Examining the value of the secondary assist

[VoluntaryDom]

at least since 1922:

1922020004

19221220

TSP 2 - 7 SEN

[TBODY] [/TBODY]

PUNCH BROADBENT (2) , ASSISTS: JACK DARRAGH (1), GEORGES BOUCHER (3)

punch broadbent is a hell of a name

 

[Michael Farkas]

theres a reason people do these studies confined to even strength

What specifically is that reason?

 

[VoluntaryDom]

What specifically is that reason?

because the thing you were describing is PP, where systems have a major effect on who gets the benefit of the points from those PPs. while at ES, offense is less system based, and players points (at least primary ones) are generally skill-based.

 

[Michael Farkas]

So this methodology (and similar), fails to account for game tactics of virtually any regard? Because the level of tactical impact on even strength varies pretty greatly from team to team, if you've ever been at a high-level team's practices or even NHL practices themselves...while I do agree that special teams is more schematic based, but it's not quite as dramatic as I read the context of your note to be...

It's not hard to make similar ES calls...like San Jose's offense running through their right defense, including how they work high-low plays is almost exclusive to the right. Burns/Karlsson/Heed/Braun vs Vlasic/Dillon/Ryan. A lot of guys in that first group at production go up, the opposite in group two...

 

[morehockeystats]

I haven't come across an actual rule that would have prevented the awarding of secondary assists in the first few years of the NHL but the first time a goal was credited with two assists happened in 1922-23 season (happened seven times that season).

You're right that there was even years when tertiary assists were allowed and even awarded but it happened in the 1930's, not 20's. Per the rules apparently tertiary assists were possible from 1930-31 season onwards but they were actually awarded only during two years:

1934-35 (57 tertiary assists awarded)
1935-36 (65 tertiary assists awarded)

Since 1936-37 season only a maximum of two assists per goal were allowed.

Just came across such a situation. Game 1934020010, Eagles - Bruins (how fitting these days), 11/17/1934. The lone goal in the boxscore is credited to Babe Siebert, assisted by Eddie Shore and Nels Stewart. However Hap Emms is also credited with an assist in the roster summaries, obviously tertiary.

 

[Filthy Dangles]

Great thread and work but I wanted to bump because I can't really wrap my head around why Points would vary less than Primary Points do year to year, since goals and primary assists have much higher correlations than second assists do.

Shouldn't adding the much more random (relatively speaking) 2nd assists to Goals and Primary Assists (much higher inter-year correlation) make them (points) vary more year to year (than primary points)?

Hopefully it's not too dumb a quesiton....

 

[Hockey Outsider]

Great thread and work but I wanted to bump because I can't really wrap my head around why Points would vary less than Primary Points do year to year, since goals and primary assists have much higher correlations than second assists do.

Shouldn't adding the much more random (relatively speaking) 2nd assists to Goals and Primary Assists (much higher inter-year correlation) make them (points) vary more year to year (than primary points)?

Hopefully it's not too dumb a quesiton....

Not a dumb question at all. What I found is secondary assists are, on their own, nearly useless in predicting secondary assists in the following year. Primary assists are decent (but not great) in predicting primary assists in the following year (the same is true for goals - so this tells me what I already knew, which is hockey isn't an overly predictable sport from year to year). But looking at total assists - so taking primaries and secondaries together - are stronger still at predicting total assists in the following year.

I think your question is - if secondary assists don't have much informational value, why does adding it to the mix give us better results? It's because secondary assists do have informational value. They don't have much informational value at predicting themselves, but they have informational value in predicting other things.

All things being equal, a player who gets a secondary assist was probably more involved in the play than someone who recorded nothing. Sure, on average they were less involved than the player actually scoring the goal, or directly setting it up, but if I'm assessing who's a productive talented forward, I'd rather have a player partially involved through a secondary assist, rather than not at all. That's why the correlation is higher - a player recording a secondary assist was at least indirectly involved in the play and, all things being equal, we'd expect someone with partial involvement in the future to score more than someone with no involvement.

 

[Hockey Outsider]

Here's an idea (to further look into the value of secondary assists) - looking for feedback before I spend the time running the numbers.

I think the issue is, ultimately - are secondary assists indicative of a player contributing to their team's offense? Or is it just statistical noise?

What may be useful is a prediction exercise. Let's say we've defined a certain population (say, all forwards over X seasons with at least XXX minutes, at even strength). We know how many goals their team scores per 60 minutes when they're on the ice (the dependent variable) and their personal rate of scoring goals, primary assists, and secondary assists per 60 minutes (the independent variables).

If secondary assists are just statistical noise, if I were to put together a model, then we would expect that a model with just G+1A would be at least as good at predicting their team's offensive output as a model with G+1A+2A. Another way of testing this - if I put together a model using all three variables, and secondary assists are just statistical noise, we'd expect the predictive power of the secondary assists variable to be much less than that of goals and primary assist variables.

Why this makes sense to me - what we ultimately care about is how much a player helps their team score goals (even if they're not the ones scoring the goals themselves). So, if we compare two players, and their goal and primary assist rates are the same, but one records far more secondary assists - does the player generating more secondary assists actually contribute to his team scoring more goals overall? Comparing two players, of course, is meaningless, but looking at a sample of several hundred players over several years might be interesting/useful.

Does this approach make sense? I can run the numbers, but open to feedback before I invest the time in doing this.

 

[Snippit]

Not a dumb question at all. What I found is secondary assists are, on their own, nearly useless in predicting secondary assists in the following year. Primary assists are decent (but not great) in predicting primary assists in the following year (the same is true for goals - so this tells me what I already knew, which is hockey isn't an overly predictable sport from year to year). But looking at total assists - so taking primaries and secondaries together - are stronger still at predicting total assists in the following year.

I think your question is - if secondary assists don't have much informational value, why does adding it to the mix give us better results? It's because secondary assists do have informational value. They don't have much informational value at predicting themselves, but they have informational value in predicting other things.

All things being equal, a player who gets a secondary assist was probably more involved in the play than someone who recorded nothing. Sure, on average they were less involved than the player actually scoring the goal, or directly setting it up, but if I'm assessing who's a productive talented forward, I'd rather have a player partially involved through a secondary assist, rather than not at all. That's why the correlation is higher - a player recording a secondary assist was at least indirectly involved in the play and, all things being equal, we'd expect someone with partial involvement in the future to score more than someone with no involvement.

I wonder if your weighted points (G+A1+0.66 A2) ranking would show more correlation than points?

In theory, if one was to kept adjusting the weight of A2 until you maximized the R^2 value, would that A2 value be a good estimate at the true value of a secondary assist?

 

[Hockey Outsider]

I wonder if your weighted points (G+A1+0.66 A2) ranking would show more correlation than points?

In theory, if one was to kept adjusting the weight of A2 until you maximized the R^2 value, would that A2 value be a good estimate at the true value of a secondary assist?

Good question. Yes - weighted points (G + A1 + 0.66 A2) gives a better correlation than total points.

Actually, the correlation is maximized when A2's are given a weight of 0.69. I'm not sure why I'm getting 0.69 here and 0.66 before (it's been several months since I've gone through the data in detail) but, in any case, they both support secondary assists as being roughly two-thirds the predictive value of a goal.

 

[supsens]

Here's an idea (to further look into the value of secondary assists) - looking for feedback before I spend the time running the numbers.

I think the issue is, ultimately - are secondary assists indicative of a player contributing to their team's offense? Or is it just statistical noise?

What may be useful is a prediction exercise. Let's say we've defined a certain population (say, all forwards over X seasons with at least XXX minutes, at even strength). We know how many goals their team scores per 60 minutes when they're on the ice (the dependent variable) and their personal rate of scoring goals, primary assists, and secondary assists per 60 minutes (the independent variables).

If secondary assists are just statistical noise, if I were to put together a model, then we would expect that a model with just G+1A would be at least as good at predicting their team's offensive output as a model with G+1A+2A. Another way of testing this - if I put together a model using all three variables, and secondary assists are just statistical noise, we'd expect the predictive power of the secondary assists variable to be much less than that of goals and primary assist variables.

Why this makes sense to me - what we ultimately care about is how much a player helps their team score goals (even if they're not the ones scoring the goals themselves). So, if we compare two players, and their goal and primary assist rates are the same, but one records far more secondary assists - does the player generating more secondary assists actually contribute to his team scoring more goals overall? Comparing two players, of course, is meaningless, but looking at a sample of several hundred players over several years might be interesting/useful.

Does this approach make sense? I can run the numbers, but open to feedback before I invest the time in doing this.

When you see guys doing perfect break out passes up to the blue line giving your team a full speed 2 on one I’m not sure how it can be written off as noise. Have you tried to only use say the center position? Just to find a common link?
A team that likes to drive the puck at the net and jump on rebounds might depend on the second assist guy to move the puck up the ice and pass it to the shooter, a team that likes to pass and one time it might not need that assist as bad. I think it’s way to random. It seems like a line stat that can’t be fairly broken up.

 

[Yzing]

Great thread and nice work.

One thing I'm thinking of: Often when there's a 2-on-1 or 2-on-0, the goal scorer is effectively making the play and makes what in theory should be a secondary assist. Like this goal yesterday.

In this case, Ehlers showed great offensive skills, speed, puckhandling and passing with one hand. But a player obviously can't get assists too when he scores. So no secondary assists were given in this case.

Another example - also from yesterday - was Draisatl's goal against Toronto. He was moving the puck almost from his own goal line to the red line before he passed the puck, got it back, and scored. In both these cases, the goal scorers were contributing significantly in setting up the plays but this is ignored in the secondary assists stats. And this is probably one of the reasons why secondary assist stats are more "noisy" than goals or primary asists stats. Secondary assists are either ignored or sometimes given to the player with the tertiary assist and thus moved even further from the action in setting up the play. The latter was the case with the Draisaitl goal I just mentioned.

Even the first Oilers goal yesterday was such a case when the secondary assist happened pretty far (action-wise) from the goal. Here is the passing sequence leading up to the goal: McDavid-Klefbom-Larsson-Andersen(save)-Larsson-Klefbom. McDavid got the secondary assist although the puck was touched 5 times after him.

Goals and primary assists are never skewed this way. They are always given to the players touching the puck the last and second last time before it crosses the goal line. All this is probably terribly obvious, but I'm new here, and I did't see it mentioned in the 2011 article on page 1 in the thread.

Another thing is that secondary assists are somewhat rarer than goals or primary assists (this was mentioned in the 2011 article). Using the numbers posted on page 1 suggests that out of all the points scored by the top 50 players, 41 % were goals, 36 % primary assists, and only 22 % secondary assists. I'm not sure though how or if this smaller sample size affects the correlation from year to year.

 

[supsens]

Great thread and nice work.

One thing I'm thinking of: Often when there's a 2-on-1 or 2-on-0, the goal scorer is effectively making the play and makes what in theory should be a secondary assist. Like this goal yesterday.

In this case, Ehlers showed great offensive skills, speed, puckhandling and passing with one hand. But a player obviously can't get assists too when he scores. So no secondary assists were given in this case.

Another example - also from yesterday - was Draisatl's goal against Toronto. He was moving the puck almost from his own goal line to the red line before he passed the puck, got it back, and scored. In both these cases, the goal scorers were contributing significantly in setting up the plays but this is ignored in the secondary assists stats. And this is probably one of the reasons why secondary assist stats are more "noisy" than goals or primary asists stats. Secondary assists are either ignored or sometimes given to the player with the tertiary assist and thus moved even further from the action in setting up the play. The latter was the case with the Draisaitl goal I just mentioned.

Even the first Oilers goal yesterday was such a case when the secondary assist happened pretty far (action-wise) from the goal. Here is the passing sequence leading up to the goal: McDavid-Klefbom-Larsson-Andersen(save)-Larsson-Klefbom. McDavid got the secondary assist although the puck was touched 5 times after him.

Goals and primary assists are never skewed this way. They are always given to the players touching the puck the last and second last time before it crosses the goal line. All this is probably terribly obvious, but I'm new here, and I did't see it mentioned in the 2011 article on page 1 in the thread.

Another thing is that secondary assists are somewhat rarer than goals or primary assists (this was mentioned in the 2011 article). Using the numbers posted on page 1 suggests that out of all the points scored by the top 50 players, 41 % were goals, 36 % primary assists, and only 22 % secondary assists. I'm not sure though how or if this smaller sample size affects the correlation from year to year.

The part that interests me is if one player is getting roughly the same points each year and primary/secondary assists seem random if the players line is getting roughly the same amount of goals for. Making the value of secondary and primary pretty much the same. Placing the value more on possession and ability to get the puck down the ice. Just random luck on what forward touched it last before the shot
Although if the extra secondary assists come from your line scoring more goals resulting if free points that goes the other way and the secondary assists don’t have the same value. Good god you have to dig deep to make any sort of determination on what’s going on.
I personally think it’s just the number of shots on goal the line can produce and random what 9th or 10th shot goes in resulting in either a primary or secondary point. Meaning they should have around the same value

 

[plusandminus]

Interesting thread and well done work.
I've been away from this board and advanced hockey stats for 4-5 years, so I'm not so up to date.

Could HockeyOutsider or someone else summarize the conclusions made so far?

You only looked at 5 vs 5 play? Does that include empty net goals at even strength (6 skaters vs 5 skaters + 1 goalie)? Perhaps you would like to compare if EN goals differ from other goals?

Your conclusions are that goals and primary assists are stable, while secondary assists is more unstable and is influenced more by environment?

I saw some value here, like 0.65 or similar. Do you mean that secondary assists could be multiplied with that number in order to get a "fairer" Pts total? (Instead 1+1+1 for G, A1 and A2, you suggest 1+1+0.66?)

What's the use of all this excellent work? What difference does it make? What have we learnt?

Isn't handling out points rather arbitrary anyway?
Guys who obscures the goalie's view don't get a point, but may have been crucial in the goal happening?
Guys who takes defensive responsibility so that others can go "all in" offensively doesn't get a point? Without their defensive responsibility there might not have been a goal anyway?

 

[kugelbahn]

Would be interesting to find correlation beyond secondary assists. I mean get number of goals while player is on ice at even strength (On-Ice EV GF at nhl.com), subtract event strength points (EVP) from that, calculate inter year correlation for this number.
Or just calculate On-Ice EV GF inter year correlation. It is another parameter to play with...

 

[Hockey Outsider]

Interesting thread and well done work.
I've been away from this board and advanced hockey stats for 4-5 years, so I'm not so up to date.

Could HockeyOutsider or someone else summarize the conclusions made so far?

You only looked at 5 vs 5 play? Does that include empty net goals at even strength (6 skaters vs 5 skaters + 1 goalie)? Perhaps you would like to compare if EN goals differ from other goals?

Your conclusions are that goals and primary assists are stable, while secondary assists is more unstable and is influenced more by environment?

I saw some value here, like 0.65 or similar. Do you mean that secondary assists could be multiplied with that number in order to get a "fairer" Pts total? (Instead 1+1+1 for G, A1 and A2, you suggest 1+1+0.66?)

What's the use of all this excellent work? What difference does it make? What have we learnt?

Isn't handling out points rather arbitrary anyway?
Guys who obscures the goalie's view don't get a point, but may have been crucial in the goal happening?
Guys who takes defensive responsibility so that others can go "all in" offensively doesn't get a point? Without their defensive responsibility there might not have been a goal anyway?

Welcome back to the forum. Sorry for the slow response. To address your points:

In this study, I used all ES data (so that would include empty net goals). It may have been possible to remove empty net goals somehow, but I couldn't figure out how to do that. Obtaining data is easier now than it was 5 years ago, and certainly much easier than it was 15 years ago, but it's still a headache most of the time.

Yes, goals and primary assists (at ES, for forwards) both appear to be repeatable year over year. Secondary assists are much less so (but are probably a good indicator of possession or offensive involvement - as secondary assists are useful in predicting total assists and points in the following year).

Yes, my best estimate is, for the set of data I studied, secondary assists are probably worth around two-thirds of a goal (or primary assist). Another study that I posted, using a different approach, comes to a similar conclusion - that there's clearly value in secondary assists, even if they're not as valuable as goals.

What's the purpose of all this work? I find it interesting, some other people apparently do as well. But the reality is it's unlikely to ever be ready by more than a few dozen people who have an interest in both hockey and stats. (In a perfect world, I would discover some great, previously unknown truth about player performance, and use that to win all my hockey pools - but hockey simply isn't that predictable and/or I haven't discovered anything truly groundbreaking yet).

I agree, points are arbitrary (but it's highly unlikely the NHL will change perhaps the most famous individual statistic, which has been unchanged for a century). Look at Tomas Holmstrom - fantastic at screening the goalie. His career high in assists is just 30. He would have earned dozens of more assists over the course of his career if he got credit for screens.