Statistics for defensive performance

[Blue Blooded]

There are a lot of Corsi-derived statistics out there, but I can't find any that measures the difference in Corsi of their opponents between their average and when they match up against the player in question.

Example: Daniel Sedin had a Corsi On of 21.09. Let's say that when he played against Rob Scuderi he only managed 13.09, this would give Scuderi a Corsi Opp of 8.00 vs Daniel Sedin. His total Corsi Opp would be the average of all the players he has been on the ice against weighted for icetime.

All the statistics required for calculating this is already being measured for use in Corsi and Qualcomp, why haven't this been put together before? I'd say it could be a good way of measuring a player's defensive performance. Of course it still doesn't account for playing style, but the guys that have bad Corsi Rel due to playing difficult minutes could be accurately measured in how good they are at containing the opposition.

Addendum:

This was something that just popped into my head last night, and after mulling it over and doing some calculations this morning I realized it was an incomplete measure of how a player affects the flow of play since the QoT wasn't accounted for.

A good measure would be the difference of expected Corsi and the Corsi of the player in question.

Expected Corsi would be: Corsi Qot - Corsi QoC. If your teammates have an average Corsi of -4 and the opposition an average Corsi of 2, the expected Corsi would be a -6.

If a player in that situation would have a Corsi of -2, it would be (-2) - (-6) = 4 point higher than expected, i.e. this player pushed the flow of play 4 points in the favour of his team when he was on the ice.

Corsi On - (Corsi QoT - Corsi QoC)

Click to read more...

 

[Blue Blooded]

I did some calculations on the 2010-2011 season among players that faced CorsiRelQoC greater or equal to 1:

E: Even with CODE-tags the formatting was horrible.

Only Lidström (1,275), Erat (0.445), and Ward (0.024) managed to beat expected Corsi among those who faced the opposition's best players. Suter (-0,066) and Weber (-1.655) were barely minus.

And CorsiQoC greter than or equal to 1:

E: Same here.

Of those who faced the toughest competition from a Corsi perspective Datsyuk (11.132) and Zeterberg (7.431) did the best job. Bergeron was also not an unexpected face at the top with his (6.194).

At the bottom we can also see that Keith Aulie was given some very tough minutes that he clearly couldn't handle being a whopping 25 points under expected Corsi.

 

[Blue Blooded]

The top 30 in CorsiRelQoC from 2011-2012:

Name        	         Corsi On	Corsi QoT	Corsi QoC	 +/- Exp.
Thornton    	 14,90	 6,900	0,977 	 8,977
Weber        	 -0,90	-6,090	2,015 	 7,205
Tyutin        	  2,32	        -3,309	1,356 	 6,985
Pavelski       	 12,32	 7,462	1,150 	 6,008
Marleau      	 10,22	 6,681	1,040 	 4,579
Lidström     	 15,24	11,463	0,759 	 4,536
Elias           	  4,44	         1,644	0,920 	 3,716
Phaneuf      	 -0,36	-2,559	1,049 	 3,248
McDonagh    	 -3,03	-4,283	1,442 	 2,695
Seabrook     	  7,67	         6,139	1,099 	 2,630
Jo. Staal     	 12,02	10,195	0,264 	 2,089
Girardi        	 -3,88	-4,207	1,400 	 1,727
S. Kostitsyn	 -8,59	-7,402	1,926 	 0,738
Smid          	 -5,60	-5,036	1,101 	 0,537
Erat           	 -9,05	-7,386	2,137 	 0,473
Horcoff       	 -6,82	-5,832	1,223 	 0,235
Butler	                 -8,74	-6,598	1,883 	-0,259
Glencross     	 -9,09	-7,185	1,116 	-0,789
Gorges       	 -4,91	-2,947	1,068 	-0,895
Carter        	 -1,56	 0,396	0,590 	-1,366
Jokinen        	-10,25	-7,085	1,658 	-1,507
Keith          	  4,02	         6,968	1,335 	-1,613
Bouwmeester	 -9,38	-6,193	1,427 	-1,760
Dwyer        	 -8,53	-5,036	1,179 	-2,315
Bra. Sutter   	 -8,90	-4,909	0,829 	-3,162
Iginla         	-11,53	-6,814	1,287 	-3,429
Laich          	 -7,49	-3,256	0,729 	-3,505
Fisher	                -12,87	-6,473	2,281 	-4,116
Alzner         	 -7,99	-3,036	0,781 	-4,173
Bolland        	 -1,35	 4,827	1,270 	-4,907

The top 30 in CorsiQoC would basically only be Nashville, Calgary, and Columbus players and would not be that interesting to look at. I'll be able to do more comprehensive studies on 2011-2012 when behindthenet.ca releases an .xls file for that season. Now it's just too much work.

 

[wgknestrick]

There are a lot of Corsi-derived statistics out there, but I can't find any that measures the difference in Corsi of their opponents between their average and when they match up against the player in question.

Example: Daniel Sedin had a Corsi On of 21.09. Let's say that when he played against Rob Scuderi he only managed 13.09, this would give Scuderi a Corsi Opp of 8.00 vs Daniel Sedin. His total Corsi Opp would be the average of all the players he has been on the ice against weighted for icetime.

All the statistics required for calculating this is already being measured for use in Corsi and Qualcomp, why haven't this been put together before? I'd say it could be a good way of measuring a player's defensive performance. Of course it still doesn't account for playing style, but the guys that have bad Corsi Rel due to playing difficult minutes could be accurately measured in how good they are at containing the opposition.




Addendum:

This was something that just popped into my head last night, and after mulling it over and doing some calculations this morning I realized it was an incomplete measure of how a player affects the flow of play since the QoT wasn't accounted for.

A good measure would be the difference of expected Corsi and the Corsi of the player in question.

Expected Corsi would be: Corsi Qot - Corsi QoC. If your teammates have an average Corsi of -4 and the opposition an average Corsi of 2, the expected Corsi would be a -6.

If a player in that situation would have a Corsi of -2, it would be (-2) - (-6) = 4 point higher than expected, i.e. this player pushed the flow of play 4 points in the favour of his team when he was on the ice.

Corsi On - (Corsi QoT - Corsi QoC)

You also have to account for shot quality allowed on ice. (and yes I've heard all the arguments that SV% and SH% are not repeatable across the entire NHL population). I don't buy it for D men population though. They have much more impact of what goes on in front of the netminder than forwards who just add to the noise.

On ice SV% 5v5 the last 4 years:

Dman A
.924
.930
.927
.936

Dman B
.896
.915
.909
.905

Example A is Rob Scuderi (defensive Dman with relatively - corsi)
Example B is Alex Goligoski (offensive Dman) with + corsi)

The last year listed (08-09), they were both playing in front of the same goaltender, yet Scuds "lucked" (sarcasm) out by having a huge SV% advantage (.031) while on ice, AND while facing tougher qual comp.


There are players that allow a lot of low quality shots (Good defensive only Dmen) and players that allow few, but high quality chances (offensive defensemen). Then you have your elite that are good at both and your scum which aren't good at anything.

2 parts


- # of shots allowed vs expected
- SV% of goalie on ice vs expected

 

[seventieslord]

You also have to account for shot quality allowed on ice. (and yes I've heard all the arguments that SV% and SH% are not repeatable across the entire NHL population). I don't buy it for D men population though. They have much more impact of what goes on in front of the netminder than forwards who just add to the noise.

On ice SV% 5v5 the last 4 years:

Dman A
.924
.930
.927
.936

Dman B
.896
.915
.909
.905

Example A is Rob Scuderi (defensive Dman with relatively - corsi)
Example B is Alex Goligoski (offensive Dman) with + corsi)

The last year listed (08-09), they were both playing in front of the same goaltender, yet Scuds "lucked" (sarcasm) out by having a huge SV% advantage (.031) while on ice, AND while facing tougher qual comp.


There are players that allow a lot of low quality shots (Good defensive only Dmen) and players that allow few, but high quality chances (offensive defensemen). Then you have your elite that are good at both and your scum which aren't good at anything.

2 parts


- # of shots allowed vs expected
- SV% of goalie on ice vs expected

I agree that Corsi, adjusted for strength of teammates and opposition, and based on quality of shots allowed, not just quantity, would be as perfect a stat as you could get.

And we have everything we need to calculate this, IIRC, so what are we waiting for?

then again there is one hitch. Corsi is based on all shots directed at the net, not just SOG. Shot quality "rates" shots based on their likelihood of scoring, with a base of 1.0 being an "average" shot. tough shots are higher, easy shots are lower. what are missed shots? how do you assign them a rating? it would have to be based on a percentage chance that the missed shot could have been a legitimate shot, based on more data. also, a blocked shot would have to still count as a shot directed at the net (which is bad) but could be counted as "less of a shot" since it didn't get there (which is good). that's all well and good, but then what is it? 0.2? 0.9? somewhere in between? It would have to be somehow quantified logically.

 

[Pietraneglo222]

The problem with Corsi stats against certain opponents is that you have a very small sample size.

 

[seventieslord]

The problem with Corsi stats against certain opponents is that you have a very small sample size.

but the aggregate is a large sample size. Yes, your final result is based on just 10 minutes against player A, 8 minutes against player B, etc... but the total is all weighted properly and based on 2000 minutes.

same thing as shot quality. don't focus on the fact that it includes "only" 12 point blank wrist shots and "only" 14 wraparound attempts and "only" 22 slap shots from the point. it's still a large sample that includes 1000+ shots in total.

In other words, I don't care what someone's Corsi was in particular against a certain player. I want to know about their season-long tendencies.

 

[DL44]

Last year i put together a formula to calculate the top rated Defensive Dmen...
Obviously a very cumbersome task as we all know.

This is how it worked out......
http://hfboards.mandatory.com/showthread.php?t=985575&highlight=


Based on the 2010-2011 season

- 50 games min played so i had a manageable sample size (173 players).

5-on-5 formula weighing each category differently...

=5*(M5)+4*(G5)+3*(V5)+2*(AB5)+1.75*(AI5)+1.5*(AP5) +Y5

M5 - ice time
G5 - qualcomp
v5 - Diff in GA On vs Off
ab5 - Corsi rel
AI5 - Diff in Blocked shots On vs Off
AP5 - Diff in Zone fin and Zone start
Y5 - PDO

Results: 5on5:

1. DAN HAMHUIS
2. ZDENO CHARA
3. TONI LYDMAN
4. KEVIN BIEKSA
5. LUBOMIR VISNOVSKY
6. DREW DOUGHTY
7. JOHN CARLSON
8. RYAN SUTER
9. BRENT BURNS
10. MATTHEW CARLE


Just tried to take into account numerous defensively related categories and go from there...


Would love to someone apply this to last yr's qualified dmen..

(All numbers were obviously from behind the net)

 

[seventieslord]

hmmm, in theory it looks ok, and it appears to identify "some" of the correct players. However, why do four clear offensive specialists get on there?

Something about QoC not being factored in properly, I imagine. Perhaps they were quite good at not getting scored on by second rate forwards, but that should almost preclude them from being compared to the guys who played against first rate forwards. Just thinking out loud.

 

[DL44]

hmmm, in theory it looks ok, and it appears to identify "some" of the correct players. However, why do four clear offensive specialists get on there?

Something about QoC not being factored in properly, I imagine. Perhaps they were quite good at not getting scored on by second rate forwards, but that should almost preclude them from being compared to the guys who played against first rate forwards. Just thinking out loud.

Right on... i was starting from stratch, so i'm sure there are a bunch of modifications that can be made to improve the formula.

Those offensive specialists were actually forced to play in their zone and did pretty well.
When i have access to my spreadsheet (work computer) I'll post their ranking in each category to show how they ranked...
conversely if there is another specific dman you were expecting to be ranked, post, i;ll post their numbers to show how they placed...

 

[GKJ]

Last year i put together a formula to calculate the top rated Defensive Dmen...
Obviously a very cumbersome task as we all know.

This is how it worked out......
http://hfboards.mandatory.com/showthread.php?t=985575&highlight=


Based on the 2010-2011 season

- 50 games min played so i had a manageable sample size (173 players).

5-on-5 formula weighing each category differently...

=5*(M5)+4*(G5)+3*(V5)+2*(AB5)+1.75*(AI5)+1.5*(AP5) +Y5

M5 - ice time
G5 - qualcomp
v5 - Diff in GA On vs Off
ab5 - Corsi rel
AI5 - Diff in Blocked shots On vs Off
AP5 - Diff in Zone fin and Zone start
Y5 - PDO

Results: 5on5:

1. DAN HAMHUIS
2. ZDENO CHARA
3. TONI LYDMAN
4. KEVIN BIEKSA
5. LUBOMIR VISNOVSKY
6. DREW DOUGHTY
7. JOHN CARLSON
8. RYAN SUTER
9. BRENT BURNS
10. MATTHEW CARLE


Just tried to take into account numerous defensively related categories and go from there...


Would love to someone apply this to last yr's qualified dmen..

(All numbers were obviously from behind the net)

Just a question about this, where exactly are you getting the evidence in how much each stat is weighed?

 

[begbeee]

Last year i put together a formula to calculate the top rated Defensive Dmen...
Obviously a very cumbersome task as we all know.

This is how it worked out......
http://hfboards.mandatory.com/showthread.php?t=985575&highlight=


Based on the 2010-2011 season

- 50 games min played so i had a manageable sample size (173 players).

5-on-5 formula weighing each category differently...

=5*(M5)+4*(G5)+3*(V5)+2*(AB5)+1.75*(AI5)+1.5*(AP5) +Y5

M5 - ice time
G5 - qualcomp
v5 - Diff in GA On vs Off
ab5 - Corsi rel
AI5 - Diff in Blocked shots On vs Off
AP5 - Diff in Zone fin and Zone start
Y5 - PDO

Results: 5on5:

1. DAN HAMHUIS
2. ZDENO CHARA
3. TONI LYDMAN
4. KEVIN BIEKSA
5. LUBOMIR VISNOVSKY
6. DREW DOUGHTY
7. JOHN CARLSON
8. RYAN SUTER
9. BRENT BURNS
10. MATTHEW CARLE


Just tried to take into account numerous defensively related categories and go from there...


Would love to someone apply this to last yr's qualified dmen..

(All numbers were obviously from behind the net)

I'm not interested in numbers much, but how can you explain Visnovsky as TOP5 defensive defensman?

 

[DL44]

Just a question about this, where exactly are you getting the evidence in how much each stat is weighed?

The goal was to select only statistical categories which reflected the defensive part the the game. So thats difficult in itself.

But after selecting the handful of categories above, I.took poll and tried to determine which categories were relatively considered more important.

So that's the subjective part. In an earlier version, I ran the numbers as all equal, but there was obvious difference in importance of some of the stats and a significant effect of superior team play.... But I.did love the look of the shorthanded results of the list...

Quality of competition Corsi quality of Comp Relative Corsi ... ( i don't know the difference between relative and regular) (Goals against while Off/60) - (Goals against while On/60) Blocked shots PDO - Team shooting % + team save % while player is On the ice Zone start % Difference in Zone finish % vs zone start %

Categories weighed equally, with significant weight on ice time.

Top 10 5on5 Defensive Defensmen:
1. Zdeno Chara 2. Ryan Suter 3. Dan Hamhius 4. Marc Staal 5. Matt Carle 6. Dan Garardi 7. Brett Clarke 8. Dion Phaneuf 9. Toni Lydman 10. Travis Harmonic

Top 10 Shorthanded Defensive Defensmen:
1. Hal Gill 2. Brooks Orpik 3. Zbynek Michalek 4. Willie Mitchell 5. Marc Staal 6. Mike Weaver 7. Jason Garrison 8. Dan Hamhuis 9. Stephan Robidas 10. Nik Kronwall

Top 10 Overall Defensive Defensmen: 1. Dan Hamhuis 2. Marc Staal 3. Dan Girardi 4. Greg Zanon 5. Toni Lydman 6. Brent Burns 7. Zdeno Chara 8. Kimmo Timonen 9. Willie Mitchell 10. Ryan Suter

 

[seventieslord]

Right on... i was starting from stratch, so i'm sure there are a bunch of modifications that can be made to improve the formula.

Those offensive specialists were actually forced to play in their zone and did pretty well.
When i have access to my spreadsheet (work computer) I'll post their ranking in each category to show how they ranked...
conversely if there is another specific dman you were expecting to be ranked, post, i;ll post their numbers to show how they placed...

Robidas and Mitchell, for starters.

I'm not interested in numbers much, but how can you explain Visnovsky as TOP5 defensive defensman?

it looks like the factors chosen, although they all make sense, were just not weighted "properly". It was a subjective exercise and it appeared to work, to some degree.

the CORSI part is probably a major flaw. Just like people correctly say "you can't use +/- as an indication of defensive ability because goals for are half the equation", the same is true about CORSI, only with shots instead of goals. So these offensive specialists are getting shots to the other team's net and advancing the puck, and helping their corsi, but not really proving themselves defensively in the process.

also, I see that zone finishes minus zone starts is a factor. I think this is a mistake. If players like Matt Carle and Visnovsky are getting a defensive zone start, it won't be against the best players. therefore, they will defend the zone, and advance the puck for the next faceoff. Doesn't prove they're that good defensively. I would use raw zone starts. I know that doesn't say what happened on that shift, BUT the fact that their coach is using them in that situation on a regular basis says much more, IMO.

I'm not sure PDO is important to use. It's very random. It's true that some players will "earn" better PDOs by limiting shot quality against and creating higher quality scoring chances, but that isn't extremely common, and PDO is still a reflection of both ends of the ice, like CORSI.

DL44 - It sounds like you have these numbers all in a spreadsheet. If you do, try this subjective weighting and get back to me with a top-20:

=10*(G5)+6*(ZS)+4*(M5)+3*(AB5)

basically, who's getting the big minutes in the defensive zone against the best players, and to a lesser extent, who is still helping out on the CORSI side (keeping corsi in there does introduce a little bit of offensive "noise" but only about 6.5% of the total)

 

[Doomsday Device]

also, I see that zone finishes minus zone starts is a factor. I think this is a mistake. If players like Matt Carle and Visnovsky are getting a defensive zone start, it won't be against the best players. therefore, they will defend the zone, and advance the puck for the next faceoff.

I'm not sure why you're assuming this. Visnovsky and Lydman faced easily the toughest quality of competition on Anaheim in 2010-11 and absolutely killed it. Visnovsky might not have a reputation as a strong defensive player who matches up against top lines but that's how he was used that season.

 

[wgknestrick]

Cough, cough GAA/60 corrected for goaltending and qual comp.

Output of shots against and SV% on ice in one easy to carry stat.

 

[seventieslord]

I'm not sure why you're assuming this. Visnovsky and Lydman faced easily the toughest quality of competition on Anaheim in 2010-11 and absolutely killed it. Visnovsky might not have a reputation as a strong defensive player who matches up against top lines but that's how he was used that season.

Thanks for the correction. I think we're getting to the bottom of it.

Lydman is the one with the exemplary defensive reputation, going back almost a decade now. If Visnovsky was his partner, then to some degree he was carried to a better defensive rating. Of course, like all pairings, they weren't joined at the hip, and in the numbers you can see, Visnovsky got some more offensive shifts against lesser competition and boosted his corsi, so he'd look better by the metrics DL44 is using.

usually he has faced the weaker competition based on qualcomp in 08, 09, 10, and 12, but you're right, he did get tougher shifts in 2011.

 

[DL44]

5-on-5

Visnovsky
M5 - ice time - 37th
G5 - qualcomp - 33rd
v5 - Diff in GA On vs Off - 2nd
ab5 - Corsi rel - 2nd
AI5 - Diff in Blocked shots On vs Off - 155nd
AP5 - Diff in Zone fin and Zone start - 63rd
Y5 - PDO - 13th



Mitchell
M5 - ice time - 26th
G5 - qualcomp - 10th
v5 - Diff in GA On vs Off - 135th
ab5 - Corsi rel - 96th
AI5 - Diff in Blocked shots On vs Off - 129th
AP5 - Diff in Zone fin and Zone start - 74th
Y5 - PDO - 64th


Robidas
M5 - ice time - 31st
G5 - qualcomp - 5th
v5 - Diff in GA On vs Off - 125th
ab5 - Corsi rel - 37th
AI5 - Diff in Blocked shots On vs Off - 136th
AP5 - Diff in Zone fin and Zone start - 25th
Y5 - PDO - 110th

 

[DL44]

=10*(G5)+6*(ZS)+4*(M5)+3*(AB5)

What's the typo? assuming V5?

 

[DL44]

With your formula it looks like: ( 2010-2011 season)

1 DANHAMHUIS
2 ZDENOCHARA
3 TONILYDMAN
4 DREWDOUGHTY
5 LUBOMIRVISNOVSKY
6 KEVINBIEKSA
7 JOHNCARLSON
8 RYANSUTER
9 TRAVISHAMONIC
10 JASONGARRISON
11 ALEXPIETRANGELO
12 MIKEWEAVER
13 KARLALZNER
14 CHRISTIANEHRHOFF
15 STEPHANEROBIDAS
16 MARCMETHOT
17 MARCSTAAL
18 DUNCANKEITH
19 JAROSLAVSPACEK
20 BRENTSEABROOK

Vs the original:

1. DAN HAMHUIS
2. ZDENO CHARA
3. TONI LYDMAN
4. KEVIN BIEKSA
5. LUBOMIR VISNOVSKY
6. DREW DOUGHTY
7. JOHN CARLSON
8. RYAN SUTER
9. BRENT BURNS
10. MATTHEW CARLE

Burns downs to 26th..
Carle down to 52nd.. Hmmm... not exactly sure why. anyways...

The old spreadsheet is a mess right now... maybe i'll re-do it with better suggestions that come out of this thread...

 

[seventieslord]

What's the typo? assuming V5?

no, sorry, I meant to clarify, ZS = zone starts, a variable you didn't define yourself. you'd need to use a 1/ZS or something though, to assign higher value to a lower number, obviously.

 

[DL44]

no, sorry, I meant to clarify, ZS = zone starts, a variable you didn't define yourself. you'd need to use a 1/ZS or something though, to assign higher value to a lower number, obviously.

Like i said in my edit, when i get a chance, i'll redo the spreadsheet with last yr's numbers...
SO i'll redo the formula with what people think should be appropriate...


In terms of what i did for rankings.. for each category the players were ranked 1 thru - 173. It was thos rankings i used in the formula... not the actual raw numbers.

 

[seventieslord]

In terms of what i did for rankings.. for each category the players were ranked 1 thru - 173. It was thos rankings i used in the formula... not the actual raw numbers.

Oh, I see.

I'm not sure how that would affect it, but it would certainly insulate exceptionally good and bad players from looking like they deserve, and it would also make a player not too much above average who happens to be 24th out of 173 and only 10% better than the guy in 140th, look a lot better than they deserve.

These numbers should probably be averaged, then each factor for each player can be expressed as a percentage of that average.

 

[DL44]

Oh, I see.

I'm not sure how that would affect it, but it would certainly insulate exceptionally good and bad players from looking like they deserve, and it would also make a player not too much above average who happens to be 24th out of 173 and only 10% better than the guy in 140th, look a lot better than they deserve.

These numbers should probably be averaged, then each factor for each player can be expressed as a percentage of that average.

thought about that..
that's where sample size comes in a bit - over a range of 173 players, there really isn't much tiering of players... its a pretty gradual progression.

And if there was to be tiering in any particular category where a player jumps up as an outlier vs the comp, its effects are squashed out with the multiple categories.



So i was going to start with a new spreadsheet based on the 2011 - 2012 season...

What do we suppose the categories should be and the make up of the formula we implement?

 

[seventieslord]

thought about that..
that's where sample size comes in a bit - over a range of 173 players, there really isn't much tiering of players... its a pretty gradual progression.

And if there was to be tiering in any particular category where a player jumps up as an outlier vs the comp, its effects are squashed out with the multiple categories.



So i was going to start with a new spreadsheet based on the 2011 - 2012 season...

What do we suppose the categories should be and the make up of the formula we implement?

Your guess is as good as mine. you started with something, I made what appeared to be improvements, but until I see the results I don't even know if they really are. In any case it's probably just subjective weighting of objective numbers. The heavy stats guys tend to have ways to logically lay out how different factors should be weighted.