Will there ever be an all-encompassing stat to quantify a forward's defensive ability?

[seventieslord]

There's just so much to consider when it comes to a forward's defensive ability, but nowadays there are tons of statistics that can help with that. But hockey is so fluid that one stat can affect another so easily - has there ever been an attempt at an omnibus defensive stat that really passes the smell test (as in, a lot of the names you expect to see on the top based on eye test and reputation show up near the top)?

I think that there are a lot of factors that can go into it; for starters, we have our now-archaic RTSS stats like giveaways and takeaways. Do they even matter, or does the end result matter - in other words, if you're taking and giving up the puck at a certain frequency, and it's having a real impact on the play, then it will translate into shots for and against.

Other things to consider:

- corsi against rates, high danger chances against, goals against (not just raw, but also relative to team! not just relative to team, but also considering quality of linemates and quality of competition! not just that, but zone starts impact those numbers too! and so on...)

- the zone starts, TOI and QoC themselves aren't a "result" per se, but they do say a lot about what a coaching staff thinks about a player's ability - should these factor in as a main component or should it be all results-driven and just adjusted for these kinds of factors?

- penalty kill performance, based on shots against and goals against, but also relative to team (QoC not as important here as it's near constant)

- other micro-stats that I've seen tossed around seem to be important and seem to show the best reputational players doing well - "successful defensive touches causing a change in possession" and "puck battles won", etc. they seem like meaningful stats because the results match the eye test. Similar to giveaways and takeaways, do these matter on a micro level or are they just a means to an end?

- corsi against and goals against rates are only half of the corsi and goal conversation, too. When you look at the leaders in ES GAA you usually see a bunch of low-event 4th line players who play against other offensively inept 4th liners so are they really the guys you want to recognize? I've always said that we shouldn't just call the players with the most points (or the best scoring rates) "the best offensively" because hockey is fluid and their level of defensive attention affects their ability to score (i.e. I would say a player with 70 points and solid defensively is way better offensively than a guy who cheats his way yo 85 points, I don't think it's hard to make the case that if the former cheated he could also score 85 points himself) . In the same way, I would hesitate to call the guys with the lowest corsi against and goals against the "best defensively" simply because they're basically the opposite case - all defense, very little effort/ability offensively. There is merit to including some sort of offensive component to a defensive forward calculation that respects the ability to drive play forward and accepts that their greater focus on offense limits their ability to post stellar defensive stats.

There's so much more that could be included, and it's completely subjective how to roll it all together into one stat. But if it ended up that we saw Bergeron, Kopitar, Barkov, Danault and other guys like them near the top, I'd personally be pretty satisfied that someone had managed to capture defensive ability in a stat reasonably well. Does anything like this exist? Who has tried and come close?

Click to read more...

 

[morehockeystats]

Granger causality - Wikipedia

 

[Filthy Dangles]

I think there are Above/Below replacement stats for defense specifically. That would probably be the closest and best thing. You try to take in all aspects of defense using various metrics, assign their value and add them up.

WAR is cumulative so those 4th line guys who just grind and cycle the puck in the zone for a few minutes that you mention won't have as high a number as their rate statistics like their GAA or Shot Share%, unless they are adding a lot of value elsewhere

 

[seventieslord]

Granger causality - Wikipedia

I'm sorry but I don't get it.

 

[morehockeystats]

I'm sorry but I don't get it.

It means that when you begin to measure X to say the player is better at Y, players work on their X rather than on their Y.
The Israeli Army has a saying about it: "What is being measured, always improves."

 

[seventieslord]

It means that when you begin to measure X to say the player is better at Y, players work on their X rather than on their Y.
The Israeli Army has a saying about it: "What is being measured, always improves."

yes, I've heard of this concept before when discussing Corsi. It makes intuitive sense because a team/player can start throwing extra shots on net and then say, "see? I'm dominating possession, I'm a great driver of play" when the real goal is to score goals and win, and possession is only a predictor of winning, not the end goal itself.

But there's no reason to believe that this applies to stats for defensive forwards. First, we have to believe that a player like this is driven to be recognized statistically as the best defensive forward. Second, we have to think that they will start doing things that only benefit their defensive statistics as opposed to, you know, actually winning.

But I don't see how a player can go, "I'm going to stop giving away the puck and start taking it away more and start having more successful defensive touches that change possession and start winning puck battles - ALL FOR MY OWN BENEFIT!!!" Those things would actually make him a better defensive forward and would actually help his team. As would driving the direction of play away from his own net and/or towards the opposition's. As would avoiding goals against by all means necessary.

 

[morehockeystats]

yes, I've heard of this concept before when discussing Corsi. It makes intuitive sense because a team/player can start throwing extra shots on net and then say, "see? I'm dominating possession, I'm a great driver of play" when the real goal is to score goals and win, and possession is only a predictor of winning, not the end goal itself.

But there's no reason to believe that this applies to stats for defensive forwards. First, we have to believe that a player like this is driven to be recognized statistically as the best defensive forward. Second, we have to think that they will start doing things that only benefit their defensive statistics as opposed to, you know, actually winning.

But I don't see how a player can go, "I'm going to stop giving away the puck and start taking it away more and start having more successful defensive touches that change possession and start winning puck battles - ALL FOR MY OWN BENEFIT!!!" Those things would actually make him a better defensive forward and would actually help his team. As would driving the direction of play away from his own net and/or towards the opposition's. As would avoiding goals against by all means necessary.

I am not arguing it must apply. I am arguing that when you devise such a stat, you need to be aware of Granger because at the end it _may_ apply.

 

[Hockey Outsider]

The most comprehensive system that I'm aware of is Alan Ryder's Player Contribution - http://www.hockeyanalytics.com/Research_files/Player_Contribution_System.pdf

Advantages:

• The system separately looks at defensive play in all three situations (ES, PK, PP), and there's a further adjustment for avoiding penalties. It makes sense that any statistical system should do this. We can look at these situations in isolation, but ultimately we care about how a player performs overall. (In recent years I've found there's an exaggerated emphasis on 5-on-5, seemingly to the exclusion of everything else, when analyzing players - other game situations still count).
• The system identifies a lot of players that you'd expect. The article I linked is from 2003, and the system identified Jere Lehtinen as the most valuable defensive forward. Patrik Elias was 2nd. John Madden and Wes Walz were high on the list as well. The top ten defensive defensemen include Nicklas Lidstrom, Scott Stevens, Wade Redden (this was before he fell off a cliff), and Eric Desjardins (though there are some surprises - Toni Lydman). You can look at other seasons and a lot the "expected" names consistently come up. I've never found another statistical system that does this with anywhere near this type of regularity.
• The system is based on contribution, which consists of both ability and ice time. If Patrice Bergeron isn't deployed on the PK (for some reason), his contribution there would be zero. I think that makes sense - a player doesn't contribute, regardless of how good they are, by sitting on the bench (or being injured).

Disadvantages:

• It would be an understatement to say that this system is convoluted. I've probably read through the method a dozen times over the years and I still don't fully understand all of it. It took years for fans to accept basic concepts like save percentage and Corsi - good luck getting anyone except the most obsessive fans to understand something like this.
• I can't even imagine how many hours must have gone into calculating everything. Ryder posted results (with a lot of thoughtful commentary) for 2003 through 2011. Good luck trying to calculate the results for other years on your own.
• There aren't any adjustments for zone starts, quality of teammates, quality of competition, score effects, etc. The results are what they are, and are taken at face value. Maybe in 2003 Toni Lydman got much easier matchups then Stevens, Lidstrom, etc, and that explains why a purely statistical system reads his performance as superior?
• Shots against and other Corsi-type factors are excluded - only the bottom line, goals against, are counted. Actually, I'm not sure if that's a disadvantage. Corsi and related stats are useful in making forward-looking predictions, because it gives you a larger sample size. But if you have a historical system (ie evaluating what's already taken place), I think goals against is the better way to go. At the risk of stating the obvious, a player who was on the ice for 50 ES goals against was on the ice for 50 ES goals against. Maybe the possession numbers suggest he was unlucky, but there's no differentiation between expected goals and unlucky goals when the score is tabulated. (I don't see any inconsistency in saying that the player had a bad year defensively in year X, but is expected to do better next year based on the underlying Corsi numbers - similarly, if a player scores 30 goals on an unsustainable 25% shooting percentage, we can give him credit for having a great year, while at the same time recognizing that it's highly unlikely he'll repeat that performance again next year).

I'll be the first to say that complex doesn't necessarily mean better (a point that Ryder also acknowledges). But this is the best "all in one" defensive value system that I've seen.