The Uncovered Truth About PDO and SH%

[Donner]

Not sure why fans of teams on a PDO bender can't just shut up and enjoy the ride

If I go to the casino and win 6 blackjack games in a row, I am going to enjoy the wins and realize it'll cool off in the future. I'm not going to delude myself that I'm a card counter now

 

[MessierII]

I'm not sick of advanced stats. I'm sick of the continuing misunderstanding - not by casual fans, by people who do teh analtyics and created advanced stats - of what some stat really means or how to practically apply it, because they should all know better. Way too many people - and especially teh analytics people - and way too many takes still see correlation and thinks aha, there's the explanation! and run off accordingly.

PDO and winning percentage are highly correlated. PDO is nothing more than an aggregate of all team and individual performance, for offense and defense, split into two numbers that are added together. Are both of those two numbers that get added together highly correlated with winning percentage? No. Or, not nearly as much as PDO. [Which, this should already be raising questions, but it doesn't - so, we move on.] So all I have to do is increase "the sum of my goalie saving more shots" and "the sum of my skaters to score on more shots" and we're more likely to win more games, right? OK. How do I use PDO to adjust the roster, adjust the offensive and defensive schemes to improve my chances of winning? Playoff series are short, there's a hell of a lot more variability in results in 4-7 games vs. 82 games, how do I use PDO from the regular season to predict what's going to happen in the playoffs? Is there any practical application of PDO that I can reliably use for decision-making for say organizational planning, sports betting, or anything? But goddamnit, PDO is highly predictive! It's highly predictive, Bob!

Some like has CF% above X, and the xGF/xGA is above some ratio? Oh, look, they're doing really good? OK, why? Does that line have special synergy? Is someone carrying the others? Is it favorable line matchups? Have they just been on a heater and they're going to calm down? How do I apply this for other guys on the roster? Can I take this and build a roster where everyone in other places has "above average" stats and now on the same roster, they're still going to have "above average" stats or maybe be even better? And this goes to all kinds of advanced stats getting cranked out: lots of ooh, aah, completely lacking in context or explanation.

Which brings me to one of my all-time hates: if a stat is higher/lower than expected, it's explained away with "well, there's going to be regression to the mean" as if they or anyone else knows what the mean is. You don't know. I mean, I'm sure you think you do, you claim you do because of long-term averages and blah blah blah, but at the micro-level, at the player level, you have no clue what any player's "mean" is at any given point in their career until you have enough data to get a reliable sense of what they've done, by which time other things have changed and the observed historical mean may not be the expected mean going forward. Quit pretending you have any clue what "mean" is for something.

I've said it a number of times, I'll say it again: whoever figures out how to take all these stats and apply them in such a way that they produce reliable results going forward is going to make an absolute f***ing shitload of money, because that's going to be applicable to a huge swath of business applications. [Spoiler: it starts with understanding why some stat looks like it does.] Until then, everyone is figuratively pissing in the kiddie pool thinking they're making great art.

Great post man. So true. I’ve always found the people who are the biggest into analytics are somehow the shittiest at actually applying them. Going way back to 2009 when people thought they had discovered some good player/bad player magical stat in Corsi.

 

[VivaLasVegas]

The regular season means nothing other than to accumulate points for the playoffs and to prepare for the playoffs. The Canucks have been the most successful at accumulating points in a season where nobody thought they would be better than a bubble team, if even that, so kudos to them.

The rest of this "expected" stuff though is just a bunch of crap, just like how that Win-O-Meter is the last refuge of butthurt fans who want to say that they had the better team even though they actually lost.

 

[andora]

SV% is the more reliable of the two components of PDO, but goalie variance season to season is still unreliable for the bulk of goalies in the NHL. Certain goalies are so elite that their teams have a higher PDO baseline than average. Lundqvist was a great example of this.

The problem with that comes playoff time, cause the truth is that MOST of the 16 teams that make the playoffs have at least GOOD goaltending. So PDO advantage in the net, even if reliable, shrinks as an advantage relative to opposition once the quality of competition is more consistently high. In Lundqvist's case, for example, he was reliably excellent in the playoffs, but the advantage that provided to his mediocre possession and xGF% team was reduced when the teams and goalies he was playing against were closer to him than they were to league average.

Yep fair enough and that is the beauty of playoffs as this exact thing has to be accounted for, adjusted to, and executed.

The 'wasting low percentage shots' argument has always been nonsense. The teams that produce more shots on goal win more games than those that produce less over large samples. Teams that generate a lot of shots not only have to put the puck toward the net, but they also have to be able to retrieve the puck, get the puck back into position for a shot, and get the puck on the net again. No teams are shooting the puck from center ice to boost their underlying numbers, every team is trying to get as many looks from the high-danger areas as possible. Some teams are simply better at maintaining pressure and maintaining possession in the opposing zone... which not only gives them more opportunities to score, it keeps the puck away from their net.

This part kind of went all over the place but using the far side of the spectrum about shots from center ice to boost numbers is not what i said nor what i meant

It depends on the team obviously, but if a team has an average or low PDO and they're winning 80% of their games, there is a high likelihood they're controlling play and dominating possession and chance share.

Unlike sh%, the number of shots a team takes and the chances they generate are controllable, and thereby both repeatable and predictive. If this hypothetical team's PDO craters further, they will still be able to control the puck and generate more chances than their opponent, thus giving them a good chance to win.

By comparison, a high PDO team with poor underlying metrics is relying entirely on an uncontrollable factor to reach victory. When the heater ends, they have none of the fundamentals that give the other team a buffer. Whereas the other team controls the puck and thereby limits chances against it, the PDO team does not. Where one team generates a ton of chances and eventually achieves a goal by virtue of volume, the PDO team cannot, and therefore when they stop scoring a disproportionate amount of their shots, their goals dry in a far more severe way.

This is all fair, however i would challenge that scoring chances are controllable. Shots taken yes - generating a scoring in itself isnt imo.

If, hypothetically, you had two teams with the exact same underlying stats as far as driving play, and maintaining possession, and generating chances, and you had one team with a high PDO and the other with a low PDO, then one team is not better than the other, one is simply luckier. And since PDO isn't controllable, their fates could swap at any time for no reason. The team that was winning more would start losing more, not because they became worse, but because they became unluckier.

But if scoring chances are controllable than wouldnt the pdo team just be better at it than the other team?

Things are meant to be done at some point in history. Somebody breaks the mold.

It's the elite system Tocchet employs. They are very very selective in what shots they take. Lots of snipers on the team. Half of them can just quickly tell within 0.01 sec if that shot will be a goal or not. They then quickly decide whether to snipe or wait it out. That explains the high shooting %. They don't get a lot of high danger chances, but they cash in.

I'm surprised more teams don't watch their video and just try to copy this system.

Defensively, they are a rock. They allow some of the most high danger chances in the league, but they have Demko and even Desmith to bail them out.

That is your high PDO in a nutshell. They just play very smart hockey. Very efficient team.

Some of the most hd chances is false no?

from the start of the season until Dec 31st, the Canucks had 5 players in the top 20 in shooting percentage
it's up to 7 now

sure, their shooting percentages are going down...but so are the rest of the leagues players

So its where they are ranked now?

 

[Ted Hoffman]

Things are meant to be done at some point in history. Somebody breaks the mold.

Gambler's fallacy - Wikipedia

en.wikipedia.org

 

[BagHead]

By this time of year, PDO is pretty well normalized, so not much should change between now and the end of the year. Any regressions will most likely be small.

PDO is not reliably predictive. That goes both ways, you can't predict a team won't do well in the playoffs just because they have high(er) PDO right now.

In playoff series, teams with lower PDO beat teams with higher PDO regularly. The opposite is also true. It really is close to a coin-toss as to which team will be victorious, if you're just looking at PDO as your only metric for evaluation.

So what use is PDO? It's slightly better than a coin-toss at predicting. I think the stat does more harm than good, personally, as it gives people false security in their convictions. People seem to think having a high PDO means the team is winning a lot more often than it normally does, but in truth they are usually winning only slightly more often, or maybe not at all. They may even be on a losing streak while rocking a high PDO.

 

[Pip]

By this time of year, PDO is pretty well normalized, so not much should change between now and the end of the year. Any regressions will most likely be small.

PDO is not reliably predictive. That goes both ways, you can't predict a team won't do well in the playoffs just because they have high(er) PDO right now.

In playoff series, teams with lower PDO beat teams with higher PDO regularly. The opposite is also true. It really is close to a coin-toss as to which team will be victorious, if you're just looking at PDO as your only metric for evaluation.

So what use is PDO? It's slightly better than a coin-toss at predicting. I think the stat does more harm than good, personally, as it gives people false security in their convictions. People seem to think having a high PDO means the team is winning a lot more often than it normally does, but in truth they are usually winning only slightly more often, or maybe not at all. They may even be on a losing streak while rocking a high PDO.

how often do teams struggle with a high PDO?

 

[BagHead]

how often do teams struggle with a high PDO?

Probably less than 50% of the time, but more than 40% of it (on average), but I guess it also depends on what you mean by "high" and "struggle". I don't have the data compiled to be able to answer that question with a mathematical result, even if you define those terms precisely.

A great resource for understanding PDO and it's nature is Nathan Gabay's site. Just scroll down to the PDO section. You'll see that, in his results, the best teams usually have better PDO over the course of a full season, but that it's not vastly better. In the playoffs, where the sample sizes are small, PDO becomes fairly meaningless in terms of predicting the winner.

Not directed at you, but something I probably should have said in my original post and did not:
Most "high" PDO teams are teams that make the playoffs. High is in quotes because the distribution of the stat is typically very small, with the top teams usually being within ~4 PDO points of the bottom teams. It's no surprise that the better the goaltending and scoring ability a team has, the better their place in the standings ends up. But that doesn't mean all high PDO teams are good, or that all low PDO teams are bad. Look at the Red Wings vs. the Hurricanes this year. The Red Wings are clinging to the last playoff spot in the East, but are 4th in the league in PDO, while the Hurricanes are 2nd in their division and have the 6th worst PDO. Both the top and bottom PDO playoff teams have won the Cup. Aside from being able to say "a high PDO generally correlates with a playoff position after a long season", there is nothing definitive to say about PDO because there's no rhyme or reason to the rest of the results on a game-by-game or series-by-series basis.

 

[andora]

how often do teams struggle with a high PDO?

I think arizona montreal and islanders are top 10 for 5on5

 

[cptjeff]

Advanced stats are not predicting anything. I think that's an important concept that a lot of people don't understand on here. The issue is that people use stats and advanced stats to make stupid predictions. It should be used to help paint a picture of what's happened during a specific period of time and some are better than others.

I think what you're trying to say is that PDO is a simple stat and you're correct because it's just team shot % + team save %. It isn't something like xGF where there are many flawed measures within the stat.

If a statistic is not at least somewhat predictive of future performance it is useless statistical noise.

Which is what PDO is. Noise.

 

[BurnabyJoe7]

If a statistic is not at least somewhat predictive of future performance it is useless statistical noise.

Which is what PDO is. Noise.

That's not what a statistic is.

But thats a common narrative on here. Some think advanced stats is a conspiracy when they're so ignorant that they don't even know what a stat is. You also don't have to be a genius to understand what an advanced stat is and how it's not predictive of future performance.

 

[Machinehead]

1,892 characters just to say you don't understand statistics.

 

[GeeoffBrown]

I think people are bringing up PDO with regard to the Canucks because they have often missed the playoffs with a similar roster

If the Avalanche or Golden Knights had a high PDO, no one would bring it up because people expect those teams to be good

Also, the Canucks just hilariously lost a game 10-7, so it is expected they would take some light playful barbs on message boards

 

[Bank Shot]

more or less. he wouldn't or couldn't exercise the discipline the coaching staff were asking for on shot selection among other things. he kept giving up possession on low percentage plays and he didn't move the puck where they wanted it. among many things, he couldn't cycle or rotate the puck to the point his teammates wouldn't give it to him. he basically could not grasp ozone possession for its own sake.

so i've answered your multiple questions, now it is your turn.

why is it that all of the canuck defence except hughes and and all of the canucks top forwards except garland are taking less shots/60 this year than last year (or even earlier this year when they played for other teams)?

i see two explanations

-are they all being more selective in shots than last year?

-are they all not playing as well this year as they did last year and thus generating less shot chances?

or is there another explanation?

So since Feb, (28 games), the Canucks are 21st in the NHL in goals for /game and 16th in the NHL in shooting % at 5v5.

If you look at all strength shooting percentage its even worse. 26th in goals for. 25th in shooting percentage.

The explanation is that the Canucks were on a giant heater for the first half of the year.

There is no magic system.

 

[BurnabyJoe7]

So since Feb, (28 games), the Canucks are 21st in the NHL in goals for /game and 16th in the NHL in shooting % at 5v5.

If you look at all strength shooting percentage its even worse. 26th in goals for. 25th in shooting percentage.

The explanation is that the Canucks were on a giant heater for the first half of the year.

There is no magic system.

Vancouver's 5v5 PDO for the season is now under 103 and at an expected range based on historical data. This is basically what every non insecure Canuck fan was expecting to happen by the end of the season after their hot start.

The OP of this thread should start an anti advanced stats club next. I would expect a lot of interest based on all the uneducated comments in here.

 

[Dekes For Days]

A team can be sustainably higher or lower than 100 in PDO, but pretty much all teams that proclaim their extreme outlier to be special fall back.

This thread seems to have been sparked because of Vancouver. Well...

Vancouver PDO:
Season - Feb 17th (when you created this thread): 1.044
Feb 18th - Present: 0.973

Pretty much exactly what was expected to happen.

 

[HolyHagelin]

Whatever, as long as the Rangers defy the PDO gods for 4 rounds if the playoffds all is good

 

[MikeyMike01]

Not sure why fans of teams on a PDO bender can't just shut up and enjoy the ride

If I go to the casino and win 6 blackjack games in a row, I am going to enjoy the wins and realize it'll cool off in the future. I'm not going to delude myself that I'm a card counter now

Not sure why the cult of advanced stats can’t figure out that their models are wrong, not reality.

 

[Brock Boeser Laser Show]

A team can be sustainably higher or lower than 100 in PDO, but pretty much all teams that proclaim their extreme outlier to be special fall back.

This thread seems to have been sparked because of Vancouver. Well...

Vancouver PDO:
Season - Feb 17th (when you created this thread): 1.044
Feb 18th - Present: 0.973

Pretty much exactly what was expected to happen.

Since Feb 18th the Canucks are tops in the league in HDCF%. Allowed the fewest HDCA with 154. Offense has dried up a bit but Desmith stinks.

I doubt many of the haters expected the Canucks to be one of top 2nd half teams analytically.

 

[syz]

Since Feb 18th the Canucks are tops in the league in HDCF%. Allowed the fewest HDCA with 154. Offense has dried up a bit but Desmith stinks.

I doubt many of the haters expected the Canucks to be one of top 2nd half teams analytically.

They're below Dallas and Edmonton since Feb 18th.

 

[Korpse]

Not sure why the cult of advanced stats can’t figure out that their models are wrong, not reality.

I don’t know. Kind of looks like the law of averages is winning out here.

 

[ijuka]

From season start until January 22, Vancouver Canucks had 105.0% PDO.

From season start until January 22, Vancouver Canucks earned the most points in the entire league.

Since January 23, Vancouver Canucks have had 99.5% PDO.

Since January 23, Vancouver Canucks have earned the 19th-most points in the entire league.


This is PDO.

 

[TooManyHumans]

From season start until January 22, Vancouver Canucks had 105.0% PDO.

From season start until January 22, Vancouver Canucks earned the most points in the entire league.

Since January 23, Vancouver Canucks have had 99.5% PDO.

Since January 23, Vancouver Canucks have earned the 19th-most points in the entire league.


This is PDO.

It's like arguing with people in MLB about BABIP. These things always find their level.

 

[Dekes For Days]

Since Feb 18th the Canucks are tops in the league in HDCF%. Allowed the fewest HDCA with 154. Offense has dried up a bit but Desmith stinks.
I doubt many of the haters expected the Canucks to be one of top 2nd half teams analytically.

They're actually 3rd in HDCF%. They're 1st at 5v5, but I'm not sure why we're focusing specifically on HDCF% anyway. They're 7th in xGF% over that time.
I don't think anybody was arguing that Vancouver isn't a good team this year. Just that they were overachieving as well, which was true.

 

[PocketNines]

It seems as though NHL fans oft forget the somewhat obscure rule disqualifying teams from winning the Stanley Cup if they have a PDO that is too high. Similarly, and perhaps more obscure, is the rule stating that if a player on a team has too high of a shooting percentage, that also disqualifies the them.


Cited here:

46.15 Excessive PDO – Should a team's regular season PDO exceed 1050, that team will be ineligible to win the Stanley Cup, regardless of the outcome of the games in which they participate.

and

50.12 Unsustainable Shooting Percentage - When any individual player on a team has a regular season shooting percentage higher than 33%, that team will be disqualified and hence ineligible from winning the Stanley Cup


The infamous case of these rules actually being enforced was in 1928 when the Montreal Maroons advanced to the Stanley Cup Final against the New York Rangers. The Maroons went on to win that series, based on actual on-ice results, and capture the Stanley Cup with Babe Siebert scoring a series-clinching goal late in game 5 (the Final was a best-of-3 back then). Unfortunately for them and their fans, these rules, which had been buried in the rulebook, were brought to light. It turns out that Siebert's teammate Hooley Smith had a shooting percentage of 33.4%. Even more egregious, the team collectively had a regular season shooting percentage of 18.6%. Most appalling, and what ultimately ended up costing the Maroons the Stanley Cup, was their team regular season PDO of 1051.

This entire ordeal likely couldn't happen today with the speed of information transfer we now have, but one can see how in 1928 the coaches and players perhaps were not even aware of these rules - and how the Maroons were allowed to play all the way to the Stanley Cup Final before NHL President Frank Calder had to rightfully rip the Cup out of the hands of Maroons captain Dunc Munro.

Wasn't this the series where Mud Bunson got tomahawked by the goalie stick of Jumbo Mullins?