As part of the ongoing ES outshooting discussion that’s been going on here, I took a quick look at the splits for teams in terms of at home ice and on the road at ES. Team summaries, from 2003-08, are as follows:
Interesting thing about that - the home team enjoys an edge of one extra goal scored and one extra goal saved per thousand shots for/against. That’s pretty small. On average, the home team outshoots the road team 22.3 to 20.9 at ES. By my math, that means that the average home team gets an extra goal scored and prevented every 740 games. There was a discussion at Lowetide’s a while back about the impact that individual players have on save percentage. I’ve come around to the idea that it’s pretty small and stuff like this is the season why. If there was really such an advantage to be had in terms of save percentage, I’d expect it to show up in things like this, where one coach has a real advantage over the other in terms of deciding who plays against who.
I put this information into a different form, just looking at the ratios of RSF/RSA, RGF/RGA, HSF/HSA and HGF/HGA. This helps in terms of clarifying the impact of home ice, I think. While you do see above that individual teams do better and worse with their save percentage and shooting percentage on the road, the league average is that there’s little difference in S% and SV% based on whether you’re a home or road team.
The same cannot be said for SF/SA. That’s where the home ice advantage looks to lie.
Some teams, of course, have wider spreads in their S% home and away and SV% home and away. I’m inclined to think that that’s nothing more than randomness or, at the very least, that you can’t draw any conclusions from it - I’m essentially just looking at two seasons of home games and two seasons road games when I do it on a team by team basis. The trend for home ice being a shooting advantage is an awful lot easier to see.
Interesting stuff as usual, Tyler, but I have to admit to having a serious disconnect here. Perhaps you can explain it.
In your Sunday morning data dump your data clearly showed teams with the shots advantage have a much poorer shooting percentage (7.557%)than the teams getting outshot (9.318%). This efficiency difference of some 23.3% is one of the most statistically significant trends I can remember in any of these studies.
Now there is new data, from those same three seasons plus one more, that expresses two things:
1) Home teams significantly outshoot road teams
2) This does NOT translate to any sort of deficit in S%, in fact home teams have slightly better S% (8.204%) than road teams (8.140%)
Expressed another way, adding S% and Sv% together:
Outshooting: 7.557% + .90682 = 0.98239
Outshot: 9.318% + .92443 = 1.01761
Home: 8.204% + .91860 = 1.00064
Road: 8.140% + .91796 = 0.99936
The home teams have a shots advantage of around 6.5%, which is hardly the 36.8% spread between shooting/outshot but nonetheless very significant. Indeed you conclude “that’s where home ice advantage looks to lie”. All things being equal, I would have expected home teams with that sort of shots advantage to have lower percentages by something like ~.003 (Home S% + Sv% = ~0.997). Instead the “correlation” is actually negative! How can this be?
Good question Bruce. Entirely possible I buggered something up adding things. I’ll double check the math tonight. If every team sees their shooting percentage go down when they outshoot and home teams are more likely to outshoot…you’re right, something doesn’t sound right.
Assuming the math is correct, what is going on with Atlanta? How are their shots 20% more accurate at home than on the road? Is this just Kovalchuk getting out against average players once in a while? Can’t be just that though, a few shifts here or there for one player can’t make that big a difference.
Tyler, I’m sure that your numbers are right.
The phenomenon of “team’s shooting% go up when they are being outshot” would seem to be entirely explained by Jeff J’s thought in the thread below. Teams are playing to the score.
This link gives all goals, shots, misses and blocks at even strength when the score in the game is tied.
http://timeonice.com/playershots0708tied.php?team=VAN&first=20001&last=21230
That’s for Vancouver. Just change the team abbreviation.
So when the score is tied, the relationship of the 30 teams shooting% to outshooting (shots for/shots against) is nonexistent. r=.03. A selection of random numbers is much more likely to produce a stronger correlation.
Goals ratio correlates to shots ratio with a ridiculous r = .74
fenwick r = .76
corsi r = .79
So once we remove Jeff’s ‘playing to the score’ effect, where in essence teams are willing to sacrifice some scoring chances in order to limit the overall number of chances in the game (watch a couple of Serie-A games this weekend for a exaggerated cross sport example).
Well there isn’t much room for the rest of the theories to live on. Which is odd, even to me, I’d bought into slipper’s thinking as well.
Also,
To check if the goalie’s save% (team shooting%) is changing with home ice advantage fairly, you should just use the shots and goals that happened when the score was tied.
Turns out it only took two minutes and one line of code to affect a change to just grab the home games/score tied.
use …playershotsteidh.php…
the rest ot the url in the same format as above.
Hey hey, score one for the broken clock!
You realize, Vic, that you’ve opened Pandora’s box with that script? There must be lots of people interested in the natural next step: players’ leading/trailing numbers. Is anyone particularly good at holding leads? How about generating offense when trailing? One could look into some interesting coaching decisions. If it’s possible to grab icetime, what happens to individual players’ icetime when leading/trailing? You expect Kesler to play more when ahead and Wellwood more when behind, but how much?
That’s not meant to be a request, BTW. How your server’s not already swamped is beyond me.
Okay, the script is slow, but the results predictable. I ran them for all 30 teams.
Overall, when the score is tied at EV strength, there is no correlation between shots+/shots- and shooting% or save%.
On home ice with the same scenario. shot+/shots- is good for the home team, as you would expect. Shooting% actually goes up, granted just a smidge.
Everything in your table in the post below is a result of playing to the score, and more specifically the fact that road teams tend to play more passively with the lead, and start doing it sooner. And teams that show a penchant for higher shooting% with fewer shots … they are teams that aren’t too worried about boring the crap out of their home fans. It comes when they are leading in the third, and would rather sit back and get outchanced 5-3 than keep playing their game and run the chances at 10-10.
Dennis’ stuff will bear this out once we’ve got a season’s worth of data.
BTW: Jeff J actually built a model for this phenomenon at sisu hockey a year or two ago. Using just goals mind, not shots, but it would have worked if he ever ran it.
BTW2: In your sims, Tyler, you use the goal scoring model from Alan Ryder. I’m sure you noticed that while it predicted number of goals in games, team by team, really well, it didn’t produce anywhere near enough ties after 60 minutes. Again, that’s playing to the score. Jeff’s model WILL produce the right number of ties, because it takes into account … say it with me
… playing to the score.
Jeff
Nah, the traffic isn’t too bad there, just the hardcore guys are into this stuff, as it should be.
My scripts are heavy handed though, so it results in way more hits than pages if scripts like ‘playershots’ are being run. I should be good for at least a million hits a month though, and TOI was just at 18,000 pages and 100,000 hits in October, looks like about the same clip this month. So I have plenty of time before I have to bother writing more efficient code. So go crazy, I don’t mind.
BTW: My thinking is that Corsi while the score is tied is the best measure of a team’s quality at 5v5, but you can just take regular corsi and apply a small exponent (1.25 or so) and get the same effect. I stumbled across that a while ago in true Bill James fashion (farting around with the numbers until something worked, I reference that in the comments of the Smid thread at LT’s a week or so ago).
Still, it’s nice to know where in the hockey game that this stuff is coming from. And I think we all love simple, commonsensical answers that work. So it’s all good imo.
You realize, Vic, that you’ve opened Pandora’s box with that script? There must be lots of people interested in the natural next step: players’ leading/trailing numbers.
Indeed. I would be one of them. For now I am more interested in teams’ leading/trailing numbers. The information when the teams are tied is most illuminating, but it tells just part of the story.
Turns out it only took two minutes and one line of code to affect a change to just grab the home games/score tied.
Vic, is it possible that you could write another line or two of your ingenious script to yield the same info for leading and trailing? Home teams leading/trailing? This would allow analysis of “playing to the score” whether tied or otherwise.
So many fucked up things are shoeing up through that new app of yours, Vic.
Montreal and Pittsburgh: Terrible in 07-08 with the score tied.
So many other teams: averagish. Like they’re playing a slow, deliberate game of ping-pong.
Sometimes more information is just more confusing.
slipper:
I think it really shows the power of a good PP in general, and good special teams in general. Look at how Montreal is no hell at evens when the score is tied, even with good goaltending they can barely hang in, then they get a PP, score … and then choke the life out of the game. Minny the same.
Vancouver also, except they hang in there when the score is tied until eventually Luongo stops more pucks than the other guy, then it’s Alain Vigneault joy killing time.
And let’s face it, people are going to keep showing up for home games in VAN, MIN and MTL no matter how they play.
To get rid of this shit, they need to change the risk/reward structure for the coaches, and the only way to do that (that I can think of) is to reward 3 points for a regulation win.
That would knock the crap out of the number of regulation ties though, and the NHL loves OT and esp shootouts (I love shootouts too, btw). So it’s not going to happen any time soon.
To get rid of this shit, they need to change the risk/reward structure for the coaches, and the only way to do that (that I can think of) is to reward 3 points for a regulation win.
Hear, hear, hear, hear, hear!!! Make all games worth the same. Of all the stupid things that have occurred on Gary Bettman’s watch, the system where “a win is a win, but a loss could be a tie” has most directly affected the competitive integrity of the game.
Holy crap. This is amazing stuff, guys.
Vic (or anyone who knows),
How does the script work when two players use the same jersey number in a single season?
I was just looking at the Canucks numbers for last year and I think that Sopel and Bourdon used the same number that year. Does it just synthesize them into one result?
To get rid of this shit, they need to change the risk/reward structure for the coaches, and the only way to do that (that I can think of) is to reward 3 points for a regulation win.
I’m still partial to the simple baseball-esque W-L system, but anything that makes for sanity in the standings and stops encouraging playing for the tie in the last ten minutes of the game would be just great.
Scott
My scripts use the jersey numbers to id the players, this for ease of programming. So if Sopel and Bourdon wore the same number for VAN one year, their results would show up as the combined totals for the jersey number.
Jeff J actually built a model for this phenomenon at sisu hockey a year or two ago.
See this post, from nearly two years ago.
There were some problems - I should have used the Poisson function for the scoring rate, it should have considered scoring rates when trailing as well as leading, should have only changed rates in the 3rd period rather than whenever a lead was taken… probably lots more.
I had to dig up the 71Mb(!) spreadsheet to point out the biggest problem I had with it. I just ran 10,000 games for a team that posts 2.2 GF / 2.2 GA when tied or trailing, and 1.5 GF / 1.6 GA when leading. They averaged 1.98 GF, 2.02 GA and posted a .519 record. Their expected w% for their scoring rate is .492. This just doesn’t happen in the NHL.
You would expect some clubs (Min, NJ) to be particularly good at holding leads. If they were significantly better at holding leads than coming back when trailing, they should slightly outperform their pythagorean win%. I looked at all the seasons available at nhl.com (so I could pull out the ENGs) and no team did. The Devils’ low-event game when leading would be partly balanced by their opponents’ high-event game. Their own high-event game when trailing would be partly balanced by their opponents’ traps. If they are truly better at one aspect of the game than the other, the difference is small enough to be lost in the noise.
And +1 on the 3 point reg. win.
The bonus point promotes defensive hockey. League bigwigs are either too stupid to realize it or are lying through their teeth when they talk about increasing scoring.
Jeff
Years ago I remember reading an article from Al Strachan that said that the guys who were promoting the OT loss point were suggesting 3 points for the regulation win as part of the same package. There was a split amongst the governors on whether to go ahead with it. In the end the NHL went half way with it.
I would be really interested to know who was promoting this idea (BUF or S.J would be good guesses) and which teams were staunchly opposed.
To me, this is the kind of thing that the NHLPA should have an enormous say in. As much as the entire board of governors, or close to it. This new CBA was promoted as a partnership between the NHL and NHLPA, but it is nothing of the sort.
There are encouraging signs, though. The NHL wants to expand the schedule to 84 games, and NHLPA President Paul Kelly is refusing unless the schedule is changed so that all NHL teams play each other at least twice a year.
I would trust Kelly’s NHLPA with stewardship of the game 1000 times before I would trust the NHL board of governors once.
I would be really interested to know who was promoting this idea (BUF or S.J would be good guesses) and .
On the GMs voting down the “3 point must” system in Feb 2007:
“Because it’s a terrible idea. That’s why it didn’t have any support. We made some radical changes when we came back from the work stoppage. The game is faster, the game is better, and the game is more entertaining. If something ain’t broke, there’s no reason to try and fix it.”
— Brian Burke
“We’ve got some exciting races in both conferences. The parity is great. Why change anything? I think we’ve had too much change of late. It’s a good game, let’s enjoy it.”
— Lou Lamoriello
“It’s time to establish continuity, you can’t keep making changes. Let’s not confuse the fans.”
— Colin Campbell
Coding error #64783 from this technoklutz … the above meant to emphasize, but instead expunged, the part of Vic’s question I was addressing, namely:
which teams were staunchly opposed
To which the answer would appear to be, the usual suspects. Among all those quotes “parity” would seem to be the keyword. Campbell’s pious concern about not confusing the fans came about ten years too late.
This is really interesting stuff guys - very cool. But I have to play devils advocate for a second. Isn’t “playing to the score, …would rather sit back and get outchanced 5-3 than keep playing their game and run the chances at 10-10, they hang in there when the score is tied”, kind of rope-a-dopey in nature? Thought that didn’t exist in hockey?;o)
oilman:
You’ve missed the point I think. “rope-a-dope” has just been blown to smithereens.
Say what?
In your sims, Tyler, you use the goal scoring model from Alan Ryder. I’m sure you noticed that while it predicted number of goals in games, team by team, really well, it didn’t produce anywhere near enough ties after 60 minutes.
Yeah, it’s funny but I was fooling around with that model a month or so ago and I noticed that. It consistently underestimates the amount of ties. Something I was thinking about doing is grinding through the data to see at what point coaches start playing to the score and what happens when they start doing so.
For some reason, I have the idea that sitting on the lead is a bad idea and that it hurts more than it helps. I’m sure that this is based on data, I’ll just be damned if I can remember how I arrived at this conclusion.
Also, if anyone is going to write a script to check this stuff, grabbing the ice time would be cool.
And of course teams ‘play to the score’ (i.e. play less aggressively in all three zones) when the game is tied in the third period as well. Both teams usually do it then. So just using the first 40 or 50 minutes of a game is probably favourite.
Tyler
Yeah, the other team is less likely to score if you are leading and PTTS (hockey needs more acronyms), but at the same time they are more likely to be the team that scores next, because you are even more less likely to score. Does that wording make any sense?
So you cut down your chance of a reg W, cut down your chance of a reg L, and increase the chance of a tie.
And with the way the math works out with the guaranteed point, you can’t blame the coaching staffs for rolling that way. And if the game ends in a shootout or OT goal, we tend to forget how tedious the third period was anyways, I suppose.
It also creates the illusion of greater parity, which is completely unnecessary in the past couple of seasons.
oilman:
You’ve missed the point I think. “rope-a-dope” has just been blown to smithereens.
Can you explain that Vic - my mensa card must be expired.
Totally off-topic, but I want to post a note where I’m sure Tyler will see it to say Thank You for posting that link to the Cards-Braves box score of Aug 23, 1962. You obviously went to some trouble to dig it up based on the scant clues in my post on Oil Droppings. Said boxscore provided all sorts of context to an event I was too young to remember in any depth. Thanks so much … that made my day!
Tyler
I’ve been checking on the IOF site stats the last couple of days to see how many people from OilersNation clicked on the link to Viscoff (so far zero btw, though I started a bit late, so maybe jonathon checked his own link before I started looking) And as an aside, when Matt suggested that people who don’t normally read comments should read the Viscoff replies, there was a rash of visitors referred by feed readers. And PDO seems to be arguing with Oiler fans about Horcoff’s value (everyone else thinks that he’s crap) … only two people have clicked his link so far, one of them presumably him.
But I digress, turns out that if you google Prom night be damed you get a link to an old article by Dennis where he’s tracking the scoring chances for a VAN EDM tilt a a year ago. VAN gets the lead early in the second and falls into a hard trap according to Dennis. Scoring chances disappear for both teams, VAN actually has none at all before the Oilers eventually score a goal late in the second.
Alain “The Thrill” Vigneault is now officially more boring than church.
Also:
The top google referrals for IOF are “Pythagorean Expectation”, “Soap Candy”, “But what does it all mean, Basil” and image searches for pics of a drunken David Hasselhoff. In the case of the latter, I don’t know what search terms they were using.
It would be nice if there was something hockey related near the top beyond “joffrey lupul family” (Ouch! btw). And speaking of lupul, a commenter called him a “big vagina” once upon a time. Someone arrived at IOF using that search, and iirc it was on about page 60,000 of a squillion. This dude must be dedicating his life to reading absolutely everything on the subject, which is disturbing and strangely admirable all at once.
Interesting thread has developed over at the Contrarian Goaltender’s site in response to the data presented by Tyler here.