the massive out-shooting totals, but they have a strong D corps.
CHI gets 7 shots per game from their Dmen or 23.5%. They get 2 more shots per game than NSH, but it looks like it’s more because of their strong offensive forwards.
WSH is an out-shooter at 31.5/g, but only get 22.9% of the total from Dmen.
I don’t think the correlation between out-shooting and D shots is going ot be all that strong.
Spoiler:
I only used shots strictly from defencemen because I don’t have the aptitude to strip the shot distance data or do anything with it. The less limited view of the theory would be out-possession or out-shooting teams have a higher ammount of perimeter shots.
I didn’t use two of the best and two of the worst out-shooting teams to sandbag my point, either. I think the better method would be to source the data not by team but by games where one team exceeded the league average of shot attempts. Being limited in that ability I chose a team like Detroit and Calgary because their underlying numbers infer that they have outshot their opponents in most of their games. The inverse of this reasoning is why I chose Tampa and NYI.
A more comprehensive study would look at shot attempts strictly by distance and not by who attempted the shot, and seperate the data not by team, but by gam numbers where one team exceed the league average for shot attempt.
Or some hybrid of that working theory.
I remember a conversation with Jeff months, maybe even years, ago regarding the team that’s trailing in a game being more likely to score the next one. I mean in hockey it’s not as obvious as soccer, but teams really do play to the score. I don’t think that any of us are disputing that at all, are we?
So would it make sense to just look at what has happened in games just when the score is tied? I would think that the best teams would look even better and the poor teams even worse, no?
I think that would remove a variable. And the shots+/- from the time leading up to a goal that broke the deadlock … more reliable I think.
I’m open to criticism here, not maried to this idea, just a thought.
Sunny:
Not my place to answer really, but I will
. I think we’re trying to build the picture back up from things we see. Tangible stuff.
Shot quality metrics are working with strange data to my mind (if they review the Brodziak goal last night and decide it bounced of a NYR player, then the shot quality for the night moves from below average to above average).
Plus it tells no story as to whether or not shot quality, for lack of a better term, is driven by the team with the puck or the defending team. And I find out tonight that some folks think that it’s mostly down to the goalie, which strikes me as pure madness. But there you go, hockey fandom is unique.
There is a lot of luck in a hockey game (two madass bounces in the single second before the aforementioned Brodziak goal btw, check it on NHL.com) so just piling together all the recorded stats, in the form that the NHL has decided is best, then running a bunch of regressions … that’s going nowhere fast.
Start with the game, work our way outwards. That strikes me as the thinking of the folks who have contributed in this thread.
The shot quality of a perimeter shot in itself is low. Perhaps a 1-2% chance of directly going in. The events that it can lead to are incredibly high quality, though. Rebounds, deflections, tip-ins, bounces, offensive zone stoppages, etc. Not to mention the potential hooking and holding calls in front of the goal mouth in the ensuing scramble or after an attack zone draw. High penalty event situations.
Another problem I see on the play by play sheets is that they don’t record distance for blocked shots. I think most people would agree that blocked shots are usually from a distance, right?
On Monday against the Rags between 17:59 and 18:26 of the 2nd period, the play by play reads something like:
NYR offensive zone face-off win
Redden blocked by Gilbert
Fritche shot 35 feet
Girardi blocked by Pisani
Girardi shot 45 feet
Horcoff penalized for hooking
I think the no measurement for blocked shots would be an issue.
I was going to say that that summary read like 27 seconds of good possession.
Oops, here goes the neighbourhood. Somehow I missed this until today, so have a little catching up to do (again/still). My compliments to you Tyler for some truly outstanding work here, and to the commenters for the terrific thread that has ensued. Fascinating results to say the least.
One surprise (to me) from the table is the sheer margin of shots. Obviously the outshooting team should be way out in front since by definition they outshoot every game, but 88603/64778 = 1.368 which is pretty darn large. You don’t list GP but the shots taken in “tied” games are about 9% of the total shots so I will assume that they are also 9% of the total games. I also assume that you have examined games in the 2005-08 seasons but not the current season, and the Oilers W-L-T record at ES that you cite would seem to confirm this. The average margin of shots would therefore be in the range of 26.4 to 19.3, which is very significant and larger than I would have guessed for a mean differential. Factoring in the ties to include all games, the average game sees a shot spread of something over 26.0 to 19.6, pretty much 4:3.
The “stunning” part as you put it has to do with the results of those shots. Comparing columns of numbers, only one team, Pittsburgh, had a better S% than its opponents in outshooting games, and that by a bare 0.1% (8.9% to 8.8%). One other team, Ottawa, had the same 8.5% either way. Doing a similar comparison in the outshot section, again just one team bucked the (opposite) trend, in that only Tampa Bay had a worse S% than their opponents in games they were outshot (8.7% to 8.9%).
Comparing each team to itself rather than to its opponent: just one team, Pittsburgh again, had a better S% in games where they outshot the opposition compared to games they were outshot (8.9% to 8.5%); while there wasn’t a single team that had a better Sv% in outshooting games than outshot.
Combining those four measures, that’s a sample size of 120 with just 3.5 exceptions. That’s not a trend, that’s a landslide.
26 teams have done better in the ES portions of games in which they outshot their opponents than they have in the ES portions of games in which they were outshot.
So what is the collective points percentage of the outshooting teams? 53% of all points awarded? (Thanks again, Buttman)
So it’s a really tiny thing to try and measure, and very tough to separate the effect of the goalie standing around do buggerall for 10 minutes of real time … and the quality of the chance being 1% better when it does come. We’re painting with fine brushes here.
Applying my own fine brush, the difference between a S% of 9.3% and 7.6% is closer to 2% than 1%. More to the point, the ratio between the two (which are even further apart than they appear once rounding is discarded, the true percentages are 9.318…% and 7.557…%) shows that teams being outshot have a S% that is a staggering 23.3% better than the outshooting teams. Which doesn’t make up the shot differential of 36.8%, but it’s most of the way there. Thus the seemingly anomalous result of outshooting teams actually outscoring by just 10.9%. Which is still very significant, don’t get me wrong; it’s just that shot vs. goal differentials are a very long way from commensurate. and if I’m not mistaken, win differentials are lower still.
Detroit’s mammoth territorial advantage over the sample means they allowed 73 shots for every 100 they took. Going by the league average 8.3 s%, that adds up to a difference of just 2.2 goals per 100 Detroit shots. The 1 goal per 100 shots looks pretty significant in that light.
Especially if it’s 1.7 goals per 100 shots. I don’t have numbers at ES offhand, but overall Detroit posted very average numbers in both S% (8.9%) and Sv% (.909), which was all they needed given their territorial domination. I would guess that it’s very uncommon for a President’s Trophy winner or even a 100-point team to have S% + Sv% Totally agree that the rope-a-dope theory doesn’t hold water. On the s%/sv% discrepancy, how about this: defending leads late in games. To start off, if you have a lead late in a game chances are you’ve had a better shot% for the game than the other guys. If the game has been close in shots, the trailing team has a good chance of surpassing the winning team in shots as they try to tie it up while the winners play it safe. Is there any evidence of trailing teams outshooting leaders late in games?
JeffJ: The “rope-a-dope theory” and the “hang-on-Harvey theory” – both of which I have advanced at various times — are closely related. IIRC, one of the points that Tyler and I actually agreed on during one of these discussions last season was that game situation is huge. Tyler likes to break it down into manpower situations which is certainly legitimate, but in my mind an even bigger factor is game score.
I can’t tell you how many games I’ve watched, especially Oiler games it seems, where the trailing team was pouring it on in the third. That was a common theme of almost every game of that 20-game stretch run last year that I examined in detail, regardless of whether Oilers led or trailed. We’ve all seen it countless times; that’s just hockey.
Moreover the counter attack works best in what I would call “positive situations”, either leading at home, leading or tied on the road. At such times the trailing and/or home team is likely to be pressing the play. A logical result of this imperative is a higher quantity of shots at one end and perhaps a higher quality at the other. As a tactic, rope-a-dope don’t work when you’re down 2-0.
Yeah, even simpler, we could just knock out the third periods. Unless it’s a three goal lead, not many leading teams start ‘playing to the score’ before the 2nd intermission. That should catch most of it.
Good idea. Extending the concept, I would absolutely love to know the split in shots up to the moment the winning team took the lead to stay, and after that moment. I would bet a lot of money on the “getting the lead” part of the game having a much superior SF:SA ration to the “keeping the lead” part. Perhaps this is the sort of thing that can be “stripped” from play-by-play sheets; all I know for sure is that I am incapable of doing that.
A possibly useful piece of additional information might be home-road splits; it is my perception – not backed up by any actual research I have done personally – that the home team is particularly likely to pour it on when trailing, and probably even more likely to do so when it’s a two- or three-goal spread than a road team that might be a little more prepared to mail in the points at that stage. But I’m cheerfully ready to be proven wrong by actual facts.
Funny thing - this topic came up back in the day and it was pretty much the same crew discussing it. Plus Lowetide trying to wreck it all with baseball references.
Wow, what a great thread. Very high level stuff all the way through, with great baseball references to boot. I particularly noted Igor’s comments about Turco and Brodeur getting the puck moving in the right direction, which predated my very similar comment to Vic of yesterday (!) by about four years. (Come back, Igor, wherever you are.) I sincerely wish I had been around for all this discussion rather than being left to my own devices all these years. Anyway, thanks for the link to that excellent discussion in the past, and for allowing me to participate in this one.
the usual oops on these edit-free sites — the paragraph addressed to JeffJ was in response to this that he wrote:
Totally agree that the rope-a-dope theory doesn’t hold water. On the s%/sv% discrepancy, how about this: defending leads late in games. To start off, if you have a lead late in a game chances are you’ve had a better shot% for the game than the other guys. If the game has been close in shots, the trailing team has a good chance of surpassing the winning team in shots as they try to tie it up while the winners play it safe. Is there any evidence of trailing teams outshooting leaders late in games?
Somehow it got expunged in the publishing process.
AAARRRGGGHHH!
Very sorry, but now I see what happened. Some kind of formatting error expunged an important section of my own comment from the above. (I thought it looked short. 
I used a “lesser than” symbol in an equation which had the effect of erasing everything up to the next formatting change.
This is now out of logical sequence, but here goes:
***
Detroit’s mammoth territorial advantage over the sample means they allowed 73 shots for every 100 they took. Going by the league average 8.3 s%, that adds up to a difference of just 2.2 goals per 100 Detroit shots. The 1 goal per 100 shots looks pretty significant in that light.
Especially if it’s 1.7 goals per 100 shots. I don’t have numbers at ES offhand, but overall Detroit posted very average numbers in both S% (8.9%) and Sv% (.909), which was all they needed given their territorial domination. I would guess that it’s very uncommon for a President’s Trophy winner or even a 100-point team to have S% + Sv% “lesser than” 1.000 – which is a study for another day, this post is long-winded enough – but when you outshoot the other guys by 3:2, “average” percentages mean you also outscore them by 3:2, which is a winning formula in any league.
There is a danger IMO in using such an exceptional team as the current Red Wings as the poster boys of how things “should” be done, not quite as dangerous as using the 80s Oilers maybe, but nowhere near representative. I can’t think of many teams since the merger that would have had a shot differential anywhere close to the +890 that Detroit posted last year. Hmmm, another study for a rainy day.
A conclusion for today, however, is that whatever the reason “why”, All Shots Are Not Created Equal. Which is what I have been saying all along.
Extrapolating slightly from your data, a shot in a low-shot game has more value than one in a high-shot game. That such a huge deviation from the mean S% can be found in any identified Large sample suggests that particularly on teams with strong tendencies on the shots clock, there should be an expected distortion from the norm in Corsi numbers and especially Save Percentages. I say “especially” Sv% because it can be argued a Zetterberg or a Datsyuk or a Lidstrom or especially all three together, are the ones driving the shot clock, therefore individual/team Corsi results are hopelessly entangled. Whereas the standard argument w.r.t. goalies – to which I do not entirely ascribe, btw — is that the netminders have much less/little/no effect on the number of shots that they face, all they do is stop the puck. So why do they collectively outperform themselves when they are being bombarded?
Totally agree that the rope-a-dope theory doesn’t hold water. On the s%/sv% discrepancy, how about this: defending leads late in games…
JeffJ: The “rope-a-dope theory” and the “hang-on-Harvey” theory – both of which I have advanced at various times — are closely related….
etc.
***
Again, my apologies. I am spoiled by the edit function. Tyler, if you would like to splice the foregoing and post #35 into one contiguous post and delete the argh posts, I would be further in your debt.
Vic,
I agree about the luck thing. It’s gotten to the point where I watch hockey games now and feel like I don’t know anything. Like I have no idea if what I’m watching is actual repeatable skill, or just a puck randomly bouncing around the ice.
Sometimes things make sense intuitively. When I watched Detroit last season, it certainly SEEMED like they were the best team. (Is it sick that I was rooting for them in the playoffs because I felt it would be a win for skill over luck?) But the NHL in general appears more and more like a bunch of coin flipping to me. Good thing I’m a degenerate gambler.
But seriously…
If we get to the point where we can start computerizing movements on the ice, etc, we might be able to start building decent models. But that’s probably science fiction at this point.
It does seem logical to build things up from what we see, as you say. But my eyes deceive me all the time. Plus, when I see a play on the ice it’s already hard enough to process it on its own - it’s impossible for me to process it in terms of it being played better or worse than the average of infinite trials of NHL players performing the exact same play, movement, shot, check, save, etc.
Bruce:
I believe that you and Igor would have hit it off famously.
Sunny
Just because it’s a luck soaked, and you’re aware of that, it doesn’t mean that it’s not fun to watch, does it? There are still the same great plays to watch, and the importance of the nuanced plays takes on more value I think. The Hockey Gods are often unjust, that’s all, we can all live with that I think.
And JeffJ is enormously correct here, he can take the rest of the day off
I wrote a quick script to churn out the EV results only at times the game was tied (playershots0708tied.php with the usual url extensions).
It turns out that every team is pretty damn similar when the score is tied. Frighteningly so, more than I would have thought possible. And the territorial advantage metrics rule with authority.
We’re getting there.
Why was Edmonton so much fucking worse than any other team when the score was tied?
Christ.
I was just looking at the Corsi # for some of the bottom feeders from 07-08. When the score was tied Chicago was -26, Phoenix was -18, Columbus was +100.
Edmonton was -318.
Well, on the whole PHX, CHI and CBJ were much better 5v5 teams than the Oilers. The Oilers were an outrageously good shootout team, and had better special teams last year.
The Oilers also played a a bit more with the score tied in games than a lot of teams did. So their suckage is exaggerated a bit.
Much, much better this year to date, schedule considered.