Good players with bad advanced stats

[OskarOskarius]

How do you explain great players with bad/mediocre advanced stats. For instance McDavid who arguably is the best in the world has CF% of just 50.9 and relative CF% of 1.70. Is it some sort of statistical effect?

Click to read more...

 

[BigBadBruins7708]

How do you explain great players with bad/mediocre advanced stats. For instance McDavid who arguably is the best in the world has CF% of just 50.9 and relative CF% of 1.70. Is it some sort of statistical effect?

That's his 5v5 CF and CFrel, not overall. But to your point, CF and CFrel are heavily tied to teammate's performance. Basically you can lead the league in ES shots and scoring but if the defense gives up a ton of shots you'll have a lower CF and CFrel. It seems with CF and CFrel, guys who excel at (or play with those that do) at shot suppression have higher CF% numbers than those that are more offensively minded.

Which is a long winded way of saying, it's mostly because McDavid is average defensively and he plays on a team that is well below average defensively.

 

[OskarOskarius]

That's his 5v5 CF and CFrel, not overall. But to your point, CF and CFrel are heavily tied to teammate's performance. Basically you can lead the league in ES shots and scoring but if the defense gives up a ton of shots you'll have a lower CF and CFrel. It seems with CF and CFrel, guys who excel at (or play with those that do) at shot suppression have higher CF% numbers than those that are more offensively minded.

Which is a long winded way of saying, it's mostly because McDavid is average defensively and he plays on a team that is well below average defensively.

Yeah but shouldn’t a good offensive lead to more shots for than shots against thus leading to a high CF%?

 

[Hockey Outsider]

A few points:

I believe the CF% you're quoting is only looking at five-on-five situations. McDavid is (so far, anyway) having one of the most productive seasons in NHL history in terms of powerplay production. That accounts for half of his production, and is excluded from the stat you're looking at.

Also as @BigBadBruins7708 mentioned, any corsi-type stats are heavily influenced by the quality of the team. In 2023, looking at 5v5 only, min 200 minutes, as of today, Chicago has five of the bottom 20 players (by CF%), St. Louis and Arizona have four, and Anahiem as two. At the top of the list, Carolina alone has eight of the top fifteen (and five of the top seven), with NJ having three. Calgary has three more in the top 20. The point is - these stats are heavily based on the quality of the player's team and a straight-up comparison can be misleading without taking the team context into account.

Plus, the sample sizes (at roughly 40 games) are pretty small. Warren Foegele and Ryan McLeod currently have better CF% than McDavid. It's essentially impossible for that to persist over the long-term. Fluky results can happen over a small sample size.

Lastly, don't get too focused on corsi. The benefit of corsi is it gives you a larger sample size, but ultimately goals for and against are what makes a team win (and lose) games. McLeod and Foegele have slightly better CF% than McDavid. But looking at actuals goal for and against, McDavid blows them away. McDavid is on the ice for 13% more goals against/60 compared to Foegele. But he's on the ice for 47% more goals for/60. In the long-run, I don't really care about corsi, I care about which player will maximize my team's goal differential, and that's what McDavid is doing.

 

[OskarOskarius]

A few points:

I believe the CF% you're quoting is only looking at five-on-five situations. McDavid is (so far, anyway) having one of the most productive seasons in NHL history in terms of powerplay production. That accounts for half of his production, and is excluded from the stat you're looking at.

Also as @BigBadBruins7708 mentioned, any corsi-type stats are heavily influenced by the quality of the team. In 2023, looking at 5v5 only, min 200 minutes, as of today, Chicago has five of the bottom 20 players (by CF%), St. Louis and Arizona have four, and Anahiem as two. At the top of the list, Carolina alone has eight of the top fifteen (and five of the top seven), with NJ having three. Calgary has three more in the top 20. The point is - these stats are heavily based on the quality of the player's team and a straight-up comparison can be misleading without taking the team context into account.

Plus, the sample sizes (at roughly 40 games) are pretty small. Warren Foegele and Ryan McLeod currently have better CF% than McDavid. It's essentially impossible for that to persist over the long-term. Fluky results can happen over a small sample size.

Lastly, don't get too focused on corsi. The benefit of corsi is it gives you a larger sample size, but ultimately goals for and against are what makes a team win (and lose) games. McLeod and Foegele have slightly better CF% than McDavid. But looking at actuals goal for and against, McDavid blows them away. McDavid is on the ice for 13% more goals against/60 compared to Foegele. But he's on the ice for 47% more goals for/60. In the long-run, I don't really care about corsi, I care about which player will maximize my team's goal differential, and that's what McDavid is doing.

In other words corsi is good for analyzing a few games but over a season +\- is better?

 

[BigBadBruins7708]

Yeah but shouldn’t a good offensive lead to more shots for than shots against thus leading to a high CF%?

The Oilers aren't a good offense. They just happen to have by far the best player in hockey (arguably of the last 30 years) and a 2nd offensive weapon that is in the top 2-5 players in the league.

Look at it this way, a team that is great offensively but equally bad defensively will net close to zero in shots for vs shots allowed

 

[Kaners Bald Spot]

As someone who watches Seth Jones on a nightly basis, this thread defines him.

When the Hawks traded for him I laughed at it, thought it was a ridiculous move for a player that was a borderline NHLer, because that's what the stats say. Then I saw him play, and now I get it.

Hockey stats are beyond imperfect. It's not baseball and you can't quantify everything.

 

[x Tame Impala]

IIRC Patrick Kane usually has/had poor advanced stats

 

[Kaners Bald Spot]

IIRC Patrick Kane usually has/had poor advanced stats

Yep, him too.

 

[Rengorlex]

Corsi is a terrible stat.

 

[Rengorlex]

In other words corsi is good for analyzing a few games but over a season +\- is better?

Expected goals is considered to be the preferred stat, I follow a lot of the stats people on Twitter and haven't seen corsi being used to evaluate player performance in a long time.

 

[RapidKnight]

I go a step further (then xGF) and favor HDCF because lines shouldn’t be rewarded for point shots because others lines can’t be rewarded for maintaining possession.

Secondly CF significantly disfavors pass first players like McDavid. I hated to admit this in the past because it seemed like an excuse but it’s true.

xGF mitigates these phenomenons compared to CF but HDCF will eliminate these altogether. Basically what I’m trying to say is HDCF is most immune to shooting tendencies. xGF however accounts for shooting talent which HDCF obviously doesn’t. xGF is better on it’s own but used as context (how they should be used) to their actual production I’d argue HDCF is equally or more valuable than xGF.

Others have mentioned that it’s only 5v5 production which is completely valid but I can’t stress enough that you shouldn’t combine it with power play production. Shots in both situations are vastly different which will make the end the result highly dependent on power play ice time. Secondly these stats on the powerplay is even more useless. Vegas used to slaughter other teams in PP advanced stats but still had a bad powerplay prior to this season. Combining does more harm than good imo.

 

[Captain97]

90% of defensive Dmen.

Shea Weber was the king of this

 

[wgknestrick]

How do you explain great players with bad/mediocre advanced stats. For instance McDavid who arguably is the best in the world has CF% of just 50.9 and relative CF% of 1.70. Is it some sort of statistical effect?

How do you explain using just shot counts as "advanced"? Using # of shots is just using "bad" stats vs that player having actual bad advanced stats (IE ones that correlate to winning).

However, when McDavid takes a set of shots where an NHL average player would be expected to score 20 goals and gets 32 goals are we going to say he has bad advanced stats?

 

[Mickey Marner]

90% of defensive Dmen.

Shea Weber was the king of this

Agreed, defensive defensemen are probably the best example. Ryan McDonagh has a career 49% corsi & -1.6 rel corsi, yet is +150 at 5v5.

 

[FlameChampion]

I don’t think (some) analytics work great on McDavid and Draisaitl. They are more focused on high dangerous chances rather than shot metrics. They both pass up on chances a lot, unless they think they have a good chance to score. Its also why both are like 20%+ shooters typically.

 

[Rengorlex]

I don’t think (some) analytics work great on McDavid and Draisaitl. They are more focused on high dangerous chances rather than shot metrics. They both pass up on chances a lot, unless they think they have a good chance to score. Its also why both are like 20%+ shooters typically.

The all-encompassing WAR type models that are used in the Twitter analytics sphere rate McDavid extremely highly. The last three years by McDavid have been probably the most dominant hockey witnessed in the advanced stats era. McDavid having bad advanced stats is a pretty wild take.

 

[isles55]

McDavid’s relative xg% the last 3 seasons are 7.97%, 11.46%, and 12.17%. Those are superstar numbers, but when you look at only offense it gets even more ridiculous, 1.04/60, 1.22/60, and 1.27/60. These are likely the best numbers ever recorded over a three season span.

Numbers from naturalstattrick, other sources might be slightly different.

 

[lomiller1]

How do you explain great players with bad/mediocre advanced stats. For instance McDavid who arguably is the best in the world has CF% of just 50.9 and relative CF% of 1.70. Is it some sort of statistical effect?

The short version is that McDavid does not have mediocre advanced stats, his advanced stats are actually really good.

raw CF% and xGF% are significantly impacted by team performance and should not be used to compare players across teams. The fact is that McDavid leads the Oilers in xGF% is significantly more important in evaluating his play than the fact his CF% is near league average. In fact the latter is viewed as almost worthless by people skilled in advanced stats.

Furthermore CF% and to a lesser degree xGF% are not the whole story. Some player are significantly better or worse point producers, even if their rel CF% or rel xGF% are similar. While xGF has closed this gap it has not eliminated it completely. This means you still need to factor in other advanced stats like points/60 and goals/60 when evaluating a player, and of course these are areas where McDavid dominates.

In terms of rel xGF I think McDavid could do better if didn't have one of the longest average shift lengths in the NHL. It doesn't follow that this would improve the Oilers results though. He's so good that even a slightly tired McDavid is still better than most NHL players. I expect this would tend to impact his defensive play more than his offensive play, and indeed we see that in his numbers.

 

[Rufus T Firefly]

The short version is that McDavid does not have mediocre advanced stats, his advanced stats are actually really good.

raw CF% and xGF% are significantly impacted by team performance and should not be used to compare players across teams. The fact is that McDavid leads the Oilers in xGF% is significantly more important in evaluating his play than the fact his CF% is near league average. In fact the latter is viewed as almost worthless by people skilled in advanced stats.

Furthermore CF% and to a lesser degree xGF% are not the whole story. Some player are significantly better or worse point producers, even if their rel CF% or rel xGF% are similar. While xGF has closed this gap it has not eliminated it completely. This means you still need to factor in other advanced stats like points/60 and goals/60 when evaluating a player, and of course these are areas where McDavid dominates.

In terms of rel xGF I think McDavid could do better if didn't have one of the longest average shift lengths in the NHL. It doesn't follow that this would improve the Oilers results though. He's so good that even a slightly tired McDavid is still better than most NHL players. I expect this would tend to impact his defensive play more than his offensive play, and indeed we see that in his numbers.

I go a step further (then xGF) and favor HDCF because lines shouldn’t be rewarded for point shots because others lines can’t be rewarded for maintaining possession.

Secondly CF significantly disfavors pass first players like McDavid. I hated to admit this in the past because it seemed like an excuse but it’s true.

xGF mitigates these phenomenons compared to CF but HDCF will eliminate these altogether. Basically what I’m trying to say is HDCF is most immune to shooting tendencies. xGF however accounts for shooting talent which HDCF obviously doesn’t. xGF is better on it’s own but used as context (how they should be used) to their actual production I’d argue HDCF is equally or more valuable than xGF.

Others have mentioned that it’s only 5v5 production which is completely valid but I can’t stress enough that you shouldn’t combine it with power play production. Shots in both situations are vastly different which will make the end the result highly dependent on power play ice time. Secondly these stats on the powerplay is even more useless. Vegas used to slaughter other teams in PP advanced stats but still had a bad powerplay prior to this season. Combining does more harm than good imo.

A few points:

I believe the CF% you're quoting is only looking at five-on-five situations. McDavid is (so far, anyway) having one of the most productive seasons in NHL history in terms of powerplay production. That accounts for half of his production, and is excluded from the stat you're looking at.

Also as @BigBadBruins7708 mentioned, any corsi-type stats are heavily influenced by the quality of the team. In 2023, looking at 5v5 only, min 200 minutes, as of today, Chicago has five of the bottom 20 players (by CF%), St. Louis and Arizona have four, and Anahiem as two. At the top of the list, Carolina alone has eight of the top fifteen (and five of the top seven), with NJ having three. Calgary has three more in the top 20. The point is - these stats are heavily based on the quality of the player's team and a straight-up comparison can be misleading without taking the team context into account.

Plus, the sample sizes (at roughly 40 games) are pretty small. Warren Foegele and Ryan McLeod currently have better CF% than McDavid. It's essentially impossible for that to persist over the long-term. Fluky results can happen over a small sample size.

Lastly, don't get too focused on corsi. The benefit of corsi is it gives you a larger sample size, but ultimately goals for and against are what makes a team win (and lose) games. McLeod and Foegele have slightly better CF% than McDavid. But looking at actuals goal for and against, McDavid blows them away. McDavid is on the ice for 13% more goals against/60 compared to Foegele. But he's on the ice for 47% more goals for/60. In the long-run, I don't really care about corsi, I care about which player will maximize my team's goal differential, and that's what McDavid is doing.

Are there any particular advanced stats or combination that you recommend looking at in order to evaluate defensive play? It seems like most stats (advanced or regular) focus on offensive opportunities as those are much easier to track. Charting the inverse (chances/shots/goals against) seems problematic.

I guess another way of asking this question is if there is any stat that you would like to see that would help evaluate this? I think something like Corsi against/HDCA would be interesting that would show the percentage of dangerous chances against amongst total shots against. But I don't know if that kind of data is out there.

 

[lomiller1]

I guess another way of asking this question is if there is any stat that you would like to see that would help evaluate this? I think something like Corsi against/HDCA would be interesting that would show the percentage of dangerous chances against amongst total shots against. But I don't know if that kind of data is out there.

Expected goals against (xGA/60) and defensive WAR\GAR are the normal go to for defensive play but they need to be used cautiously. It's difficult to get a defensive stat that is actually predictive of future success, which tends to suggest defensive success (relative to team and system) isn't a highly repeatable skill. Some models have seem to have detected repeatable skill others don't seem to. IMO such skill probably does exist but isn't all that different between NHL players.

This actually isn't as surprising at it may seem at first when you think about the way interference penalties work. You can "discourage" a player from going to high danger areas but you can't physically stop them, so if the offensive player has sufficient will to get to the dangerous spots the defender can't stop them without taking a penalty. This means good scoring chances are a much more repeatable via offensive play than preventing them is via defensive play.

 

[RapidKnight]

Are there any particular advanced stats or combination that you recommend looking at in order to evaluate defensive play? It seems like most stats (advanced or regular) focus on offensive opportunities as those are much easier to track. Charting the inverse (chances/shots/goals against) seems problematic.

I guess another way of asking this question is if there is any stat that you would like to see that would help evaluate this? I think something like Corsi against/HDCA would be interesting that would show the percentage of dangerous chances against amongst total shots against. But I don't know if that kind of data is out there.

xGA is the definitely the best for defense but it only gets you so close to the answer. I’m being inconsistent but the low dangers chances aren’t as problematic for defense because teams play against various offensive strategies so it essentially reduces the small benefit of using HDCF over xGF. CA is probably not as bad for defense for the same reason but it still fails to represent the bigger picture. OZS% and PKTOI / 5v5TOI% (an unquantifiable indicator of carry over 5v5TOI%) are other things that can impact xGA.

 

[Oleksiak]

nm

 

[Notsince67]

xGA is the definitely the best for defense but it only gets you so close to the answer. I’m being inconsistent but the low dangers chances aren’t as problematic for defense because teams play against various offensive strategies so it essentially reduces the small benefit of using HDCF over xGF. CA is probably not as bad for defense for the same reason but it still fails to represent the bigger picture. OZS% and PKTOI / 5v5TOI% (an unquantifiable indicator of carry over 5v5TOI%) are other things that can impact xGA.

I agree and would add that any xg captures distinct fenway shot attempts as an event which misses puck carrying, takeaways and blocks as defensive quantifiers. When puck carrying stats are finally released via puck tracking tech, I would bet a whole new level of predictive metrics show themselves as all 3 of these are linked beyond the fenway shot.

 

[Golden_Jet]

I agree and would add that any xg captures distinct fenway shot attempts as an event which misses puck carrying, takeaways and blocks as defensive quantifiers. When puck carrying stats are finally released via puck tracking tech, I would bet a whole new level of predictive metrics show themselves as all 3 of these are linked beyond the fenway shot.

Not sure they will, NHL , teams and broadcasters pay for sportlogiq data that has it. Teams pay about 150-200K for that.
So I assume sportlogiq in turn pays NHL