Friedman's 30 thoughts - analytics say Oilers' luck should turn

[SebC]

I took the 2007-08 to 2011-12 data for 5-on-5 PDO, Corsi%, Fenwick% and plotted each season against the previous season.

Results:

Corsi: 59% persistence, R^2=0.38
Fenwick: 54% persistence, R^2=0.33
PDO: 23% persistence, R^2=0.05

So yeah, PDO is more a measure of luck than Corsi or Fenwick. But if it were purely a measure of luck, that 23% would be 0%.

[doctajones428]

Am I able to block words on CP? Specifically:

Corsi
Fenwick
PDO
Percentage
Advanced
Stats
Unsustainable

[crapshoot]

Being stuck in the bottom of the barrell the way the Oilers are is not bad luck. However I do think that the Oilers management interpret the advanced stats in such a way that they think they are better than they are, so right now advanced stats might play a huge role in they Oilers staying the course.

[Caged Great]

Originally Posted by doctajones428

Am I able to block words on CP? Specifically:

Corsi
Fenwick
PDO
Percentage
Advanced
Stats
Unsustainable

It's just a new topic that hasn't been beaten to death yet. It will be in due time. Just have to endure this discussion because it's unsustainable over the long term. The percentage of users discussing advanced stats will decrease as more understand the inherent flaws in the Corsi, Fenwick, and PDO systems.

[Geeoff]

I really hope that the Oilers do start regressing to the mean. The idea that they could potentially draft and ruin McDavid and Eichel is extremely distasteful to me.

[Jay Random]

Originally Posted by malcolmk14

I created a normal distribution curve using PDO numbers for all NHL teams for the last six years and it came up with the following results:

Mean: 100.01
Standard Deviation: 1.234

Within 1 standard deviation of the mean (68.27% of teams will fall within this range):

98.776 - 101.244

Within 2 standard deviations of the mean (95.45% of teams will fall in this range):

97.542 - 102.478

Within 3 standard deviations of the mean (99.73% of teams will fall in this range):

96.307 - 103.713


The Oilers are outside 3 standard deviations from the mean at 96.1. This means they are either historically bad (bottom ~0.08% of teams), or they are unlucky.

The Flames are just outside 2 standard deviations from the mean at 102.9. This means they are either historically good (top ~1% of teams), or they are getting lucky.

The explanation is not that the Oilers have been unlucky or the Flames lucky, but that a smaller sample size means more randomness and a larger standard deviation. It is not correct to compare the distribution over a stretch of about 25 games per team with the distribution over a full season.

Furthermore, there is no prima facie reason to suppose that PDO will follow a normal distribution. If all shots on goal were perfectly random and stochastically independent, then PDO would tend to form a normal curve. They’re not. It therefore does not follow (for instance) that 99.73% of teams will fall within 3 standard deviations of the mean. All you can say is that at least 8/9 of all teams will do so, according to Chebychev’s rule. Beyond that, you have to look at the actual data and count how many teams actually fell inside that range.

Some people (I’m not accusing you; you just reminded me to rant about this) have this damned silly idea that because PDO has a mean of 100 by definition, everything will regress to the mean – that any deviation is strictly random. This is an error. You can see it if you apply it to other things.

If you add up all the games won and games lost in the NHL (counting OTL and SOL as losses), the aggregate winning percentage of the entire league each season will be .500 by definition. But nobody expects that every team will regress to .500 over time, because not all teams are equal. Why would anyone expect all teams to regress to an equal PDO?

[malcolmk14]

Thanks for the explanation. When I was thinking about it last night that was kind of where my thinking was going - I just couldn't explain it eloquently. All teams are not equal and shots are not a random distribution, so why would pdo regress to the mean? Makes perfect sense.

[JayP]

Originally Posted by Caged Great

It's just a new topic that hasn't been beaten to death yet. It will be in due time. Just have to endure this discussion because it's unsustainable over the long term. The percentage of users discussing advanced stats will decrease as more understand the inherent flaws in the Corsi, Fenwick, and PDO systems.

I'm sure you said the same thing about OBP and OPS in baseball 10-12 years ago. Advanced stats aren't going anywhere - sorry to burst your bubble.

You are right though - the percentage of users discussing Corsi, Fenwick, and PDO will decrease. That's largely because their inherent flaws are evident to the advanced stats community and they will be replaced by better, more comprehensive stats as they are developed. The exact same thing has happened in baseball, basketball, and football.

[FBI]

How about Corsi and fenwick within a defined quality shot area instead?

[FBI]

Or like shots right in the crease and shots from the slot are worth more than shots outside the blue line etc.

[Erick Estrada]

The analytics say the Oilers luck should turn but after last night I tend to believe the hockey gods don't particularly care what any of these concocted formulas spit out.

[Fan in Exile]

There's a reason they play the games and give the two points to the team that puts more pucks in the net rather than shots on goal.

Look at stats in other sports: was Ali lucky when he knocked out Foreman? Stats would have told you that Foreman threw and landed several times more punches and power punches, and controlled the centre of the ring.

Soccer games always report who has more possession, who has attempted and completed more passes but just as in hockey, there are teams that play different styles, teams that keep possession and pass it twenty times before taking a shot and others that rely on swift counter-attacks. The team that has the most possession is not necessarily the better or more dominant team. The same goes with basketball.

The Flames play a quick transition counter-attack game. The Avalanche play this way too. It also doesn't take much in-depth analysis to figure out that teams with good goaltending and skilled forwards that create quality scoring chances are likely to have higher PDOs.

I like what I see from the Flames. I think they'll make the playoffs if they continue to play in the same way. The Oilers despite the advanced stats will not improve unless they start playing differently.

[Street Pharmacist]

Originally Posted by FBI

How about Corsi and fenwick within a defined quality shot area instead?

There's work on this, but it's not easy to do. Firstly, what would the measurement be for quality scoring chances? How do you make that strandard? Secondly, how do you mathematically involve it with unblocked shots? Most quality scoring chances aren't blocked I would think.

[Flaming Choy]

Originally Posted by doctajones428

Am I able to block words on CP? Specifically:

Corsi
Fenwick
PDO
Percentage
Advanced
Stats
Unsustainable

You could install a chrome extension that replaces those words with anything you want.

https://chrome.google.com/webstore/d...lkjdpbfh?hl=en

[Resolute 14]

Or he could just not enter a thread that has "analytics" in the title.

[Caged Great]

Originally Posted by JayP

I'm sure you said the same thing about OBP and OPS in baseball 10-12 years ago. Advanced stats aren't going anywhere - sorry to burst your bubble.

You are right though - the percentage of users discussing Corsi, Fenwick, and PDO will decrease. That's largely because their inherent flaws are evident to the advanced stats community and they will be replaced by better, more comprehensive stats as they are developed. The exact same thing has happened in baseball, basketball, and football.

Actually contrarily I'm a huge proponent of the advanced stats in baseball, because they are actually beneficial and more indicative of the true talent of the player.

These stats, while helpful to a limited extent are flawed each in their own ways and don't account for the talent of players like the ones in baseball do. That's why the good teams always seem to outperform their underlying stats. It also doesn't help to pick out which areas that the player is skilled in. There is no creativity index in order to generate offensive chances, just that chances are being created. How is it being accomplished. I liken it to batting average in baseball. It's helpful to tell how good of a hitter the batter is, but it isn't OPS or anything that might tell you if they are getting hits are they doubles or all squeaker base hits. A player like Backlund gets highlighted as being a great corsi player, yet he doesn't generate enough points to account for his excellent #'s. He's basically a singles hitter. It's useful as he's great in that regard, but it's also limited as he can't drive the runs in (score 25+ goals 50+ points). Any stat should be able to provide context as well as a simple guideline. Unfortunately Corsi/Fenwick only provide the guideline without the context.

Also in baseball, it's virtually impossible to rig the stats in your favour. In hockey though, if you just start shooting from everywhere regardless of positioning or context, then you'll be better in those stats. Basically this is what Edmonton has done.

There are just too many variables unaccounted for with them. They are a step in the right direction though, and hopefully there are new stats that can provide the context of the players' talent in addition to their efficiency. We don't have that now.

[joe_mullen]

Originally Posted by Caged Great

It's just a new topic that hasn't been beaten to death yet. It will be in due time. Just have to endure this discussion because it's unsustainable over the long term. The percentage of users discussing advanced stats will decrease as more understand the inherent flaws in the Corsi, Fenwick, and PDO systems.

As opposed to the inherent flaws in the old fashioned eye test, +/-, etc. Why the backlash against potentially understanding the game at a more intricate level?

[Caged Great]

Originally Posted by joe_mullen

As opposed to the inherent flaws in the old fashioned eye test, +/-, etc. Why the backlash against potentially understanding the game at a more intricate level?

They don't go far enough to give context, which leaves too much up for interpretation still. That's fine as it'll give a bit more understanding of the game, and that's a good thing. However, it should not be the be all and end all like some of the metrics in baseball are.

OPS (On base % (batting avg + walks and being hit by the pitch) + Slugging %) gives the user context of how good they are overall. While each of the individual stats helps, they don't tell the entire story. JP Arencibia could hit 20+ homeruns, but was terrible at getting on base to the extent that it didn't balance out and he's not a regular any more.

Until there's a stat package that's as good as OPS is in baseball at giving context, then it's just another stat. Useful, but it requires the user to fill in the blanks that the stats can't account for.

[Itse]

Originally Posted by joe_mullen

As opposed to the inherent flaws in the old fashioned eye test, +/-, etc. Why the backlash against potentially understanding the game at a more intricate level?

For the most part those "advanced stats" do not actually provide any new understanding, let alone at "intricate level". Saying that they do doesn't make it real.

[doctajones428]

Originally Posted by Resolute 14

Or he could just not enter a thread that has "analytics" in the title.

Or maybe I wanted to read "Friedman's 30 thoughts" and the discussion related to all his thoughts, not just trash advanced stats.