Quantifying the clutchness of a goal

[seventieslord]

This is something I thought of the other day as a study that has to be done, and then I immediately assumed it has already been done, since it's so relatively easy.

There's a perception that some players have more or less of a propensity to be a part of more important goals more often. I think that rudimentary attempts at quantifying this have been highly inadequate. The most basic is the GWG statistic, which is so flawed that I don't really need to explain why to this crowd.

From there, there have been attempts to credit players for their contributions to winning and tying goals, which makes sense fundamentally, because those are the goals that make a difference in the standings. I remember the original 1987 Hockey Compendium tried such a thing.

In such a low-scoring NHL, though, just about every goal is important. A case could be made that every goal scored when the game is within three goals is at least somewhat important (maybe even four!). But are they all equally important? No.

If you were down two and you tie the game, that tying goal is very important, but is it much more important than the goal that made it within one? If you are ahead by a goal midway through, and you score again, that goal greatly increased your likelihood of holding onto the lead, but it is not a "game winning" goal... unless, of course, you get scored on one more time. And so on.

We could assign arbitrary values to some goals, like on a 5 to 1 scale, where a 5 is an OT goal, a 4 is a tiebreaking goal late in the 3rd, a 3 is a tying goal, etc. While that would be making an earnest attempt to quantify a goal's importance, it would still be arbitrary.

But I believe we have the data available (going back quite a few years now) to quantify the "clutchness" or "importance" of every goal scored. I think the most important thing to consider is, what was the expected number of points your team would receive in the situation it was in prior to the goal, and what was the expected number of points following it?

For example: Your team is down a goal with 1:00 left in the 3rd. There's a 3% chance your team wins the game at this point. You score a goal. That makes it a coin flip from here. It's 50%. You just made your team's likelihood of a win rise by 47% with that goal.

Another example: It's midway through the 2nd and you're up 3-0. You're 98% likely to win this game. You score again. Your chances of winning just went up to 99.5. Your goal only made the likelihood of a win go up by 1.5%. It is not a very important goal.

Of course, that late tying goal could be rendered meaningless if the other team comes back and scores with 15 seconds remaining, but that doesn't change that the tying goal was very clutch. Similarly, the team down 4-0 could in fact come back to tie it, making that 4-0 goal very important, but that doesn't change that it was very unimportant when it was scored.

It would not be difficult at all for someone with the knowhow, to scour game sheets and create a huge data set that calculates the average number of points earned by a team in a certain situation.

i.e.:

- down by one goal at midway point: 0.67 pts on average
- down by one goal at end of 2nd: 0.55 pts
- down by one goal 14:00 into 3rd: 0.26 pts
- down by one goal 19:00 into 3rd: 0.08 pts

Using the data from that season, you could then definitively state the importance of each goal by the rise in expected points that goal caused. You could then add up any individual player's contributions to important points.

Thoughts? (this has to already be done before, right? I'm not known for my originality)

Click to read more...

 

[Doctor No]

I've started something like this on a few occasions - pretty much exactly as you've described. I should even have some Excel frameworks around somewhere if it would prove helpful (although they weren't particularly hard to set up, so probably not helpful).

The win probability framework (or in pro hockey nomenclature, points expectation) is definitely the way to go.

Ultimately, I kept losing interest because I have an irrational interest in goalies, and I couldn't do the thing that I really wanted to do (incorporate saves - which I'm sure the timing of is available but I didn't want to hunt for it then).

 

[morehockeystats]

This is something I thought of the other day as a study that has to be done, and then I immediately assumed it has already been done, since it's so relatively easy.

There's a perception that some players have more or less of a propensity to be a part of more important goals more often. I think that rudimentary attempts at quantifying this have been highly inadequate. The most basic is the GWG statistic, which is so flawed that I don't really need to explain why to this crowd.

From there, there have been attempts to credit players for their contributions to winning and tying goals, which makes sense fundamentally, because those are the goals that make a difference in the standings. I remember the original 1987 Hockey Compendium tried such a thing.

In such a low-scoring NHL, though, just about every goal is important. A case could be made that every goal scored when the game is within three goals is at least somewhat important (maybe even four!). But are they all equally important? No.

If you were down two and you tie the game, that tying goal is very important, but is it much more important than the goal that made it within one? If you are ahead by a goal midway through, and you score again, that goal greatly increased your likelihood of holding onto the lead, but it is not a "game winning" goal... unless, of course, you get scored on one more time. And so on.

We could assign arbitrary values to some goals, like on a 5 to 1 scale, where a 5 is an OT goal, a 4 is a tiebreaking goal late in the 3rd, a 3 is a tying goal, etc. While that would be making an earnest attempt to quantify a goal's importance, it would still be arbitrary.

But I believe we have the data available (going back quite a few years now) to quantify the "clutchness" or "importance" of every goal scored. I think the most important thing to consider is, what was the expected number of points your team would receive in the situation it was in prior to the goal, and what was the expected number of points following it?

For example: Your team is down a goal with 1:00 left in the 3rd. There's a 3% chance your team wins the game at this point. You score a goal. That makes it a coin flip from here. It's 50%. You just made your team's likelihood of a win rise by 47% with that goal.

Another example: It's midway through the 2nd and you're up 3-0. You're 98% likely to win this game. You score again. Your chances of winning just went up to 99.5. Your goal only made the likelihood of a win go up by 1.5%. It is not a very important goal.

Of course, that late tying goal could be rendered meaningless if the other team comes back and scores with 15 seconds remaining, but that doesn't change that the tying goal was very clutch. Similarly, the team down 4-0 could in fact come back to tie it, making that 4-0 goal very important, but that doesn't change that it was very unimportant when it was scored.

It would not be difficult at all for someone with the knowhow, to scour game sheets and create a huge data set that calculates the average number of points earned by a team in a certain situation.

i.e.:

- down by one goal at midway point: 0.67 pts on average
- down by one goal at end of 2nd: 0.55 pts
- down by one goal 14:00 into 3rd: 0.26 pts
- down by one goal 19:00 into 3rd: 0.08 pts

Using the data from that season, you could then definitively state the importance of each goal by the rise in expected points that goal caused. You could then add up any individual player's contributions to important points.

Thoughts? (this has to already be done before, right? I'm not known for my originality)

Have you seen my page on the Clutch?
http://morehockeystats.com/players/clutch

 

[Canadiens1958]

First Goal

Overpass started a great thread.

http://hfboards.mandatory.com/showthread.php?t=2159233&highlight=overpass+first+goal

Going back to the first recorded game in history you cannot dispute the importance of the first goal of the game, how it defines the future of the event, how it reflects the value and importance of each goal that follows.

Any other starting point guaranteees an inadequate result that may be interesting.

 

[seventieslord]

Have you seen my page on the Clutch?
http://morehockeystats.com/players/clutch

Good work. Definitly leaves me wanting, though. Basically the same thing I described in the OP:

We could assign arbitrary values to some goals, like on a 5 to 1 scale, where a 5 is an OT goal, a 4 is a tiebreaking goal late in the 3rd, a 3 is a tying goal, etc. While that would be making an earnest attempt to quantify a goal's importance, it would still be arbitrary.

also, should include the work of the other players in producing that goal, not just the guy who took the shot.

Overpass started a great thread.

http://hfboards.mandatory.com/sh...ass+first+goal

Going back to the first recorded game in history you cannot dispute the importance of the first goal of the game, how it defines the future of the event, how it reflects the value and importance of each goal that follows.

Any other starting point guaranteees an inadequate result that may be interesting.

First goals are demonstrably important, because we have easily available statistics that show what the first goal does to a team's win likelihood. They're not the only goals that do, though, and not the most important ones, either. I'm interested in something full-blown, that looks at every goal individually.

 

[Canadiens1958]

First Goal

Good work. Definitly leaves me wanting, though. Basically the same thing I described in the OP:



also, should include the work of the other players in producing that goal, not just the guy who took the shot.



First goals are demonstrably important, because we have easily available statistics that show what the first goal does to a team's win likelihood. They're not the only goals that do, though, and not the most important ones, either. I'm interested in something full-blown, that looks at every goal individually.

First goals are the most important ones. Fundamental.

Twice as important as any other goal. 1-0 game is winnable while a 0-0 game is not. 1-0 means the opposition has to score at least once to tie and at least twice in a row during the game to win. Look at the incidence of a team scoring two goals in a row on a goalie to appreciate this. So the first goal is worth at least twice that of any goal. Makes the game winnable = 2points. A second goal at worst creates a tie = 1 point. The difficulty factor of scoring two goals in a row is roughly double that of a single goal.

You can define other goals any way you want. GWG, Clutch , etc all have multiple interpretations but there is only one first goal and it becomes the foundation fro the game.

 

[Doctor No]

First goals are the most important ones. Fundamental.

Twice as important as any other goal.

You're painting with too broad of a brush here.

 

[Canadiens1958]

Primer

You're painting with too broad of a brush here.

Exactly. How else do you suggest applying the primer?

Yet without the primer there is nothing to support the fine and intricate strokes.

 

[seventieslord]

First goals are the most important ones. Fundamental.

Twice as important as any other goal. 1-0 game is winnable while a 0-0 game is not. 1-0 means the opposition has to score at least once to tie and at least twice in a row during the game to win. Look at the incidence of a team scoring two goals in a row on a goalie to appreciate this. So the first goal is worth at least twice that of any goal. Makes the game winnable = 2points. A second goal at worst creates a tie = 1 point. The difficulty factor of scoring two goals in a row is roughly double that of a single goal.

You can define other goals any way you want. GWG, Clutch , etc all have multiple interpretations but there is only one first goal and it becomes the foundation fro the game.

At the moment a game starts, the average team's expected points from a game are 1.15 (thanks to the loser point). Teams that scored first this season had a win% of .671, or 1.34 points per game. Therefore, a first goal is worth, on average, a 17% greater chance of winning the game, or 0.19 points in the standings (there are two ways it could be expressed).

While this is important, it makes no sense to attempt to claim it's more impactful than a 3rd period goal that ties the game or takes the lead... or an OT goal.

 

[Canadiens1958]

Historically

At the moment a game starts, the average team's expected points from a game are 1.15 (thanks to the loser point). Teams that scored first this season had a win% of .671, or 1.34 points per game. Therefore, a first goal is worth, on average, a 17% greater chance of winning the game, or 0.19 points in the standings (there are two ways it could be expressed).

While this is important, it makes no sense to attempt to claim it's more impactful than a 3rd period goal that ties the game or takes the lead... or an OT goal.

This past season would not reflect the NHL historically.

The impact of a 3rd period tying goal is measurable only in regards to the first goal.A previous goal or goals have to be scored to frame the tying goal. Likewise the OT goal only measurable in regards to the first goal. No first goal and both become the first goal only with greater weight and significance.

 

[seventieslord]

This past season would not reflect the NHL historically.

No, I'm sure it's a little different from season to season. I can't imagine there's any season you can point to where it's as important as you're making it out to be, though. The stats are all there on nhl.com - please show me a season that makes the case for first goals being significantly more important than I'm saying.

It's as simple as this - lower scoring leagues, first goal will be more important. Higher scoring leagues, less important.

The impact of a 3rd period tying goal is measurable only in regards to the first goal.A previous goal or goals have to be scored to frame the tying goal. Likewise the OT goal only measurable in regards to the first goal. No first goal and both become the first goal only with greater weight and significance.

The clutchness or importance of a goal has to be expressed in a way that reflects how it affected the game at that moment. Otherwise every single goal in a 4-3 game is treated as equally important, which is exactly the kind of thing we should try to get away from.

 

[Doctor No]

It's as simple as this - lower scoring leagues, first goal will be more important. Higher scoring leagues, less important.

Exactly. If the NHL ever switches to "first goal win", for instance, the first goal is very important.

Is the first basket in an NBA gane of critical importance?

 

[Canadiens1958]

Except

Exactly. If the NHL ever switches to "first goal win", for instance, the first goal is very important.

Is the first basket in an NBA gane of critical importance?

Except in the NBA or basketball do the stats reflect tying or winning baskets? No.

Also in the NHL you do not have a shot clock. The first basket is viewed differently as a result.Pre shot clock basketball may have viewed the first basket differently.

 

[Canadiens1958]

First Goal

No, I'm sure it's a little different from season to season. I can't imagine there's any season you can point to where it's as important as you're making it out to be, though. The stats are all there on nhl.com - please show me a season that makes the case for first goals being significantly more important than I'm saying.

It's as simple as this - lower scoring leagues, first goal will be more important. Higher scoring leagues, less important.



The clutchness or importance of a goal has to be expressed in a way that reflects how it affected the game at that moment. Otherwise every single goal in a 4-3 game is treated as equally important, which is exactly the kind of thing we should try to get away from.

So you admit that it is a question of just how important the first goal really is relative to scoring levels.

Easy to live with compromise. Move on.

 

[Doctor No]

Except in the NBA or basketball do the stats reflect tying or winning baskets? No.

Also in the NHL you do not have a shot clock. The first basket is viewed differently as a result.Pre shot clock basketball may have viewed the first basket differently.

Not the point.

 

[seventieslord]

So you admit that it is a question of just how important the first goal really is relative to scoring levels.

Easy to live with compromise. Move on.

Not the point.

The value of a first goal will fluctuate slightly as scoring levels do. It's still never as important as a late tying/winning goal. The NHL's goals per game would have to approach soccer levels for that to be the case.

 

[Canadiens1958]

Exactly

Not the point.

The value of a first goal will fluctuate slightly as scoring levels do. It's still never as important as a late tying/winning goal. The NHL's goals per game would have to accept soccer levels for that to be the case.

Exactly. No expectation of a fixed value. Red flags raised if it is a fixed value.

It is critical to defining the value of a late winning or tying goal. Time and scoring in between. How long the first goal held or did it produce a scoring explosion?Just as it is critical to defining team, individual player, goalie, coaching performance. How the referees call a game and any other derivatives you may wish to generate.

 

[seventieslord]

Exactly. No expectation of a fixed value. Red flags raised if it is a fixed value.

It is critical to defining the value of a late winning or tying goal. Time and scoring in between. How long the first goal held or did it produce a scoring explosion?Just as it is critical to defining team, individual player, goalie, coaching performance. How the referees call a game and any other derivatives you may wish to generate.

What is your point? Did I ever give you the impression that I would expect a goal importance quantification system to assume the same parameters season after season?

 

[Canadiens1958]

My Point

What is your point? Did I ever give you the impression that I would expect a goal importance quantification system to assume the same parameters season after season?

My point is that unless you ground the data to a basic reference point - the first goal of a game you will have disjointed and contradictory results that risk not making sense. See your ETOI effort about Bill Nyrop and Don Awrey in 1976. Major oops happened.

First goal as a foundation avoids such situations and allows the comparables across seasons, eras(defined by rules), strategies(value of goals when a team wishes to play an offensive vs a defensive game), looking a sitting at home waiting situations across the years, etc. with built in safety valves against questionable results.

 

[seventieslord]

My point is that unless you ground the data to a basic reference point - the first goal of a game you will have disjointed and contradictory results that risk not making sense. See your ETOI effort about Bill Nyrop and Don Awrey in 1976. Major oops happened.

There's no oops at all. Nothing I said was inaccurate. Awrey was the #4 in the regular season (which anyone should know I was referring to, since there is no data with which to estimate playoff TOI)

First goal as a foundation avoids such situations and allows the comparables across seasons, eras(defined by rules), strategies(value of goals when a team wishes to play an offensive vs a defensive game), looking a sitting at home waiting situations across the years, etc. with built in safety valves against questionable results.

Why do more people not just call these posts out for the gobbledygook that they are?

 

[Canadiens1958]

I See.......

There's no oops at all. Nothing I said was inaccurate. Awrey was the #4 in the regular season (which anyone should know I was referring to, since there is no data with which to estimate playoff TOI)



Why do more people not just call these posts out for the gobbledygook that they are?

I see .... kindly review the following thread where ETOI is used for playoff performance. My contributions seem to be appreciated.

http://hfboards.mandatory.com/showthread.php?t=2208643&highlight=black+gold+extractor

 

[BenchBrawl]

I wanted to do this once, gave up.

http://hfboards.mandatory.com/showthread.php?t=2109413&highlight=

 

[Uncle Rotter]

For example: Your team is down a goal with 1:00 left in the 3rd. There's a 3% chance your team wins the game at this point. You score a goal. That makes it a coin flip from here. It's 50%. You just made your team's likelihood of a win rise by 47% with that goal.

No, a coin flip would be a tie game. A one goal lead in the final minute is still a favourite.

 

[93LEAFS]

This stat exists in baseball, it is called Win Probability Added. This could pretty easily be done just by using a technique similar to the handicapping methods sportsbooks employ for shifts in odds. Although, you may have a problem with empty net goals in 1 lead games being slightly overvalued.

The issue with this stat in baseball, is that it isn't very repeatable compared to techniques used to measure high-level performance such as WAR. It may though provide a more accurate telling of a player's value, but just not its repeatability. Most of the analytics community is more interested it seems in finding out what leads to success, if so, is it repeatable and how do you quantify it. I think it'd be hard to argue that clutchness is a repeatable trait. It would help in a historical context though.

 

[Canadiens1958]

Baseball

This stat exists in baseball, it is called Win Probability Added. This could pretty easily be done just by using a technique similar to the handicapping methods sportsbooks employ for shifts in odds. Although, you may have a problem with empty net goals in 1 lead games being slightly overvalued.

The issue with this stat in baseball, is that it isn't very repeatable compared to techniques used to measure high-level performance such as WAR. It may though provide a more accurate telling of a player's value, but just not its repeatability. Most of the analytics community is more interested it seems in finding out what leads to success, if so, is it repeatable and how do you quantify it. I think it'd be hard to argue that clutchness is a repeatable trait. It would help in a historical context though.

Interesting post.

Repeatable vs predictable has to be differentiated. Baseball places certain hitters in preferred positions offensively, defining roles. Lead-off, clean-up, etc. Likewise certain hitters are in less preferred roles - second place, eighth place if the pitcher hits. Then there is the DH which is a wild card, used differently depending on the team line-up.

Hockey does not have such preferred roles for scorers. No clean-up or designated scorers. Likewise TOI is not a guarantee of scoring opportunity.