Waiting for the regression...
by richer44 on Oct 19, 2010 6:33 PM EDT
Email
Print...to the mean, that is.
Editor's Note: This is a good, short analysis and has a proper perspective. The comments have been interesting too, so check them out.
TG posted two quotations, from Lou and MacLean that piqued my interest.
First, from Lou yesterday (emphasis mine):
"Devils general manager Lou Lamoriello is understandably disappointed in his team’s 1-4-1 start, but believes things would look a lot better if it had capitalized on more of its chances in the first six games.
"If we would have taken advantage of a couple of opportunities we’ve had at different times, our record would be the other way," he said today.
"...if we weren’t getting chances and we weren’t getting opportunities, that’s a different story. We’re getting them and the right people are getting them, so they have to go in."
And from Johnny Mac this afternoon (again, emphasis mine):
"As far as scoring chances, we are getting some quality chances," MacLean said after today’s practice. "We are definitely getting quality chances and I’m confident with this group if we keep getting those chances, the puck is going to go in."
What struck me about these quotes - besides how well the new head coach stays on-message - is the implication that a significant part of the Devils' recent apocalypse struggles is due essentially to luck - the bad kind, clearly, but luck nonetheless. The games against Buffalo, Colorado, and Boston saw the Devils outshoot their opponents 110-81, but they were outscored 4-7. Extending out to all 6 games, the Devils have outshot their opponents 204-169, but have been outscored 10-21. So what if a few more of those shots went in? Should we really expect them to? My thoughts after the jump.
UPDATE: I did a quick-and-dirty simulation last night to see what the probability might be for the Devils shooting 4.9% over the 204 shots they've taken this season. Based on the established mean (which I STILL contend is a big enough sample - 12,000-13,000 shots is a LOT of chance events), we can expect the Devils to shoot 4.9% over 204 shots 60 times out of 1000, or about 6%. The vast majority of results have the Devils shooting between 6.5-10% over that span.
Fair warning: I'm focusing solely on the offense here. 3.5 Goals Against/game is an atrocious average and clearly defensive lapses play a major part in the 1-4-1 record thus far, but for the purposes of this FanPost I'm ignoring the D.
Fair warning 2: I am NOT an advanced-stat guy - working on it, but not quite there - so I'm working purely from the team stats available on NHL.com.
It's been demonstrated that one of the more stable/reliable predictive statistics (in terms of scoring talent) is shot percentage and that, statistically speaking, teams who tend to outshoot their opponents on average also tend to win more games. I decided to look at team shot % rather than individual because the sample sizes are just too small.
Thus far this season, the Devils average 34 shots/game (6th in the NHL) and their team S% is 4.9, the lowest among the top 10 teams in the NHL for shots/game. Contrast that with the Lightning (3rd) and Avs' (10th) 9.3% or Nashville's (5th) 9.5% and it paints a pretty good picture of the Devils' offensive season so far: plenty of shots, not enough goals.
As a 'baseline' reading, since the lockout the Devils have averaged 29.2, 28.7, 28.8, 32.9, and 30 S/G in 2005-2009, with S% of 9.7(!), 8.74, 8.4, 8.8, and 8.76 in each year. See? Pretty stable.
Applying this season's averages to that reading, we see a team that is grossly underperforming relative to its established goalscoring talent. The shots per game are close to the mean, but that 4.9 S% is the difference-maker. Consider this:
If the Devils scored this season at their team average since the lockout (8.88%) with the same shots/game, they'd have 18 goals, or an average of 3.01 goals per game. That's nearly a goal and a half better on average than their current, abysmal 1.667 G/G. Truncating the decimal, let's add those goals to the Devils' scorelines for each game, changing only that variable:
Devils 4-4 Dallas (OT)
Devils 3-7 Washington
Devils 2-3 Pittsburgh
Devils 2-0 Buffalo
Devils 3-3 Colorado
Devils 2-4 Boston
Clearly, the Caps game is an outlier, but *ALL THINGS BEING EQUAL EXCEPT* that the Devils shoot at a similar percentage to their established mean, those games look considerably closer. If the Devils scored 4 in regular time against Dallas, they would've won that game. Scoring a second against Pittsburgh negates the empty-net goal and most likely sends that game to overtime, as does scoring the extra goal against Colorado. At that point, flip a coin and the Devils are 3-1-1 at worst and are third in the division with a game in hand against the Penguins.
Now, I'm not saying that's how those games would have gone if the Devils shot closer to their true mean, but I am saying they could have. And before we storm the Rock with torches and pitchforks or burn our Kovalchuk jerseys or whatever else fans do when they panic (Cry? Drink Heavily? Write FanPosts on ILWT?), let's all take a deep breath and wait patiently for the pro-gression to the mean.
All FanPosts and FanShots are the respective work of the author and not representative of the writers or other users of In Lou We Trust.
108 comments
|
1 recs |
Do you like this story?
Comments
One of the problems with “luck” and “regression to the mean” and all these fun terms is that they were the explanation for why the team struggled in the playoffs last year too. At some point (though I don’t think we’re there yet) the new averages aren’t outliers that need to regress, but just the level the team performs at.
Playing Devils' advocate since 1982.
Look at it this way:
Zach Parise is currently shooting at 3.8% and he has shot at least 10.5% in every season of his career, a career S% of 11.6.
Ilya Kovalchuk is currently shooting at 8.7% and his career S% is at 14.7.
Patrik Elias is shooting at 7.7%, and while he’s had 2 seasons at that level, his career S% is 12.3.
Jason Arnott is actually shooting better than last season at 10%, but it is still below his career S% of 12.1.
Travis Zajac is shooting at 9.1% and the last time it was that low was in his sophmore season; his career S% is 11.1.
Dainius Zubrus is shooting at 9.1% as well and his career S% is 11.6.
Then there are players like Jamie Langenbrunner and David Clarkson who haven’t scored yet, so their S% is well below their career total.
While I’m not saying everyone will shoot at exactly their career S%, there will be some variation, it is highly unlikely that all of them will remain below their career S% by this much and go on to have career lows across the board. Especially in Parise’s and Kovalchuk’s case.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 19, 2010 5:54 PM EDT up reply actions
Like I said, we’re not there yet, but if lower rates continue for long enough one has to adjust one’s expectations, regardless of previous performance.
In other words, calling it “bad luck” only works for so long. Eventually one has to admit that there’s more to it and maybe the baseline percentages were just “good luck.”
Playing Devils' advocate since 1982.
Right, but my whole point is that if it continues long enough, the expectation of a regression to the mean has to be adjusted, if not entirely thrown out.
Playing Devils' advocate since 1982.
I disagree – I think the mean is strong enough in this case that if these numbers were to continue we’d call it a statistically aberrant season. But if it gets to that, we’re really just arguing over whether the drapes on the Titanic are lavender or periwinkle.
Things change. Looking back at historical precedent is all well and good when determining what one should expect going forward… but when those expectations are consistently not being met, one has to rethink the expectations.
I’m not saying we’re there, or that we will be, or even where that line might be… but this whole regression to the mean thing has gotten out of hand. It’s presented as a unalterable truth when it’s really just a guiding principle.
Playing Devils' advocate since 1982.
well, clearly we’re at an impasse – I think I just have more faith in the predictive value of statistics than you do, at least in these circumstances. I wouldn’t say it’s an ‘inalterable truth,’ because we don’t know how many sides the die has – if we did, sports wouldn’t be any fun – but given the amount of data available and its stability, I’d bet good money these last 6 games are an aberration and not a new norm.
All I’m saying is that looking at the past results of an individual or group of players in different circumstances over a period of time doesn’t mean that it being different now means that it has to go back to where it was.
Playing Devils' advocate since 1982.
no, and that’s not what regression to the mean means, either. what it means is that their future performance should approach their past performance.
Even though I should know better than to ask, based on past experiences with your ridiculous argumentative style of taking things out of context, misrepresenting statements and arguing petty semantics, how is that any different than what I said?
Playing Devils' advocate since 1982.
now can we drop the nonsense?
Ah, yes, the unexpected misrepresentation of my statement technique. Didn’t see that one coming.
I said, as you point out, “go back where it was.” I didn’t say, “raise their shooting percentages to such a level that it all averages out to the previous level.” I said “go back where it was,” meaning, as you put it, “their future performance should approach their past performance.”
So, your “correction” not only doesn’t address my original point, but it just restates exactly what I said, only in different words.
Seems to me that the only limited powers of knowledge around here are a certain someone’s reading comprehension.
now can we drop the nonsense?
Any time you’re ready.
Playing Devils' advocate since 1982.
If it doesn’t, it’s a huge outlier.
The bigger issue IMO is not IF it’ll regress, but WHEN. It’s possible to overshoot or undershoot for entire seasons (see WSH last year). That being said, you probably don’t undershoot by 4% over 82 games. WSH’s “overshooting” was along the lines of 2% or so.
by red army line on Oct 20, 2010 12:51 PM EDT up reply actions
well, if you want to go back to before the lockout, the S% look like this: 8.75, 8.3, 8.38, 11.4 (2000-2001), 9.24 (1999-2000), 9.6, and 9.35.
Based on the mean S% since the lockout – which I contend is a resonable sample size, given the Devils took over 12,000 shots in those 5 years – the probability that the Devils shoot 4.9% the remainder of this year is .003%. Even including the high seasons prior to the lockout, you’re still only talking about a standard deviation of .85%. It’s slightly more likely that Kovy gets killed by a lightning strike this year than the Devils finishing the season at 4.9%.
I understand your skepticism – statistics can only ever tell part of the story, but in this case I believe it tells a significant part, and we’re much safer assuming the Devils will finish the season with a S% close to 8.8% than 4.9%
well, right – but inside the population that makes up the team average is each individual player’s S%. I just tried to maximize my sample size for this particular analysis by going across the whole team – that’d also even out any outlier years from an individual player.
right, that’s fair. i meant a summation of the devils’ individual players for their careers – kovalchuk and zubrus are good shooters who’ve spent most of their shooting career away from NJ. i think it drives the point home better than the team shooting percentages over that time, but it’s a minor point, i admit.
richer explained it above but you don’t seem to understand how unlikely it would be for the devils to shoot even 6% as a team for an entire season.
Huh? I simply asked if that was also about their shooting percentage in the playoffs?
Playing Devils' advocate since 1982.
Depends
Gabe Desjardens actually compared the expected shooting percentage based on location to what the first round teams shot in the playoffs. Here’s the Devils numbers:
Actual even strength S%: 2.67%
Actual power play S%: 6.67%
In both cases, the Devils shot well under expectations based on location. The shooting percentage of 2.67% at even strength hockey was especially putrid.
For what it’s worth, the Devils took a total of 2,459 shots last season. If they shot at 4.9% for that season, then they’d score only 120.491 goals. It would take a lot for any NHL team to put up so few goals in 82 games.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 19, 2010 10:50 PM EDT up reply actions
Seems to me that if the “expected” numbers aren’t synching up with reality, it’s not due to “bad luck” but that the expectations were wrong.
Bad luck is a cop-out excuse for why predictive stats don’t accurately predict the outcome.
Playing Devils' advocate since 1982.
The expected numbers are based on shot location data Gabe’s been collecting for years; but he can attest to their meaning better than I.
That said, how is this a cop out excuse? Shooting percentage involves a lot of variables that the shooter can’t really control; bad luck is just as good explanation to refer to that for the shooter. If there was a more accurate & intuitive term for this, then fine. But it’s not a cop out at all.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 20, 2010 8:27 AM EDT via mobile up reply actions
how is this a cop out excuse?
Shooting percentage involves a lot of variables that the shooter can’t really control
Exactly.
Calling it bad luck is a cop-out because it’s implying that the calculated expectations are correct despite an incorrect rate of 100%, and that it’s the players that didn’t perform as they should… and not because of the countless variables involved in the complex act of shooting and scoring a goal, but simply because of some “bad luck.”
If no one meets the expectations, the expectations are wrong.
Playing Devils' advocate since 1982.
there are going to be huge variations in these numbers in 5 game samples. the point is that everyone was saying how the devils were taking outside shots and they got what they deserved, but when confronted with numbers like this, their tune changes (well, those numbers don’t mean anything, and variables aren’t being accounted for, and so forth). whether or not you want to attribute it to bad luck or some number of variables, your point remains that it has to do with VARIABLE conditions – i.e. chance events. a point most hockey fans don’t or won’t acknowledge.
It’s probably impossible to exactly hit the mean each and every time, elesias. Stuff varies, and it’s normal to observe this variation. Sometimes it varies a lot higher than normal, and sometimes a lot lower. For various reasons, it was lower. Point is, we can’t really look at those factors and guess when they’ll show up again, and what effect they’ll have.
by red army line on Oct 20, 2010 12:57 PM EDT up reply actions
Which is to say that for all the number gathering and discussion, there is about as much predictive value to these sorts of things as someone just making observational guesses.
Playing Devils' advocate since 1982.
Not really. You can get a pretty good estimate with the numbers, and over time they’ll converge to somewhere around the estimate (unless something outlier-ish happens). I mean, Puck Prospectus is already getting decent GVT projections with VUKOTA or whatever they call it.
by red army line on Oct 20, 2010 3:19 PM EDT up reply actions
Calling it bad luck is a cop-out because it’s implying that the calculated expectations are correct despite an incorrect rate of 100%, and that it’s the players that didn’t perform as they should… and not because of the countless variables involved in the complex act of shooting and scoring a goal, but simply because of some "bad luck."
These two posts by Gabe Desjardens at Behind the Net blog will be of your interest. The first establishes that the major factor in scoring goals is transient skill (a.k.a. uncaptured performance, luck) and the second breaks down the talents involved on scoring on an initial shot at even strength. Gabe concluded that on a given shot, luck is the dominant factor. It’s over the course of a season, then talent is the dominant factor.
Given that the Devils’ problems are scoring in the short-term, the issue isn’t talent. That’s why I (and others) are attributing rotten luck to how the Devils keep running into strong goaltenders, pucks aren’t bouncing right in, etc, etc, etc despite bombing netminders with all sorts of shots.
That’s not a cop-out, it’s a frustrating reality.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 20, 2010 5:13 PM EDT up reply actions
These things all rely on “expected” statistics based on flawed data and a flawed concept: that because it was so in the past, it must be in the future.
Which is my point. “Expecting” certain outcomes that are impossible to rightfully expect, and saying that the reality isn’t meeting the statistical expectations because of “luck” is a cop-out. It’s not matching up because the expectations are wrong. Simple as that.
Playing Devils' advocate since 1982.
So it’s impossible for something unforseen to happen? All the statistics give are the most likely scenario. All the hockey stats guys always stress that there is a heavy uncertainty factor (which they call “luck” for shorthand). I don’t know, are you hung up on the semantics? Wouldn’t be the first time I’ve come across someone like that.
by red army line on Oct 22, 2010 11:19 AM EDT up reply actions
No, that’s not what I’m saying at all. Actually, I’m saying just the opposite: that the unforeseen is bound to happen.
Which is why I don’t expect that statistical predictions will ever be anything more than semi-educated guesses.
I’d be curious to see how, if at all, advances in the attention paid to statistics in hockey has affected something like the betting in Vegas. It stands to reason that if advances are being made and these projected expectations are becoming more accurate, then if anyone at all were to be taking advantage of them it would be the casinos et al that have real money on the line.
Perhaps it is just a semantics thing. It cheapens the actual factors involved to pass it off as luck… factors that those without their heads just in the numbers can take into account.
For instance, five years ago the Redskins (forgive the football reference, but it’s the only example I can think off of the top of my head, and I expect that you can appreciate the team involved) were coming off of a 52 point offensive explosion against the 49ers and facing the defensively struggling NY Giants. I don’t know what the exact spread was, but the Skins were pretty favored and had every reason to be confident going in… except that the Giants owner, Wellington Mara, passed away during the week and I knew at that moment that the Skins had no chance against an emotionally charged Giants team. One of the best offenses in the League up to that point ended up losing 36-0.
In the stat world that’s “an anomaly” or “luck,” but it was neither when one looks at things realistically.
Playing Devils' advocate since 1982.
I’d be curious to see how, if at all, advances in the attention paid to statistics in hockey has affected something like the betting in Vegas. It stands to reason that if advances are being made and these projected expectations are becoming more accurate, then if anyone at all were to be taking advantage of them it would be the casinos et al that have real money on the line.
people have done it in baseball and basketball. i don’t know that anyone’s done it for hockey because hockey statistics are still way behind those sports.
For instance, five years ago the Redskins (forgive the football reference, but it’s the only example I can think off of the top of my head, and I expect that you can appreciate the team involved) were coming off of a 52 point offensive explosion against the 49ers and facing the defensively struggling NY Giants. I don’t know what the exact spread was, but the Skins were pretty favored and had every reason to be confident going in… except that the Giants owner, Wellington Mara, passed away during the week and I knew at that moment that the Skins had no chance against an emotionally charged Giants team. One of the best offenses in the League up to that point ended up losing 36-0.
this is a huge case of selection bias. there are lots of instances where teams ‘had to’ win games because of someone dying or something, and yet they lost. this, however, rarely gets mentioned, because it doesn’t fit in with a journalist’s narrative. we don’t know how things are going to affect people.
Perhaps it is just a semantics thing. It cheapens the actual factors involved to pass it off as luck… factors that those without their heads just in the numbers can take into account.
which factors are these, and how do they project forward? or are they just factors in hindsight, predictably loaded with confirmation bias?
people have done it in baseball and basketball. i don’t know that anyone’s done it for hockey because hockey statistics are still way behind those sports.
And what were the results? They’re different beasts, but I’d still be curious to see.
this is a huge case of selection bias
It’s a case of my providing an example to demonstrate that the numbers don’t always tell the story, and relying on the past to predict the future doesn’t account for major outside influences… but instead passes off the contradictory results as “luck.” But you knew that and omitted the final sentence in my post on purpose because it doesn’t support your narrative.
which factors are these, and how do they project forward? or are they just factors in hindsight, predictably loaded with confirmation bias?
So, if I’m understanding correctly, you’re accusing me of retroactively pointing to bogus outside influences to explain why the reality of the results isn’t synching up with the statistical expectations?
Playing Devils' advocate since 1982.
And what were the results? They’re different beasts, but I’d still be curious to see.
winning sports bettors tend not to give out results or spell out what they are doing, because A: people will copy them, therefore reducing their edge and B: sportsbooks may catch wind of it and deny them action.
It’s a case of my providing an example to demonstrate that the numbers don’t always tell the story, and relying on the past to predict the future doesn’t account for major outside influences… but instead passes off the contradictory results as "luck." But you knew that and omitted the final sentence in my post on purpose because it doesn’t support your narrative.
the problem is, did the giants continue winning? how can we attribute the win to this one thing? try to use your system in the future and see how good you are with it. if you were so sure, did you put money on the game?
So, if I’m understanding correctly, you’re accusing me of retroactively pointing to bogus outside influences to explain why the reality of the results isn’t synching up with the statistical expectations?
no – i’m asking you to do it again. my suspicion is that similar to your wellington mara story, you ‘knew’ the devils were going to lose to the flyers and are therefore fighting real hard for your ‘knowledge’ – luck be damned, you had that one called. so, just do it again as the season goes along, and in the playoffs. one thing this entire discussion is missing is how other teams did according to expected shooting percentage – the statistic doesn’t even come close to explaining how goals are scored, or how many goals we should exactly expect to be scored. there is a HUGE chance element in hockey – teams have not really shown the ability to shoot above these averages consistently from year-to-year. when we’re talking about a few games, just about anything can happen – like i said, most hockey teams win between 40% and 60% of their games. still, making ‘educated guesses’ is better than making guesses.
winning sports bettors tend not to give out results or spell out what they are doing, because A: people will copy them, therefore reducing their edge and B: sportsbooks may catch wind of it and deny them action.
I’m not asking for the secret formula of winning sports bettors. I was inquiring into the possibility that someone can show that advances in sports statistics has improved the ‘house advantage’ of a group we’d expect to be on the leading edge of such things… the casinos.
the problem is, did the giants continue winning?
They went 6 and 3 after that and won the division, but that’s not really relevant.
how can we attribute the win to this one thing?
We can’t attribute it to any one thing, but it was a major influence on the events in that game… far more than “luck.”
Contrarily, can you prove that it wasn’t? Or that the Redskins’ run at the end of 2007 wasn’t directly related to the murder of Sean Taylor? Or that Favre’s tremendous game just days after his father’s passing wasn’t influenced at all by that event?
try to use your system in the future and see how good you are with it.
What system? Reality? Understanding that athletes are human beings and respond to emotional influences?
It’s not a system, nor is it a predictive measure. I never claimed it was except to say that I knew at the time that the Giants would respond with an emotional game and the Redskins would not be able to match their intensity and would therefore lose.
The example was solely to demonstrate that there are a wide variety of influences on athletes from day to day and week to week, and why I feel calling their not meeting statistical expectations “luck” cheapens the actual factors involved.
if you were so sure, did you put money on the game?
No, I don’t gamble.
no – i’m asking you to do it again.
Fine, I’ll be happy to describe ahead of time how the Giants will play with nearly unmatchable intensity in the game immediately following the death of Wellington Mara next time it happens.
my suspicion is that similar to your wellington mara story, you ‘knew’ the devils were going to lose to the flyers and are therefore fighting real hard for your ‘knowledge’ – luck be damned, you had that one called. so, just do it again as the season goes along, and in the playoffs.
No, actually. I thought the Devils would win going in, their regular season record against the Flyers aside… though by the end of Game 3 I fully expected them to lose the series. Does that count?
making ‘educated guesses’ is better than making guesses.
Depends on your definition of an educated guess, but I’d like to see some evidence of this claim.
Playing Devils' advocate since 1982.
I’m not asking for the secret formula of winning sports bettors. I was inquiring into the possibility that someone can show that advances in sports statistics has improved the ‘house advantage’ of a group we’d expect to be on the leading edge of such things… the casinos.
they’re even more protective of information than winning bettors. there’s no doubt that they aren’t just creating betting lines off intuition.
Contrarily, can you prove that it wasn’t? Or that the Redskins’ run at the end of 2007 wasn’t directly related to the murder of Sean Taylor? Or that Favre’s tremendous game just days after his father’s passing wasn’t influenced at all by that event?
no, of course not, and it’s harder for me to remember contrary examples because they fall outside the narrative we’re given. but i suppose it’s not really that relevant.
Depends on your definition of an educated guess, but I’d like to see some evidence of this claim.
it’s impossible, because you will just counter-claim counter-examples. we’d need a huge amount of data, and also a ‘guesser’ to compare it against.
they’re even more protective of information than winning bettors. there’s no doubt that they aren’t just creating betting lines off intuition.
I’m not asking for their super-secret formulas, just if anyone has done any kind of study concerning the impact of advanced stats on betting.
In other words, if advanced stats can be properly predictive, and if advances in the stats are being made, then the casinos who set the betting lines should be at the vanguard of such things and, one would think, one should be able to show an appreciable increase in the house edge to coincide with the effectiveness of the data.
I know casinos are secretive, but things like the house advantage percentages are pretty well-known.
Playing Devils' advocate since 1982.
i have no idea whether casinos use advanced stats that they’ve stolen from the internet; i think it more likely that they have in-house stuff. most of what advanced stats end up saying, they have little or no interest in, and they also have to take into account market conditions when setting lines.
Once again you demonstrate why I dislike “conversing” with you.
Playing Devils' advocate since 1982.
i don’t even have any idea how anyone would undertake such a study. how does one ‘know’ when the advanced stats are being used and when they’re not? how does one filter out a definite increase in ROI from mere chance? how would one separate the fact that the casinos might be getting smarter with the fact that there are more sharp bettors out there with the increase in precise information?
once again you demonstrate why I dislike ‘conversing’ with you.
Hey, after misrepresenting what I was saying more than once in an unrelated to the discussion at hand, thinking-aloud moment, please feel free, once you’re forced to actually acknowledge what I actually said and not what you inexplicably interpreted me to have said, to dismiss it because you don’t know how it could be done.
For future reference, if you dislike conversing with someone you may want to try not replying to their posts instead of responding to them with your typical condescending, superior tone.
Playing Devils' advocate since 1982.
i’m sorry i couldn’t parse your stumbling question. yes, that was merely a curiosity of yours and not a demand. i should have said in response ‘i don’t know of anything like that, nor do i see how it could be done or why it would be done, but feel free to look for it yourself.’
For future reference, if you dislike conversing with someone you may want to try not replying to their posts instead of responding to them with your typical condescending, superior tone.
i suspect you should take your own advice.
If you both can’t play nice, then you both won’t play at all.
Cease the biting.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 23, 2010 1:16 PM EDT up reply actions
No, that’s not what I’m saying at all. Actually, I’m saying just the opposite: that the unforeseen is bound to happen.
Sometimes things turn out almost exactly as we’d think based on statistical guesses—take this year’s SCF for example.
Then why are you rapping on the usefulness of stats? They’re still the best predictors we have. Long term, results still tend to converge around statistically-estimated means, regardless of how they predict individual games. Is one season long term? Kind of is. It’s feasible to overshoot and undershoot, but not by much. Is it feasible to undershoot at a team level by ~5% over a month’s worth of games? I don’t think I’ve seen it happen.
by red army line on Oct 22, 2010 1:06 PM EDT up reply actions
Sometimes things turn out almost exactly as we’d think based on statistical guesses—take this year’s SCF for example.
I’m not familiar with the reference.
Regardless, one example doesn’t exactly make a compelling argument for their viability… are there more examples?
They’re still the best predictors we have.
I’ve yet to see this conclusively shown to be the case.
Besides, that’s like saying that Uncle Ron is the best babysitter we have. That may be true, but it doesn’t mean I’m leaving my kids with Uncle Ron.
Long term, results still tend to converge around statistically-estimated means, regardless of how they predict individual games.
I don’t deny this. I largely agree with this.
For things like determining whether or not to sign a player to a contract, past performance statistics are a boon and as much a part of the process as current, on-ice evaluations, I’m sure.
For things like saying this team will win or that player will score that… not so much.
It’s feasible to overshoot and undershoot, but not by much. Is it feasible to undershoot at a team level by ~5% over a month’s worth of games? I don’t think I’ve seen it happen.
I’m not hung up on the shooting percentages. In fact, I fully agree with you that the Devils will not continue to shoot so abysmally. My issue, originally, was that the term ‘regression to the mean’ is bandied about as an unalterable truth. Just because Player X shot at Y% over the past Z years, and he is not shooting at Y% now, it does not mean that he will regress back to Y%. All things being equal, it’s likely, but not all things are equal.
Playing Devils' advocate since 1982.
My issue, originally, was that the term ‘regression to the mean’ is bandied about as an unalterable truth.
ok, since that WAS the title of my post, I feel the need to dive back in.
IN THIS CASE, given the sample size we have and the stability of the numbers, AND given a statistical simulation based on those numbers, AND given that the Devils’ offensive talent (as demonstrated by career S%) is at least as good this year as last, that one would EXPECT the Devils’ S% over the rest of the season to regress back towards the established mean. It’s not a certainty, or an ‘inalterable truth’ – it’s just what you would expect given the data. In fact, based on a sample size that large – even accounting for dead executives and flu outbreaks and full moons – you can say with at least 95% statistical certainty that the Devils’ S% in their first 6 games was an aberration, NOT a new mean. That’s what the math can tell us.
And clearly I’m not the only one who thinks this way, because both Lou and Johnny Mac said on different occasions that ‘if we keep getting chances they’ll eventually go in,’ which is a numbers-free way of saying ‘this is an aberration, not a new mean.’
It’s hard not to resent being used as your straw-man, even implicitly – I took care to keep my conclusions within the scope of a single variable, and if the numbers were noisier I wouldn’t have stated my conclusion so confidently – hardly ‘bandying about.’ I don’t know if there are any fans out there who scream ‘REGRESS TO THE MEAN!’ every time the Devils get scored on, but I assure you that I’m not one of them.
by richer44 on Oct 22, 2010 5:18 PM EDT up reply actions 1 recs
I wasn’t targeting you or your usage of the term. Call it the straw that broke the camel’s back.
It’s hard not to resent being used as your straw-man, even implicitly
I apologize if that is how you feel it turned out, but that was neither my original intent nor how I feel the discussion evolved.
I’m not disagreeing with your assessment. It doesn’t take a mathematician to see that the Devils are going to score more goals than they have been.
I do disagree, however, with the concept of regression to the mean as often used to mean that because it was, it must be again. That just because there has been some statistical stability over a period of years that this season will be also.
Playing Devils' advocate since 1982.
These things all rely on "expected" statistics based on flawed data and a flawed concept: that because it was so in the past, it must be in the future.
that’s not what’s happening. what people who write about statistics expect you to understand is that there is a range of possible outcomes. when people project, e.g. a player’s batting average for the next season, they will give a specific number, like .278. that’s just the most likely outcome, but most likely here is very tiny – there’s a giant range which that player’s batting average can be expected to fall under (probably something like .230-.320 99% of the time). likewise with gabe’s shooting percentage project. gabe has shooting location data from the last several years, and he’s used that to find the shooting percentage from every point in the offensive zone. when he says the devils’ expected shooting percentage is 6.whatever, that’s just the most likely outcome out of a huge range of outcomes.
so when gabe says that the devils shot well below expectation given the devils’ shooting locations, he’s making a factual statement. whether you want to attribute that to chance or ‘bad play’ is up to you.
Expecting" certain outcomes that are impossible to rightfully expect, and saying that the reality isn’t meeting the statistical expectations because of "luck" is a cop-out. It’s not matching up because the expectations are wrong.
so what should the expectations be? the devils aren’t shooting well, therefore we should expect them to continue shooting not well? should we expect them to shoot 2.9% at even strength? do you think that’s a reasonable thing to expect and to be able to predict, even though no team has even come close to shooting that in a given season? please tell me about your systematic expectations.
Bad luck is a cop-out excuse for why predictive stats don’t accurately predict the outcome.
You can’t buy that a goalie got hot, Kovy rang a couple of layups off the pipe, Rolston was feeling a bit sick and didn’t have the usual zip on his shots, etc etc? I sure can.
We can come up with explanations after the fact. Not so much before. All we have for predictives are records, goal differential, and Corsi. That’s it.
by red army line on Oct 20, 2010 12:55 PM EDT up reply actions
I can buy it, absolutely, but it’s a retroactively applied excuse when things don’t turn out they way the stats suggested they should.
It’s common sense that those variables will affect the outcome, which is why I don’t put much stock into the predictive nature of statistics that ignore such variables.
Playing Devils' advocate since 1982.
I can buy it, absolutely, but it’s a retroactively applied excuse when things don’t turn out they way the stats suggested they should. /blockquote>
That’s why most hockey statisticians don’t even bother.
It’s common sense that those variables will affect the outcome, which is why I don’t put much stock into the predictive nature of statistics that ignore such variables.If the variables are so unpredictable to being with, what’s the point of…anything?
We don’t have the computing power to predict when Brodeur gets the sniffles, or when Pavelec suddenly collapses, or when Karl Alzner starts seeing flashes in front of his eyes. I’m not even sure the laws of quantum mechanics allow us to predict these things anyways.
We do have the computing power to predict the winner in more than half of NHL games. Sure, take it with a grain of salt, but don’t dismiss it. IMO, it’s more productive to explain why the stats won’t regress to the mean than to attack the credibility of the stats themselves. I’m pretty sure it’s not getting to the point where we blindly look at stats here.
by red army line on Oct 20, 2010 3:26 PM EDT up reply actions
Blockquote fail…first part and part in the middle should be quoted, rest not.
by red army line on Oct 20, 2010 3:27 PM EDT up reply actions
That’s not exactly a ringing endorsement—you can flip a coin and correctly predict the winner of a hockey game about 50% of the time, and if the predictor games are any measure, most people can accomplish at least that success rate with nothing but semi-informed guesses.
I’m not dismissing the credibility of stats, I’m dismissing the growing use of the term “regression to the mean” as some sort of unalterable truth that just because it was so in the past means it will be in the future.
I’m also dismissive of “bad luck” being an acceptable explanation for why stats used to predict the future are inaccurate. Random variables are a part of the game and always will be… therefore attempts to predict their outcomes on paper will always be thwarted by reality. If they weren’t, then they wouldn’t need to play the games.
Playing Devils' advocate since 1982.
red army said that corsi makes up 52% of winning, not that whoever has the highest corsi wins 52% of the time.
i will try my damnedest not to strawman here, but that last statement is puzzling on several levels.
I’m also dismissive of "bad luck" being an acceptable explanation for why stats used to predict the future are inaccurate.
This doesn’t make sense. A statistic is a record of something that has happened, not something that will happen. It may or may not correlate to future events. We can determine whether or not it does by taking large samples and seeing if it holds true (for instance, how often a home team wins a particular game – does having home ice correlate with winning? as it turns out, it does have a slight correlation with winning). so when you say “stats used to predict the future are inaccurate”, you’re not making a sensible statement – statistics don’t absolutely predict the future, and no one claimed that they did. they can, if used properly, predict the future better than mere chance. however, hockey teams tend to win between 40 and 60% of their games – meaning a lot of the time, the ‘better team’ is hard to find, and that often times, the ‘better team’ (i.e. the one with the better record coming into the game) won’t win the game.
Random variables are a part of the game and always will be… therefore attempts to predict their outcomes on paper will always be thwarted by reality.
no one is arguing this statement. absolutely no one has said that by making a prediction based on statistics, we have ‘predicted outcomes on paper’. all we have said is that it is more likely that if these predictions are based on strong foundations that they will predict the winner more often than mere chance. furthermore, most predictions in hockey that are statistically based aren’t just based on winning one game – they are based on trends and whether or not they will continue or reverse. this is because one game is an enormously chance-y event.
If they weren’t, then they wouldn’t need to play the games.
yes, obviously the games have to be played. the games have an enormous chance element that no statistics can properly describe. however, mere observation cannot winnow out these chance elements either. chance is an enormous part of a hockey game, and no one who uses statistics intelligently to discuss sports would ever deny the role that chance plays.
by Triumph44 on Oct 20, 2010 6:04 PM EDT up reply actions 1 recs
Rec’d. All the stats tell us is who is more likely to win the game. That advantage the favorite has over the underdog probably is never more lopsided than 60-40, if it even gets there. I think Hawerchuk ran the numbers and found that the MTL upset over Washington in 7 had about a 35-40% chance of happening anyways (winning in 7, not winning in 7 and coming back from a 3-1 deficit too). There’s so much parity in the NHL.
by red army line on Oct 21, 2010 1:47 PM EDT up reply actions
I remain unconvinced that “bad luck” is anything more than a cop-out reason to explain why things don’t work out the way they’re expected to based on previous results.
Playing Devils' advocate since 1982.
If we can agree that for various reasons the Devils are going to start shooting a lot better real soon, there is no issue here. The whole “luck” thing also encompasses things like strategy changes.
by red army line on Oct 22, 2010 11:21 AM EDT up reply actions
Ok. I think this whole thing has been semantics, pretty much. “Luck” unfortunately in the statistical sense encompasses everything from owner deaths to coaching adjustments. The stats suggest the Devils will do SOMETHING to get their sh% up. It may just be to keep plugging away in the same way, or minor changes, or huge changes.
by red army line on Oct 22, 2010 1:12 PM EDT up reply actions
The term intangibles seems more appropriate than luck, don’t you think?
Go Devils
Go Jets
Like sports betting? Sign up with centsports and do it for free
Depends whether I’m in the stats mood or non-stats mood.
by red army line on Oct 24, 2010 5:38 PM EDT up reply actions
That’s not exactly a ringing endorsement—you can flip a coin and correctly predict the winner of a hockey game about 50% of the time, and if the predictor games are any measure, most people can accomplish at least that success rate with nothing but semi-informed guesses.
I guess the way I put the idea before was kind of silly considering this point you bring up…like Triumph says, Corsi “explains” about 52% of winning (I think it’s a bit higher), which is the r squared. The r then is around 70%, which is quite significant.
My memory is a bit hazy on this, so I looked it up. The Puck Stops Here on Kuklas Korner got an r of 0.63 between team points and Corsi (<a href="http://www.google.com/url?sa=t&source=web&cd=3&ved=0CB0QFjAC&url=http%3A%2F%2Fwww.kuklaskorner.com%2Findex.php%2Fpsh%2Fcomments%2Fcorrelation_of_fenwick_number_with_team_points%2F&rct=j&q=hockey%20corsi%20winning&ei=V4HATO_WNpChOqrPpf4L&usg=AFQjCNFzHLypqj2vi-kLejIEuyZgWpdVow&sig2=7tl0C58LZdp7_gc_jCDSiw&cad=rja" target="new">link). Over ~1300 games in a season, that’s pretty good.
by red army line on Oct 21, 2010 2:08 PM EDT up reply actions
Try this link. The guy doesn’t use score-tied Corsi only, though, I don’t think, so in reality if you do predictions well you can do better than 63% or so.
by red army line on Oct 21, 2010 2:10 PM EDT up reply actions
Maybe. The point is that over a large enough sample we can’t expect these factors commonly denoted as part of “luck” to be sustainable. They have no predictive value.
Corsi% explains about 52% of winning (the r squared). Looks like the Devils have a big edge in that metric. I’d bet they start winning—a lot—really soon.
by red army line on Oct 20, 2010 12:49 PM EDT up reply actions
is that corsi with the score tied or corsi in general? a decent portion of the devils’ corsi has come from score effects.
Good point. I ran the numbers with score tied, I think. I only did one season (to do a quick check on whether Corsi% or Fenwick% correlates better to winning) but I’m pretty sure Vic Ferrari or someone used a bigger sample size and got similar results.
by red army line on Oct 20, 2010 3:28 PM EDT up reply actions
You could also make the arguement that if the opposition’s shooting percentage was even to the Devils, the Devils would have won the Washington game. Now obviously the Devils’ D was horrible in that game and the better team won that night, but the quality of chances against are a significant part of it. Marty hasn’t been great early on, but he wasn’ t brilliant early last season, either. He allowed 5 to PHI, 3 to NYR, 3 to TB, 2 to FLA, 2 to WSH and 4 to ATL. That’s 19 goals in 6 games. That’s not far off from his start this year, is it?
"Potential means you ain't done it yet" - Bill Parcells
By the way, the following six games he allowed 10 in the NEXT six games he played (during a 7-game span).
"Potential means you ain't done it yet" - Bill Parcells
by DownGoesAvery on Oct 19, 2010 4:22 PM EDT up reply actions
You could also make the arguement that if the opposition’s shooting percentage was even to the Devils, the Devils would have won the Washington game.
Well, right – that’s why it’s better to outshoot your opponent. Slight differences in team S% only play out after lots and lots of games, which doesn’t do you any good finding a goal on that particular night. The Caps shot close to 25% that night, which is preposterously high and a major statistical outlier. If you look at expected S% alone, the Capitals are absolutely not 5 goals better than the Devils – that’s just how it played out on this particular occasion. (Imagine if the Devils were outshot that night? Yeesh.)
In any case, I wanted to focus specifically on offense for the scope of this particular FanPost, so I left out any defensive variables. Just put a big ‘ALL THINGS BEING EQUAL’ at the top of the post as you read.
Well, you can add an “ALL THINGS BEING EQUAL” in the post where it matters.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 19, 2010 6:05 PM EDT up reply actions
I’ve long suspected the Devils are a ‘Sabermetric’ (to use a baseball term) hockey team and I imagine that they have hockey stats that would make the folks at ‘PuckProspectus.Com’ blush.
OK, enough Lou-Aid for now.
I was at the Colorado game Friday and in the first period of that game the Devils were all over the Avalanche but still down 1-0. Was it ‘luck’? I think it was ‘bad luck’. (Though I prefer the word ‘chance’ to ‘luck’) Kovalchuck had Anderson beat and the puck dribbled just wide. On another night it goes in. I believe that.
It’s an interesting analysis that gives hope but there is one flaw in it. If you shoot 20 times from the blue line the likelihood is that you won’t score any goals and your S% would be low comparing to if you have 20 two-on-ones when your S% should be significantly greater.
What would really interesting to see is how many legitimate scoring chances per game were generated comparing to other seasons but alas it’s a very subjective stat since who knows what does qualify for a legitimate scoring chance.
by jersey_devil17 on Oct 19, 2010 10:06 PM EDT reply actions
If you shoot 20 times from the blue line the likelihood is that you won’t score any goals and your S% would be low comparing to if you have 20 two-on-ones when your S% should be significantly greater.
That’s true, but it also assumes an even distribution, which I’m pretty sure is incorrect – you tend not to take as many shots from the blue line because it’s not a high-scoring area, and vice-versa for shots in close. Without consistent chance/location data, I’m just going to assume it all evens out by the time you’ve taken nearly 2500 shots over the course of a season.
What would really interesting to see is how many legitimate scoring chances per game were generated comparing to other seasons but alas it’s a very subjective stat since who knows what does qualify for a legitimate scoring chance.
This is do-able. Incredibly time-consuming but do-able if one has a lot of time, patience, and the ability to watch full games, pause them, and cross-check who is and isn’t on the ice. A number of people are counting scoring chances for teams throughout the season. Derek Zona of the Copper & Blue is doing it for Florida incidentally at Litter Box Cats; this report explains what he’s doing and shows off how to report the data.
Is it subjective? Sure, but that just means it’s important to define what a scoring chance is and stick to that definition. Am I going to do it? No. Would it tell us more about the Devils finishing chances? Sure, but it would be far more useful in comparison to total shots on net and to see whether the Devils are out-chancing the opposition as well as just out-shooting them.
Devils in my heart! Devils in my mind! Devils in my eyes! Devils until I die!
In Lou We Trust - The New Jersey Devils SBN Blog
by John Fischer on Oct 19, 2010 11:20 PM EDT up reply actions
With the number of people tracking, you should be able to get 15-20 games worth of Devils data. Rangers, Leafs, Panthers, Caps, Oilers, Flames are all being tracked…I forget if there’s anyone else.
As for the subjectivity of it all, having done this now for 4 games, and seeing how odd the shot data that NHL tracks is, I don’t think it’s any worse than what is being taken as ‘official’ stats.
Blueshirt Banter - The Season is Over - I read it on Facebook
Twitter: RangerSmurf
by George E. Ays on Oct 20, 2010 6:43 PM EDT up reply actions
Too far out or not hitting the corners?
Is it possible to analyze the puck range and placement decisions the Devils are making in their shooting opportunities this season versus aan equivalent early period in the last 2-3 seasons? Two thoughts … are we shooting too early (nerves / pressure?) giving an easier save, or in like for like distance to the net, are we not placing it in the corners / 5-hole compared to earlier seasons?
Email Gabriel Desjardins. He has that info. It’s notoriously poorly recorded, though.
by red army line on Oct 20, 2010 12:59 PM EDT up reply actions
It's not time to panic...
but while Kovalchuk does have a large disparity from his career shooting percentage to his current % remember that he only shot 9% in his 27 games with the Devils last year.
Time to worry? No, not yet but let’s keep an eye on this.
I am of the opinion that while there are plenty of shots that are generated that they are of a lower quality. I may check out the report John refers to above to see if we can track that here, because I just don’t see the opportunities that have been generated. Shots? Yeah. Opportunities? Not so much. I have witnessed way too many pucks hit goalies right in the chest to think they have simply had bad luck so far this season.
My .02, apologies if there aren’t enough statistical references in here but that is my subjective view.
I hate statistics, personally. took me far too long to understand it in college, and even longer when applied to my job. Confidence levels of performance and effectiveness make me cringe although I have to deal with it every day. It was all reduced to simple terms by my lovely wife, who, after being subjected to missing “her programs” in favor of the middle-of-the-season, really means very little in the grand scheme of things, Devil’s game, has joined me in person at 2 home games so far this season.
Her assessment? “Wow, you know, they’d score a lot more goals if they tried to shoot around the goalguy instead of through him.” I tried to explain to her that in the long run, you can expect about 1 out of every 11 recorded shots on goal equates to a goal. Give or take a shot or two on any given night, or for any given team. She followed that up with, “Wow, the Devils are going to score a lot of goals this season. Do you want me to write them a letter suggesting that they try to go around the other team’s goalguy?”
How could I not marry her?
I don’t hate them, used them a lot in college myself but I just think that sometimes they are used to state a fact instead of stating an eventual possibility. In baseball you can see the use of advanced stats as becoming a bit divisive when used by some of the analysts/fans in the way they talk about other analysts/fans points of view and negating a well reasoned, subjective argument in a snarky, condescending tone.
Nice anecedote about your wife. I have a similar situation during the middle of the season justfying not DVRing of certain programs over others to see a Devils/Sabres game. Luckily Doc Emrick makes games exciting for any fan and Chico, while a homer, is entertaining. If nothing else, I converted my wife from being a fairweather Rangers fan to a hardcore Devils fans. If that is my legacy….so be it.
by Tom Stivali on Oct 20, 2010 11:54 AM EDT up reply actions
i think the devils have outchanced teams in every game besides the game against washington – and in that game, they were about even in chances until the 3rd. it’s not just pucks hitting chests – that can result in a scoring chance (on a rebound), or can be a scoring chance (if it was taken from a high-percentage area).
lou and john maclean agree with the above view – i don’t think they’re just blowing smoke.
as for your point down below, i think there’s a lot to be learned from subjective views in hockey. stats certainly won’t tell the entire story. as an example, tallinder is leading devils’ D in fenwick, but i’d be a little surprised if he’s leading them in net scoring chances – it feels like tallinder’s unit has been out there for a ton of scoring chances against. the problem with subjective views is that they are stories, and are therefore often loaded with selection bias and confirmation bias. the scoring chances project that john mentions above is actually an attempt to put subjectivity back into the equation because raw stats don’t tell us enough.
Vic Ferrari a while back looked at Edmonton’s Fenwick% and scoring chance%, and they correlated very strongly.
Behind the Net under “individual shooting” has shot distance.
by red army line on Oct 20, 2010 1:01 PM EDT up reply actions
over large samples, they are going to correlate, of that i have little doubt. over smaller samples, they may not.
Depends on how you count them. Knee high, Neil, and I have been getting 20+ scoring chances per team, when each team takes ~30 shots. That’s decent sample over 5 games, so it shouldn’t deviate all that much, I’d guess.
by red army line on Oct 20, 2010 3:29 PM EDT up reply actions
if you insist that is so – and i believe it to be, mostly – why are you bothering to count them at all?
Because it’s slightly more accurate? It can shut up the advanced stats haters. For all the “Caps were always shooting from the outside against MTL” people, WSH had more than 1.5x the SCs than MTL, which shut up that crowd pretty fast.
I imagine scoring chances work better than Corsi for true “shutdown” or “checking” players (like Sami Pahlsson).
I guess in the anomaly case every now and then (say, Colorado last year), they do deviate. I wouldn’t bet on SCs and Fenwick deviating all that much now for the Devils, though. Although, come to think of it, even 5% is huge in our case, and you probably have about 200-300 SCs already. I withdraw what I said before. 5% is just getting lucky a handful of different times at this point. Sorry, estimation fail earlier.
by red army line on Oct 21, 2010 2:00 PM EDT up reply actions
Comments For This Post Are Closed
Secondary Sidebar
User Tools
Related FanPosts on In Lou We Trust
FanPosts
Community blog posts and discussion.
Recommended FanPosts
Recent FanPosts
In Association With
Latest Puck Daddy - NHL from Yahoo! Sports
- Thursday’s Three Stars: Kings win eighth straight; Kopitar’s sweet breakaway deke
- King leads Kings to super-meta Game 3 victory over the Coyotes
- Coyotes Coach Dave Tippett on embellishment: ‘The game is turning a little dishonest’ (VIDEO)
- No Hanzal, no Aucoin for Coyotes in Game 3 vs. Kings; Hal Gill, pylon (Playoff Puck Previews)
- Tom Renney’s contract not renewed by Edmonton Oilers; who might replace him?
Latest stories and headlines from NHL.com
- King, Quick lead LA Kings to brink of Cup final with 2-1 win over Coyotes in Game 3
- King, Quick lead Kings to brink of Cup final with 2-1 win over Coyotes in Game 3
- Milano takes over NHL's Twitter feed for Game 3
- Scrivens gets shutout as Marlies blank Barons 5-0 in Game 1 of AHL West final
- Kostka scores twice as Admirals rout IceCaps 6-1 in Eastern Conference opener
