By The NumbersHockey Analytics... the Final Frontier. Explore strange new worlds, to seek out new algorithms, to boldly go where no one has gone before.

Just throwing out a broad topic that may or may not have legs.

One way to measure the distinction between critical goals (say, an overtime goal) and non-critical goals (say, the tenth goal in a 10-1 victory) would be to credit the goal (and assists) with the difference in win probability that results from the goal.

For instance (and with made up probabilities), an overtime goal might increase the chance of victory from 50% to 100%, and so one would credit the goal 0.5 V-I Goals.

On the other end of the coin, the tenth goal in a 10-1 victory might increase the chance of victory from 99.99995% to 99.99998%, and so one would credit the goal 0.00003 V-I Goals.

This might be a way to distinguish between players who get their points in different situations. My guess it that the players who rank highly in V-I Goals and V-I Points are similar to those who rank highly in (regular) Goals and Points, but I haven't done the legwork.

I know GWG isn't the same thing (in fact it's a borderline useless stat, but I digress), but there has been work done on players with high GWG numbers, and those do tend to be players who simply score a lot of goals.

I personally think score-tied values would be enough. It's documented, in terms of scoring chances and shot differentials, that there are scoring effects (the teams that are down tend to outshoot those in the lead due to strategic changes), so score tied would be the most fair method of measuring critical goals scored. I don't see any way a weighting system could be implemented based on chance of victory, since we have no legitimate way of determining such a value; there are too many variables at a given moment in a given game for us to begin to accurately describe such a thing. I know the bookies do a decent job of it, but they only have to do it well enough to put up odds that cover their bets; they can cover any inaccuracy by altering the spread to give themselves a buffer.

Wouldn't goals that give a team a 3 or 4 goal lead increase the chance of victory more than goals that give a team a 1 or 2 goal lead? A team is more likely to win with a 3 goal lead than a 1 goal lead, but I don't know if I'd call those goals more critical.

There aren't as many individual events to collect in hockey obviously, but I'm sure you could calculate the win probability added from a goal given the context of the game (2-2, 5-1, etc.), as well as the time of the goal (a tiebreaking goal with 2 seconds remaining adding more to WPA than a tiebreaking goal with 10 minutes remaining) and minus the reverse.

If you adhere to shot percentage being more a matter of luck than skill (I'm not convinced, but just saying) you could ignore goals altogether and just calculate the above for shots for and against. If your typical shot has a 9% chance of going in, you credit the shot with 9% of the WPA a goal would've provided at that time and context and the reverse for all shots against.

It would be interesting to see if such a statistic would follow a similar trend as it does for baseball where it's not very predictive ("clutch" doesn't exist) but can serve as a "storytelling" stat.

The problem is that there are very few situations in hockey were the next goal isn't of critical importance. Most games are within 2 goals most of the time so you're mostly likely just looking at some statistical noise rather than anything meaningful.

Its like the myth of making important saves. Nearly every save made in the NHL is important. Blowout situations where they aren't are scarce.

Wouldn't goals that give a team a 3 or 4 goal lead increase the chance of victory more than goals that give a team a 1 or 2 goal lead? A team is more likely to win with a 3 goal lead than a 1 goal lead, but I don't know if I'd call those goals more critical.

No. The biggest move towards winning a game is going from being tied to being a goal ahead. Goals that increase leads after that each have less and less value, because you already have a lead, and you might not need those goals at all.

A team being 3 goals ahead is more likely to win a game than a team 1 goal ahead. But the additional likelihood of winning that the first team earned by going from two goals to three goals is significantly less that the additional likelihood gained by going from tied to 1 ahead.

Just throwing out a broad topic that may or may not have legs.

One way to measure the distinction between critical goals (say, an overtime goal) and non-critical goals (say, the tenth goal in a 10-1 victory) would be to credit the goal (and assists) with the difference in win probability that results from the goal.

For instance (and with made up probabilities), an overtime goal might increase the chance of victory from 50% to 100%, and so one would credit the goal 0.5 V-I Goals.

On the other end of the coin, the tenth goal in a 10-1 victory might increase the chance of victory from 99.99995% to 99.99998%, and so one would credit the goal 0.00003 V-I Goals.

This might be a way to distinguish between players who get their points in different situations. My guess it that the players who rank highly in V-I Goals and V-I Points are similar to those who rank highly in (regular) Goals and Points, but I haven't done the legwork.

I like it.

I started fooling around with discounting goals for importance, but this has a more scientific basis. I just used random values that I thought seemed right.

I've had this idea for several years (I don't remember if I've ever written about it here). This would be the most objective and rigorous way of quantifying the importance of players' offensive contributions.

The problem is that, in addition to the score, there are two other essential variables: the time remaining in the game, and league scoring context.

For example, a goal that gives a team a 1-0 lead at 1:00 of the first period might be worth 0.15 (hypothethically increasing the win probability from 50% to 65%). However, a goal that gives a team a 1-0 lead at 19:00 of the third period might be worth 0.45 (hypothethically increasing the win probability from 50% to 95%). This method would require a rigorous analysis of the win probability added based on the score in the game and the time remaining.

A more subtle (but equally important) consideration is the scoring context. A 2-1 goal at 15:00 of the second period might would be worth a lot more in, say, 2002, compared to 1982, given that the former season had far fewer goals per game, thus making each goal more important to a team's chance of winning. (Another way of thinking about it: given that there was so little offense in 2002, it's far more likely for a team to hang on to a 2-1 lead with 5 minutes left in the second period, compared to the wide-open 1982).

I spent some time working on this perhaps 7-8 years ago but gave up quickly after realizing how much work was required (also, the relevant data was much harder to find than it likely is today).

The problem is that there are very few situations in hockey were the next goal isn't of critical importance. Most games are within 2 goals most of the time so you're mostly likely just looking at some statistical noise rather than anything meaningful.

Its like the myth of making important saves. Nearly every save made in the NHL is important. Blowout situations where they aren't are scarce.

The reality is a lot of times the "next goal" bit is psychological. Again, this is relevant to study to see if its truly present but it only matters if giving up that goal increases the rate of a subsequent goal against as opposed to prevailing scoring rate.

Instead awarding the GWG as it is now (i.e. 4th g in a 6-3 win)
Award it to the goal that puts the team out ahead for good in a win.

i.e. Team a wins: 3-2
Presently, its the scorer of the 3rd g that gets credited...

but if scoring went..
A: 1-0
A: 2-0
B: 2-1
A: 3-1
B: 3-2

Have the 1-0 goal as the GWG.

or

B: 0-1
A: 1-1
A: 2-1
A: 3-1
B: 3-2

Have the 2-1 goal as the GWG.

in a blowout..
A: jumps to a 4-0 lead
B: scores 2 late ones

Award the 1st goal A scored as the GWG.

Basically the goal the put the team ahead without the other team able to catch up gets awarded.
I think that's better than awarding the 3rd scorer in that final example above.

I wonder what that would change in the rankings.. if something more substantial appears with the stat.

PS: i just realized this forum existed YESTERDAY! Love it. How long has it been around??

Alex Burrows, 7
Henrik Sedin, 6
Daniel Sedin, 6
Chris Higgins, 4
Sami Salo, 3
Kevin Bieksa 2
Cody Hodgson, 2
Mason Raymond, 2
Manny Malhotra, 2
Andrew Ebbett, 2
Andrew Alberts, 2
Ryan Kesler, 1
Jannik Hansen 1
David Booth, 1
Aaron Rome, 1
Samuel Pahlsson, 1

4 players wiped off the GWG scoresheet:

Kesler -1 - Zero
Alberts -2 - Zero
Bieksa -2 - Zero
Booth -1 - Zero

2 added to the list:
Lappy +1
Edler +1

Another takes a big hit:
Burrows -4

Hodgson -1
Raymond adds 1, take away 2, -1

Henrik adds 4, 2 taken away, +2
Rome adds 3, loses 1, +2
Daniel adds 2, 1 taken away, +1

Salo +2
Higgens +2
Ebbett +1

3 times players would of had their GWGs wiped away, but they had multi goal games where they earned the GWG earlier in the game..

New GWG leaders:

Henrik 8
Daniel 7
Higgens 6
Salo 5
Burrows 3
Rome 3
Ebbett 3
Malhotra 2
Hodgson 1
Raymond 1
Hansen 1
Edler 1
Pahlsson 1
Lappy 1

A sample size of 1 team over 1 season isn't adequate obviously.. so to conclude cream rising would be premature... but i think there is difference we would see when the emphasis slides towards goals that put your team ahead and taking away the benefit of falling into a GWG with late mean nothing opposition goals...

Edit:
=========================
Did CBJ since they had the smallest sample size of GWG...

Quote:

Derick Brassard 3
R.J. Umberger 3
Nikita Nikitin 3
Antoine Vermette 3
Ryan Johansen 3
Rick Nash 2
James Wisniewsk 2
Derek MacKenzie 2
Jeff Carter 1
Derek Dorsett 1
Jack Johnson 1
Grant Clitsome 1

R.J. Umberger 4
Ryan Johansen 3
Rick Nash 3
Nikita Nikitin 2
Antoine Vermette 2
Vinny Prospal 2
James Wisniewsk 2
Derick Brassard 1
Derek MacKenzie 1
Jeff Carter 1
Derek Dorsett 1
Jack Johnson 1
Grant Clitsome 1
Letestu 1

The ‘CIRV’ Stat (Comeback Impact Ratio Value) and Distribution Ratios

Ezra took a look at the numbers for four team (Jets, Leafs, Oilers and Capitals)

Quote:

At the beginning of last season, my father John Ginsburg (a Math Professor at the University of Winnipeg for 32 years, now retired) was discussing who he thought would have the most goals and points for the Winnipeg Jets in the upcoming season. He then began discussing whether or not a statistic existed that measured the impact a goal/assist/point had on a game.

For example. Evander Kane scored 30 goals and 27 assists last season but what do those 30 goals and 27 assists mean as far as having an impact on a game? The reason why this is so important is because when a player scores a goal is undeniably more important than the goal itself. For further proof, one must only look at what has happened recently with 40-year-old designated hitter Raul Ibanez of the New York Yankees. Ibanez has 19 homers this season, 10 of which have tied the game or put or his team in the lead. Now that’s impact.

Lets me first explain the CIRV Statistic and the other distribution ratios. A distribution value refers to the distribution of a player’s goals, assists and points based on the three different stages of a game: when your team is ahead, when your team is tied and when your team is behind. The CIRV or “Comback Impact Ratio Value” is specifically what percentage of a player’s points occur when his team was behind.

Without having yet read the article, the issue with this sort of analysis is that you not only have to demonstrate that what a player did had more impact in the game, but to draw any sort of conclusion from it you also have to demonstrate that a player's success, or lack thereof, in such a metric reflects something about the player rather than the player's opportunity or simple variance.

That is, I haven't yet seen anything to demonstrate that a player's apparent clutchiness is persistent, that is represents something about the player rather than statistical noise.

I've had this idea for several years (I don't remember if I've ever written about it here). This would be the most objective and rigorous way of quantifying the importance of players' offensive contributions.

The problem is that, in addition to the score, there are two other essential variables: the time remaining in the game, and league scoring context.

For example, a goal that gives a team a 1-0 lead at 1:00 of the first period might be worth 0.15 (hypothethically increasing the win probability from 50% to 65%). However, a goal that gives a team a 1-0 lead at 19:00 of the third period might be worth 0.45 (hypothethically increasing the win probability from 50% to 95%). This method would require a rigorous analysis of the win probability added based on the score in the game and the time remaining.

A more subtle (but equally important) consideration is the scoring context. A 2-1 goal at 15:00 of the second period might would be worth a lot more in, say, 2002, compared to 1982, given that the former season had far fewer goals per game, thus making each goal more important to a team's chance of winning. (Another way of thinking about it: given that there was so little offense in 2002, it's far more likely for a team to hang on to a 2-1 lead with 5 minutes left in the second period, compared to the wide-open 1982).

I spent some time working on this perhaps 7-8 years ago but gave up quickly after realizing how much work was required (also, the relevant data was much harder to find than it likely is today).

I agree that this would be the best approach in assigning importance to players' goals/points. If the data is readily available in proper format, it seems very possible. I would guess there exists the proper formula to calculate win% based on score and time remaining, given the avg. scoring rate for the league. IIRC there was at least one link to a logarithmic formula in the "do teams pull their goalie soon enough?" thread and/or the "ideas for future studies" sticky thread (both in this forum), although I didn't delve into those links, so I am not certain of their applicability to this question.

One complication since the lockout is the shootout, and there was an extra point for OT games for a while even before the SO existed. This would seem to give greater importance to a tying goal than a goal which puts a team in the lead, an unfortunate side effect of a poorly designed team point structure by the NHL.