Analytics & fancy stats thread

[Volica]

Russell at 18.95% is...just wow.

Hamilton leading the pack. It's not surprising; he's been good with Calgary, but will still get the short end of the stick no matter who talks about him.

 

[Anglesmith]

Hamilton leading the pack. It's not surprising; he's been good with Calgary, but will still get the short end of the stick no matter who talks about him.

Really? Hamilton struggled defensively to start the year, but since he's adjusted, who's been giving him the short end of the stick?

 

[Volica]

Really? Hamilton struggled defensively to start the year, but since he's adjusted, who's been giving him the short end of the stick?

Read the main forum; watch the short end of the stick be given. Listen to an after-game show and he's a popular topic on the fan.

There are even folks on this board who play it off like he's a soft butterfly that can't play D

 

[Johnny Hoxville]

Dougie reminds me a lot of Lindstrom, minus the upper echelon defensive play in his own end. He's getting there, but that's the only thing holding him back from being an elite dman.

 

[Pirate King Luffy]

Dougie reminds me a lot of Lindstrom, minus the upper echelon defensive play in his own end. He's getting there, but that's the only thing holding him back from being an elite dman.

Who?

 

[Johnny Hoxville]

He played in the 30's.

 

[Anglesmith]

Read the main forum

Well there's your problem.

 

[Lunatik]

Who?

 

[Dertell]

Breaking news: Getting hit by a puck does not feel good. Regardless of play-style.

Given the wide range of key injuries that comes from other causes and the general arbitrary nature of man game lost, the r2 value is even higher than I expected. I don't have the datas for positions, but I am curious whether goaltenders are the main reason for the rise. I mean, it has to be impossible for them to get an injury while the game is being played on the other side.

fyi, the data point on the extreme right in the 1st chart represents the 2010-11 NYI.

 

[Dertell]

Because that tank thread was going a bit off-topic with the faceoff stuff. Here's some material pro-faceoff people may find interesting:

https://hockey-graphs.com/2015/01/15/the-relationship-between-corsi-and-winning-faceoffs/

http://canucksarmy.com/2011/8/1/the-overrated-value-of-a-faceoff

tl;dr: There is a small correlation between faceoffs and possession (Corsi), and the little that exist can be explained by the centre or the players on the same team being good at winning puck battles, whether it's the FO or not.

Also, the post above oddly came at the right time, in right of Hartley's recent comments regarding blocked shots.

 

[slappipappi]

I would like a time based possession stat, you can pass the puck around for 5 minutes and if you don't try to shoot it's not reflected in possession stats.

Why should it?

You score by shooting pucks towards the net.

 

[Lunatik]

Why should it?

You score by shooting pucks towards the net.

Yes, but that is shots, not possession. Possession in every other sport is measure in time because time of possession is important as the more you have the puck, the less your opponent does and shots do not always accurately measure that.

 

[Anglesmith]

Why should it?

You score by shooting pucks towards the net.

That's not really the point, nor is it even the approach that analytics is going for (in fairness). Shot attempts are used because they are a considered a suitable approximation for possession, not because shot attempts themselves are going to necessarily help you win a game.

Given it's possible for other sports to track possession directly, though, it seems odd that possession is measured indirectly by hockey analytics.

 

[Dertell]

That's not really the point, nor is it even the approach that analytics is going for (in fairness). Shot attempts are used because they are a considered a suitable approximation for possession, not because shot attempts themselves are going to necessarily help you win a game.

Given it's possible for other sports to track possession directly, though, it seems odd that possession is measured indirectly by hockey analytics.

Corsi is better than (or favored over) pure zone time and real possession statistics, because Corsi represents meaningful possession. By that, I mean possession of the puck + control of the play + position to score. The main reason corsi is used over fenwick is it can already create large sample size quickly, so there's no point in dumping it in favor of a stat with more noise in the data.

Many people - myself included - refer to corsi as possession sometime, but it's a bad alias because it doesn't represent very well why it is tracked or what it represents.

 

[Dertell]

Must read about zone entries:

https://jenlc13.wordpress.com/2016/04/13/zone-time-goals/

More evidences the flames need to dump stretch pass in favor of short pass and carry-in, even if scoring hasn't really been the issue this year.

Dumping might be one of the reasons why the Kings' shooting% has been low for so many years, too.

 

[Dertell]

Warning: so much words, numbers.

So the other day I was looking into scoring chance prevention and if it really exists and to what extent. Turns out, my previous hypothesis that it's random and doesn't exist - was off the mark. Since scoring chances% and high-danger chances% (as defined by WOI) correlates strongly with Corsi%, the best way to see if a team generally keep the opponents outside the danger zone and helps the goaltender boost his SV% is to do a ratio between all-quality events against (corsi, fenwick, SOGs) and high-quality chances against. Based on the data collected by WOI in the past ten years, I tested the year-by-year repeatability of those ratio.

Just for clarifications' sake:

C = Corsi
F = Fenwick
S = SOG
SC = Scoring Chance
HSC = High-scoring chance
A = Against

Ratio | Repeatability Year to Year (Autocorrelation r2)
HSCA/CA | 0.126
SCA/CA | 0.777
HSCA/FA | 0.035
SCA/FA | 0.594
HSCA/SA | 0.174
SCA/SA | 0.671

To give an idea of the variance between teams: the SCA/CA ratio varied between 0.42 and 0.57 and the SCA/SA ratio varied between 0.75 and 1.05 (a scoring chances can be a missed shot). Yea, SOG might be a useful stat after all. High-scoring chances against, on the other hand, seem completely useless. No idea why Anything/Fenwick can't auto-repeat. I triple checked the numbers and I'm pretty sure I didn't mess up anything.

Regardless, there's no doubt that CA year-by-year repeatability is stronger, it has a stronger correlation with GA and the teams' CA number can vary in a far bigger range. Shots suppression a more important part of the game than scoring chances. However, teams can affect the save% to significant extent. For example: the 2015-16 flames' SCA/SA ratio is at 0.96; 0.05 higher than average and at .832 percentile.

0.05 may seem little, but middle-chances and high-chances (which together constitutes scoring chances) have a 92.5 and .833 Save% at 5v5 respectively. On the other hand, "Low-chances" have a 0.975 Save%. Statistically speaking, it's fair to say the flames system harmed the on-ice save% by at least ~1.0% at ES. In this case, 1% is the difference between league-worst and a more respectable 21st place with 92.1%.

tl;dr: Getting a new goaltender isn't a priority. Hiller's departure and fixing the defensive system will do far more to boost the team's Save%. The team's scoring chances / shots ratio is awful relative to the rest of the league and the stat can auto-repeat.

 

[RedHot]

Analytics Thread

The last one was deleted, so here's a thread to discuss anything stats related, their use in today's game, and how they may affect how teams make decisions.

Some helpful resources:
http://stats.hockeyanalysis.com/
http://ownthepuck.blogspot.com/

 

[Pirate King Luffy]

Why was the last one deleted?

 

[RedHot]

Why was the last one deleted?

Not sure. I bumped it this morning, then next time I checked it was gone.

 

[Flames Fanatic]

I'm slowly, SLOWLY coming around to advanced stats as further evidence, as opposed to the whole picture by themselves.

For example, advanced stats suggest that Cody Ceci was being carried by Patrick Weircoch. Anyone who has EVER watched the Sens will tell you its the exact opposite to the extreme.

 

[Ace Rimmer]

It's because there was a bump without any added content.

At least this time there is added content.

 

[RedHot]

An article on Troy Brouwer and his zone starts.

http://flamesnation.ca/2016/7/18/will-giving-troy-brouwer-more-offensive-zone-time-work

 

[Fig]

TBH, I wouldn't mind a quick and dirty cheat sheet thread on different types of stats, how it's calculated, what it's trying to interpret and what its short comings are.

I mean, I have the gist of things like Corsi or Fenwick, but can never really wrap my mind around why they are relevant because there's always those posters out there saying, "that one has flaws, this one is better" but I don't know what it is measuring other than % is better, is better player/team.

http://nhlnumbers.com/2014/9/1/the-nhl-advanced-stats-cheat-sheet

This for instance is great history, or what the stat is based on... but IMO isn't super helpful in helping me understand the application of the stats or the weaknesses of each stat (hence why there's so damn many).

 

[Lunatik]

TBH, I wouldn't mind a quick and dirty cheat sheet thread on different types of stats, how it's calculated, what it's trying to interpret and what its short comings are.

I mean, I have the gist of things like Corsi or Fenwick, but can never really wrap my mind around why they are relevant because there's always those posters out there saying, "that one has flaws, this one is better" but I don't know what it is measuring other than % is better, is better player/team.

http://nhlnumbers.com/2014/9/1/the-nhl-advanced-stats-cheat-sheet

This for instance is great history, or what the stat is based on... but IMO isn't super helpful in helping me understand the application of the stats or the weaknesses of each stat (hence why there's so damn many).

The biggest weakness as for any stat is context. People use them without context and ruin any value they might have. Other weaknesses are "possession" stats like Corsi and Fenwick are determined by shots and/or shot attempts, not calculated in time as the word possession would indicate it should. They also makes the assumption that blocked shocks are bad, even though players are taught to get in the shooting lanes. I mean sure that sliding blocked shot that takes the guy out of position is bad, but blocking it with your shins is just as effective at breaking up a play as knocking the puck off their stick. They also make a ton of assumptions, "high quality scoring chances" are determined by location on the ice, not if the shot was actually a high quality scoring chance or not. A wrist shot from within the home plate area into the goalies logo is a better scoring chance than a wrist shot from outside that area with 2 bodies in front of the goalie. Almost every one of these stats makes an assumption of some kind, which makes them extremely flawed.

 

[Dertell]

TBH, I wouldn't mind a quick and dirty cheat sheet thread on different types of stats, how it's calculated, what it's trying to interpret and what its short comings are.

I mean, I have the gist of things like Corsi or Fenwick, but can never really wrap my mind around why they are relevant because there's always those posters out there saying, "that one has flaws, this one is better" but I don't know what it is measuring other than % is better, is better player/team.

http://nhlnumbers.com/2014/9/1/the-nhl-advanced-stats-cheat-sheet

This for instance is great history, or what the stat is based on... but IMO isn't super helpful in helping me understand the application of the stats or the weaknesses of each stat (hence why there's so damn many).

Different people have different opinions. What is a stat's "pros and cons" will vary depending on who you ask. For example, some will say Corsi's weakness is it's strongly affected by "context". I say instead its weakness is that a forward's CF60 doesn't mean much.

But in general, it's less a matter of pros and cons and more so that different stats represent different things happening during the game. Some stats end up being replaced by newer stats serving the same purpose more efficiently, others are combined to create a value.