Non-traditional metrics: PDO

[Talks to Goalposts]

The guy who said that was Gabe Desjardins.

He was referring to how wrapping your head around the PDO concept was very important and can help you a lot in your analysis.

What PDO is about is how a player's goal stats (essentially +/-) gets monkeyed around with over short periods due to swings of chance that are largely outside a player's control. Because plus minus itself is the heart of what hockey is but is so often made useless by the vagaries of circumstance. If plus minus corresponded mechanistically with level of play it would be the perfect stat, but it doesn't.

It does a lot to adjust for what a players teammates were doing, so you can see who got screwed with or helped by goaltending or teamates converting on their chances at a greater or less rate than normal.

Teams will have high PDO long term, but that is a predicatable result of goaltending talent. Keeping in mind PDO will tell you about players that are getting screwed in goals against because of bad goaltending.

Embracing PDO is about acknowledging that results don't necessarily follow quality of play in small sample sizes and that a player will return to their average ability in time. Its also a solid thing to look at when your trying to figure out if a player is actually getting better or worse with time or just having a run on good or bad circumstances. PDO is all about figuring out what kind of performance is sustainable and what was just happenstance.

 

[Jason MacIsaac]

What is IPP?

I wouldn't call PDO an "advanced statistic", I'd call it a new statistic that is very simple and very misleading. It has two components, one of which the individual has much control over and may indicate his skill to a large degree, the other of which he has very little control over (except to prevent high quality shots) and therefore has little to do with his skill. It's like putting sauerkraut and Oreos together and calling it an advanced food.

Look at the shooting %s of Brett Hull or Stamkos. They jumped in their second seasons, but that doesn't mean they didn't improve further a season or two later. Of course such levels are not sustainable for most players over longer periods, but it doesn't mean it was mostly luck.

Players have minimal difference on on ice shooting percentage. On ice shooting % is plenty more luck then skill. Some even suggest it is all luck in the long run.

NJ's PDO was wasted in 2011 by the shooting %, which was top 10 by seasons end this season. NJ was very unlucky with ES sv% during the regular season but look what happened in the playoffs.

PDO is the name of the poster who created the metric and people in the stats comunity dubbed it "PDO".

 

[Czech Your Math]

Players have minimal difference on on ice shooting percentage. On ice shooting % is plenty more luck then skill. Some even suggest it is all luck in the long run.

NJ's PDO was wasted in 2011 by the shooting %, which was top 10 by seasons end this season. NJ was very unlucky with ES sv% during the regular season but look what happened in the playoffs.

Even on a team level, PDO is not purely luck. Neither of the components, SV% and Sh%, are pure luck. Yes, in the shorter term, luck is going to be the primary factor. However, in the longer term, each component will trend towards an expected and/or historical avg. based in large part on the skill of the players.

The reason it appears to be "luck" to some in the longer term is that:

1) The difference between goalies' SV%s tend to be very small in the longer term. For instance, the top 30 goalies in GP last season had overall SV%s that ranged from 894 to 930. So there is only so much difference to be expected from this component. This is not the same as saying that the expected SV% is the league avg., because it is not so for each team and therefore for each player on that team.

2) The differences in Sh% for a group of players also will have relatively small differences in the longer term. A line of Alfie-Spezza-Heatley would not be expected to have the same Sh% as an energy line of grinders. This seems like simple common sense.

Since each component has a relatively small range, the sum of them will also have a relatively small range. This doesn't mean that each player (or line or group) and each team will be expected to trend toward the league average.

I see limited use for this metric, but it may easily be misapplied by those who don't understand that each player/team has a separate expected PDO and that future expected PDO should not be influenced by past PDO. If Vancouver has a PDO of 1002 after 5 games, I may expect their future PDO to actually go up in the future, not down to 1000 as the metric suggests. If Vancouver had a PDO of 997 after 5 games, that shouldn't change the expected future PDO.

 

[Czech Your Math]

I found the data relating to PDO, although I have concerns about its reliability. When I calculate the Shooting%s using their GF/60 and SF/60 data (and GA & SA), the results do not generally match the numbers provided.

However, assuming their data is correct, I see the following:

1) There is a fairly strong correlation (over 60%) for opponents' shooting% (S%A) from one season to the next. Since SV% = 1 - S%A, this component of PDO does not seem close to being "almost entirely driven by luck."

2) S%A is much less and negatively correlated to the current season's SA/60. So as the opponents' shots rise, the team's S%A falls, which increases SV%, which increases PDO. Since a team which is giving up a lot of shots is considered my many to be "outplayed" (outshot), and their SV% will be higher on average (due to the correlation described), and they will be considered a bit "luckier" due to a generally higher PDO.

3) As others have said, and I may have been incorrect about, a team's shooting% 5v5 appears to be rather unpredictable. I would like to verify the data before concluding this, however. There is a small negative correlation between SF and S%F, so again a team be considered by some to be "outplaying" their opponent, while being a bit "unlucky", but this is not too significant. There is a much stronger correlation between overall S%F from season to season. I would guess the reason for this, is that the players on the PP are of much higher skill, so that their shooting%s are much more predictable from one year to the next.

4) PDO itself is somewhat correlated to the PDO from the previous season (as measured by the difference between 1000). The correlation is much stronger (over 50%) for overall PDO.

Conclusions- As I already said, I hesitate to conclude much more without the BTN's data being verified as correct. However, it seems to me that:

A) On a team level 5v5 PDO may be a product of mostly random factors and therefore could be of some use in determining the sustainability of team ES goal differential.

B) On an individual level, 5v5 PDO may also be a product of mostly random factors, but this may vary significantly by individual. It's probably especially useful for 3rd/4th line grinders, who have limited skill, as compared to top liner players whose S%F is going to be much more influenced by that line's superior skills.

C) Overall team PDO seems to be much more predictable and may be at least as much due to skill as luck. When teams have their best players on the PP & PK, the cream rises to the top much more.

D) Since the metric is used to determine the sustainability of an individual player's plus-minus, the top line players' roles on the PP & PK must also be considered, as these roles may significantly influence their plus-minus numbers (through SHG F & A).

I would like to look further at PDO, but am waiting for confirmation that the data is reliable in order to do any further study on this.

 

[Fourier]

I found the data relating to PDO, although I have concerns about its reliability. When I calculate the Shooting%s using their GF/60 and SF/60 data (and GA & SA), the results do not generally match the numbers provided.

However, assuming their data is correct, I see the following:

1) There is a fairly strong correlation (over 60%) for opponents' shooting% (S%A) from one season to the next. Since SV% = 1 - S%A, this component of PDO does not seem close to being "almost entirely driven by luck."

2) S%A is much less and negatively correlated to the current season's SA/60. So as the opponents' shots rise, the team's S%A falls, which increases SV%, which increases PDO. Since a team which is giving up a lot of shots is considered my many to be "outplayed" (outshot), and their SV% will be higher on average (due to the correlation described), and they will be considered a bit "luckier" due to a generally higher PDO.

3) As others have said, and I may have been incorrect about, a team's shooting% 5v5 appears to be rather unpredictable. I would like to verify the data before concluding this, however. There is a small negative correlation between SF and S%F, so again a team be considered by some to be "outplaying" their opponent, while being a bit "unlucky", but this is not too significant. There is a much stronger correlation between overall S%F from season to season. I would guess the reason for this, is that the players on the PP are of much higher skill, so that their shooting%s are much more predictable from one year to the next.

4) PDO itself is somewhat correlated to the PDO from the previous season (as measured by the difference between 1000). The correlation is much stronger (over 50%) for overall PDO.

Conclusions- As I already said, I hesitate to conclude much more without the BTI's data being verified as correct. However, it seems to me that:

A) On a team level 5v5 PDO may be a product of mostly random factors and therefore could be of some use in determining the sustainability of team ES goal differential.

B) On an individual level, 5v5 PDO may also be a product of mostly random factors, but this may vary significantly by individual. It's probably especially useful for 3rd/4th line grinders, who have limited skill, as compared to top liner players whose S%F is going to be much more influenced by that line's superior skills.

C) Overall team PDO seems to be much more predictable and may be at least as much due to skill as luck. When teams have their best players on the PP & PK, the cream rises to the top much more.

D) Since the metric is used to determine the sustainability of an individual player's plus-minus, the top line players' roles on the PP & PK must also be considered, as these roles may significantly influence their plus-minus numbers (through SHG F & A).

I would like to look further at PDO, but am waiting for confirmation that the data is reliable in order to do any further study on this.

Thanks for taking the time to do this.

With respect to point B) I have always felt that the leagues elite players tend to be almost de facto outliers with respect to many of the more popular statistical manipulations.

 

[Jason MacIsaac]

1) There is a fairly strong correlation (over 60%) for opponents' shooting% (S%A) from one season to the next. Since SV% = 1 - S%A, this component of PDO does not seem close to being "almost entirely driven by luck.".

PDO is not individual sh%, it is the 5 man unit's sh% while on the ice. Sure Kovalchuk and Stamkos's sh% are farily consistant. I have never seen someone use individual PDO ever, I didn't know it existed. PDO is something that needs to be taken in context. If a player plays in a defensive system it is flawed to look at sv% at all, stick to sh% as a 5 man unit.

Nobody is saying there is one super statistic. When I evaluate a player I take Zone starts, Corsi On, QOC Corsi, QOC team, Corsi Off, SH% ON, SV% ON, all into consideration

 

[Master_Of_Districts]

I found the data relating to PDO, although I have concerns about its reliability. When I calculate the Shooting%s using their GF/60 and SF/60 data (and GA & SA), the results do not generally match the numbers provided.

However, assuming their data is correct, I see the following:

1) There is a fairly strong correlation (over 60%) for opponents' shooting% (S%A) from one season to the next. Since SV% = 1 - S%A, this component of PDO does not seem close to being "almost entirely driven by luck."

2) S%A is much less and negatively correlated to the current season's SA/60. So as the opponents' shots rise, the team's S%A falls, which increases SV%, which increases PDO. Since a team which is giving up a lot of shots is considered my many to be "outplayed" (outshot), and their SV% will be higher on average (due to the correlation described), and they will be considered a bit "luckier" due to a generally higher PDO.

3) As others have said, and I may have been incorrect about, a team's shooting% 5v5 appears to be rather unpredictable. I would like to verify the data before concluding this, however. There is a small negative correlation between SF and S%F, so again a team be considered by some to be "outplaying" their opponent, while being a bit "unlucky", but this is not too significant. There is a much stronger correlation between overall S%F from season to season. I would guess the reason for this, is that the players on the PP are of much higher skill, so that their shooting%s are much more predictable from one year to the next.

4) PDO itself is somewhat correlated to the PDO from the previous season (as measured by the difference between 1000). The correlation is much stronger (over 50%) for overall PDO.

Conclusions- As I already said, I hesitate to conclude much more without the BTI's data being verified as correct. However, it seems to me that:

A) On a team level 5v5 PDO may be a product of mostly random factors and therefore could be of some use in determining the sustainability of team ES goal differential.

B) On an individual level, 5v5 PDO may also be a product of mostly random factors, but this may vary significantly by individual. It's probably especially useful for 3rd/4th line grinders, who have limited skill, as compared to top liner players whose S%F is going to be much more influenced by that line's superior skills.

C) Overall team PDO seems to be much more predictable and may be at least as much due to skill as luck. When teams have their best players on the PP & PK, the cream rises to the top much more.

D) Since the metric is used to determine the sustainability of an individual player's plus-minus, the top line players' roles on the PP & PK must also be considered, as these roles may significantly influence their plus-minus numbers (through SHG F & A).

I would like to look further at PDO, but am waiting for confirmation that the data is reliable in order to do any further study on this.

Year-to-Year Correlations for Team Shooting and Save Percentage at Even Strength (EN Goals Removed)


EV Shooting Percentage

2003-04 & 2005-06: 0.35
2005-06 & 2006-07: -0.05
2006-07 & 2007-08: 0.35
2007-08 & 2008-09: 0.16
2008-09 & 2009-10: -0.15
2009-10 & 2010-11: -0.06

Average: 0.10

EV Save Percentage

2003-04 & 2005-06: 0.27
2005-06 & 2006-07: 0.51
2006-07 & 2007-08: 0.17
2007-08 & 2008-09: 0.40
2008-09 & 2009-10: 0.26
2009-10 & 2010-11: 0.46

Average: 0.35

 

[Czech Your Math]

PDO is not individual sh%, it is the 5 man unit's sh% while on the ice. Sure Kovalchuk and Stamkos's sh% are farily consistant. I have never seen someone use individual PDO ever, I didn't know it existed.

I understand that now. However, top players are likely often playing with other players of high skill and/or are influencing those players' sh%s with their own skill. Those who are likely of less influence on others' sh%s (goal-scorers themselves) are likely taking a high % of the shots by the group on the ice.

I'm not sure why you quoted the portion of my post pertaining to SV%F (essentially opponents' sh%A) and then spoke about individual sh%F. The two seem mostly unrelated.

 

[Czech Your Math]

Year-to-Year Correlations for Team Shooting and Save Percentage at Even Strength (EN Goals Removed)


EV Shooting Percentage

2003-04 & 2005-06: 0.35
2005-06 & 2006-07: -0.05
2006-07 & 2007-08: 0.35
2007-08 & 2008-09: 0.16
2008-09 & 2009-10: -0.15
2009-10 & 2010-11: -0.06

Average: 0.10

EV Save Percentage

2003-04 & 2005-06: 0.27
2005-06 & 2006-07: 0.51
2006-07 & 2007-08: 0.17
2007-08 & 2008-09: 0.40
2008-09 & 2009-10: 0.26
2009-10 & 2010-11: 0.46

Average: 0.35

The SV% numbers look very similar to what I calculated for 5v5 for the last three season pairs (.40 as a whole from '08 & '09 to '10 & '11).

The sh% numbers look a bit suspicious, with two .35 correlations and then later negative correlations. Any idea why the sh% correlations changed from somewhat positive to neutral/slightly negative in more recent seasons?

What is the source of your data for these correlations?

 

[Jason MacIsaac]

I understand that now. However, top players are likely often playing with other players of high skill and/or are influencing those players' sh%s with their own skill. Those who are likely of less influence on others' sh%s (goal-scorers themselves) are likely taking a high % of the shots by the group on the ice.

I'm not sure why you quoted the portion of my post pertaining to SV%F (essentially opponents' sh%A) and then spoke about individual sh%F. The two seem mostly unrelated.

You would think so but Detroit's top lines had terrible on ice sh% last year.

 

[Czech Your Math]

You would think so but Detroit's top lines had terrible on ice sh% last year.

Where do you find data for 2012?

One must also consider that Detroit's best players are older and some/all may be declining. The fact that one team's best players had a poor sh% on ice at ES does not invalidate the theory that skilled players' teams will have higher, more predictable sh%s while they're on the ice, nor does it prove that the sh% component of PDO is "all luck."

This article suggests PDO's sh% component may not be so much about luck on an individual player basis:

http://hockeyanalysis.com/2012/02/05/thoughts-on-pdo-and-luck/

 

[overpass]

You would think so but Detroit's top lines had terrible on ice sh% last year.

One season numbers may show nothing more than bad luck, or Zetterberg's bad back. But I wonder if Detroit, more than most teams, tries to maximize shot volume as part of maximizing goals. Without commenting on whether that is an effective strategy, it's very likely that it would lead to a lower SH%.

 

[Czech Your Math]

Anyone who is using the downloaded .xls files from BehindTheNet, I want to find out the following:

Why, when you take the GF/60 and SF/60 for a particular situation (e.g. 5v5 column O & P) and divide GF/60 by SF/60, does it often yield results that are substantially different than SPCTF (column Q) for that situataion (it's not a rounding error)? If that is the case, how can we trust that this data is reliable?

 

[Talks to Goalposts]

Anyone who is using the downloaded .xls files from BehindTheNet, I want to find out the following:

Why, when you take the GF/60 and SF/60 for a particular situation (e.g. 5v5 column O & P) and divide GF/60 by SF/60, does it often yield results that are substantially different than SPCTF (column Q) for that situataion (it's not a rounding error)? If that is the case, how can we trust that this data is reliable?

BTN's SF doesn't include goals, its saved shots. There should be your discrepancy. Try SF/(GF+SF) for save percentages

 

[Czech Your Math]

BTN's SF doesn't include goals, its saved shots. There should be your discrepancy. Try SF/(GF+SF) for save percentages

Thanks for the tip, but that doesn't solve it.

That's poor labeling too by BTN IMO.

 

[Master_Of_Districts]

The sh% numbers look a bit suspicious, with two .35 correlations and then later negative correlations. Any idea why the sh% correlations changed from somewhat positive to neutral/slightly negative in more recent seasons?

It's just random variation.

That said, I suspect, based on other lines of evidence, that the true underlying correlation is larger (likely 0.2-0.3).

What is the source of your data for these correlations?

NHL.com play-by-play feeds.

I've checked the numbers against officially published data and it more or less perfectly corresponds.

The reason why it doesn't perfectly correspond is because there are a few games for which there is no play-by-play feed.

 

[Egblad]

PDO Stat Questions

Hi guys, I was wondering if I could get some help with this interesting stat. I get the premise behind it, that it shows how "lucky" or "unlucky" a team is, or a player for that matter as either regress generally to around 100%.

I am interested in knowing what I can take away from that stat with players for instance and trying to predict perhaps better or worse performance in the future.

One thing that I found interesting was Nicklas Backstrom for Washington, had 44 points in 42 games in 2011-2012 season, so over a ppg. Yet his PDO was quite low at 986. What should I be taking away from that? Despite him being productive on the ice, he is "unlucky?"

Is it also possible to track a player's PDO from season to season and take away anything from it? Say if a player is 103% in 2010-11, and 11-12 it was 101%, can it be assumed that it will continue to in all likelihood regress?

Thank you guys for helping me understand this stat better. If anybody could provide insight as to how I should use this stat, thanks!

 

[Feed Me A Stray Cat]

Backstrom was probably unlucky.

You'd probably need 4-5 seasons of PDOs well above 1000 to make a definitive conclusion about a player's ability to outperform the averages.

 

[GKJ]

This might help you. Even if you're just looking for a collection of further reading on it at the bottom of the article.

http://www.fearthefin.com/2012/8/2/3215351/non-traditional-metrics-glossary-pdo

Average PDO is around 1000, so 1003 and 1001 (I'm assuming that's what you meant with percentages), it may not regress at all, or only regress slightly. James van Riemsdyk had the worst PDO on the Flyers (tied with Schenn), at 978, so that's expected to bounce back. You would need a few more years though to determine if a high (or low) PDO is sustainable. That could be where things like shot quality come into play

 

[MMonarchs]

Goals that don't change the outcome of the game shouldn't be counted. In other words, any goal scored by the winning team after the GWG shouldn't count. It should go in the books just as if it was a save. This "adjusted" PDO would mean a lot more.

 

[Jarick]

After running quite a few numbers and comparing with the Wild's performance, I am leaning towards PDO being an indicator of quality scoring chances and less an indicator of luck.

Odd man rushes and breakaways are better scoring chances than perimeter shots. Good teams box out the opposition and limit odd man rushes. Good teams transition on the breakout quickly and force turnovers leading to odd man rushes. All these things drive up PDO in a way that's not related to luck.

 

[hella rights]

PDO is not individual sh%, it is the 5 man unit's sh% while on the ice. Sure Kovalchuk and Stamkos's sh% are farily consistant. I have never seen someone use individual PDO ever, I didn't know it existed. PDO is something that needs to be taken in context. If a player plays in a defensive system it is flawed to look at sv% at all, stick to sh% as a 5 man unit.

So each player in a 5-man unit has the same PDO in a given game? If that's the case, I can't see how anyone can accurately assign an individual player a PDO if it's really a group level statistic, probably why you've never seen it used that way.

 

[Larry Hoover]

PDO & Fenwick % close are only for ES play, correct?

I wonder if a person could consider several variables, and combine them into a single statistic.

Power plays per game + PP% + penalties per game + PK% + PDO + Fenwick % Close...I wonder if such a statistic could be formulated. Would it be the best overall team ranking metric?

 

[Master_Of_Districts]

After running quite a few numbers and comparing with the Wild's performance, I am leaning towards PDO being an indicator of quality scoring chances and less an indicator of luck.

Odd man rushes and breakaways are better scoring chances than perimeter shots. Good teams box out the opposition and limit odd man rushes. Good teams transition on the breakout quickly and force turnovers leading to odd man rushes. All these things drive up PDO in a way that's not related to luck.

How does PDO repeat from half of the year to the other, across teams?

If what you're saying is true, there should be some significant correlation.

 

[Jarick]

I'm not sure it's a predictive stat, but it sure seems to be a descriptive stat.

Last night, Wild got killed. Just terrible. In every sense of the word. They gave up a breakaway first shift of the game and LA scored. Then they gave up a 3-on-1 two shifts later and LA scored. Lots of turnovers and crappy shots from the perimeter.

PDO was 786, shot differential +9. Can't really tell anything from Corsi because LA scored on their first two shots and Wild never scored.

But if you look at the PDO, you think, wow the Wild were terribly unlucky. You watch the game, probably their worst game of the year.