Tuesday, October 30, 2007

De-fence

First I'll note that Behind The Net has updated as of Oct. 29, and I've also passed along a lengthy list of requests for other stats to be added to the site.

We'll see how that goes.

I've looked at defensive performance almost ad nauseam on this site in the past; call it my own little fascination, if you will. The greatest thing about a resource like BtN is that we can now measure what was previously unquantifiable (other than with the extremely limited plus-minus figure), and since I've always had a soft spot for the unheralded defensive defenders, that makes me giddy.

To me, many statheads make this far more complicated than it has to be: On some basic level, defensive performance simply comes down to playing the most minutes without having a goal go in your net. Period.

The leaders and the bottom feeders, 13 per cent of the way through 2007-08.

Defencemen with best 5-on-5 goals against average (among top 100 in ice time):


NAMETEAMGPTOI/60Q.COMPGAON/60
1ROSTISLAV KLESLACBJ1015.570.080.77
2DEREK MORRISPHX916.610.231.2
3DANIEL GIRARDINYR1014.750.051.22
4KENT HUSKINSANA1313.99-0.091.32
5JAROSLAV MODRYL.A1214.61-0.051.37
6ADRIAN AUCOINCGY1115.9-0.11.37
7ANDREJ MESZAROSOTT1016.28-0.11.47
8NICKLAS LIDSTROMDET1216.030.341.56
9BRIAN RAFALSKIDET1215.790.311.58
10BRETT LEBDADET1212.58-0.11.59
11BRAYDON COBURNPHI916.4901.62
12TOBIAS ENSTROMATL1113.390.071.63
13KIMMO TIMONENPHI916.210.121.64
14F. BEAUCHEMINANA1316.210.161.71
15CHRIS PRONGERANA1315.740.111.76
16DUNCAN KEITHCHI1117.940.171.82
17ANDREW FERENCEBOS1016.35-0.071.83
18DION PHANEUFCGY1117.6801.85
19ZDENO CHARABOS1016.070.241.87
20ROMAN HAMRLIKMTL1015.7-0.051.91

By all accounts, big Rusty Klesla is having the year everyone in the Columbus organization thought he would when he was picked fourth overall in 2000. (Unfortunately many of those souls are no longer around to see it.)

That he (and a Coyotes defenceman!) lead all minute munching blueliners in goals against per 60 minutes played is pretty darn significant if you ask me.

Also noteworthy is the fact there are quite a few guys on here who play on struggling teams, namely rookie Tobias Enstrom with those poor Thrashers.

Derek Morris deserves a medal or something: He's played big minutes against brutal opposition on what is easily the worst team in the league, yet the puck's never going in when he's on the ice.

Honestly, does anyone know what's going on in Phoenix?


Defencemen with worst 5-on-5 goals against average (among top 100 in ice time):


NAMETEAMGPTOI/60Q.COMPGAON/60
1KEVIN BIEKSAVAN1113.650.094.8
2TREVOR DALEYDAL1015.070.174.78
3ALEXEI ZHITNIKATL1114.3-0.114.58
4LUBOMIR VISNOVSKYL.A1215.340.084.56
5HENRIK TALLINDERBUF1014.640.234.51
6MATT GREENEEDM1213.81-0.074.35
7BRETT CLARKCOL1115.220.24.3
8BRENT BURNSMIN1115.580.064.2
9ROB BLAKEL.A1213.270.194.14
10J-M LILESCOL1113.81-0.13.95
11KIM JOHNSSONMIN1115.240.043.94
12SCOTT HANNANCOL1117.070.223.83
13BRYAN ALLENFLA1115.720.053.82
14DAN HAMHUISNSH1015.880.043.78
15RYAN WHITNEYPIT1014.360.023.76
16FRANK KABERLECAR1113.050.033.76
17GARNET EXELBYATL1116.04-0.053.74
18RADEK MARTINEKNYI916.080.013.73
19BRENDAN WITTNYI916.2203.7
20IAN WHITETOR1213.5703.68
21KYLE MCLARENS.J1014.68-0.063.68

So much for my crowing about the year Brent Burns has had in Minnesota.

Kevin Bieksa's really been having a tough go of it in Vancouver, apparently feeling the pressures that go along with that new big deal. Last year, they were calling him Jovo Jr. out West in reference to the recently departed ex-Canuck Ed Jovanovski, but now Bieksa's looking an awful lot like the Jovo Sr. of today.

Big, young rearguards like Matt Greene and Garnet Exelby are getting relatively easy minutes and still being punished, while it's been a terrible season for Alexei Zhitnik (pair him with Enstrom maybe?).

Players like Henrik Tallinder, Scott Hannan and Kim Johnsson were all leaders in this category last season, so it's been quite a swing for them. All three get the tough assignments and have been beaten up a bit.

N.B. Before the message boards go bonkers with stuff like 'Beaksa's the worstest!' and the like, it's worth noting that these are only defenders from the Top 100 in minutes played so far. Among all 227 defencemen who have played this season, Bieksa's 4.8 GA/60 is 12th worst, with players like Ladislav Smid (8.01), Maxim Kondratiev (6.36), Anders Eriksson (6.17) and Steve McCarthy (6.01) all pulling up the real rear.

Smid is the worst among all players who have played at least 40 5-on-5 minutes, while Derek Roy (6.33) is worst among those with 100+. Buffalo's defensive numbers are way off last year's pace, and Chris Drury's doing just fine in the Big Apple so far.

Coincidence?

24 Comments:

At 9:19 AM, October 30, 2007, Blogger voxel said...

The funny thing about Smid is that he's become a high-event player!

Not only has he got a high GAON/60, he's got a high GFON/60. He's been contributing offensively since his call up from the AHL.

I don't value defensive defensemen as much as you, so I'd like to see the GFON/60 non-zero and the +-ON/60 in the positive range.

Randy Jones was doing great until his suspension

 
At 11:13 AM, October 30, 2007, Anonymous PPP said...

And people still like Ian White.

 
At 12:13 PM, October 30, 2007, Anonymous Anonymous said...

Too bad stats won't show who they're playing against.

Tallinder and Lydman plays every 5-5 situation against Heatley-Spezza-Alfredsson, Jagr line, Briere line, Koivu line, Sundin line etc.

Some D men face lesser lights.

 
At 12:38 PM, October 30, 2007, Blogger Pleasure Motors said...

Too bad stats won't show who they're playing against.

Right, except for that little quality of competition thing right beside the GAON/60 stat, which very clearly shows what kind of competition they're playing against.

 
At 2:22 PM, October 30, 2007, Blogger J. Michael Neal said...

To me, many statheads make this far more complicated than it has to be:

The reason we do that is that this is a more complicated question than you are making it out to be. How many goals you let in is not just a function of your own performance. It's also a function of the performance of the teammates you were on the ice with, and the performance of the opposing players you were on the ice with. The stats try to account for this, but they do so in a problematic way.

It tries to measure the quality of your teammates and your opponents by how they perform, using the same metric. Of course, how each of those guys perform is also affected not only by the other guys on the ice are doing, but also by your performance. Given that Andy MacDonald was on the ice more than 90% of the time that Teemu Selanne was on the ice, trying to separate their effects is extremely difficult.

You have a system in which there are complicated interactions between a number of moving pieces. There are statistical tools that can be used to try to account for this. They rely upon the concept of recursion, and the Bradley-Terry method is a good example. This is used in KRACH, or Ken's Rankings for American College Hockey. More information on it can be found in the KRACH FAQ.

If you don't use some form of recursion, you can't separate out the pollution of the various interactions. You can rig up the system so that the numbers come out looking like they do now, and can be analyzed just like they are now. There's no reason that interpretations of the data need to be any more complicated. Generating the data does have to be more complicated.

 
At 2:33 PM, October 30, 2007, Blogger James Mirtle said...

Nope. The question I'm asking is 'which defencemen have been on the ice for the fewest goals against? Which ones have been on the ice for the most?'

That is the question.

 
At 2:47 PM, October 30, 2007, Anonymous ken said...

Anders Eriksson? Seriously, how is that guy in the NHL? If ever there was proof the NHL has expanded too much and diluted the quality of talent, it's right there.

 
At 3:12 PM, October 30, 2007, Anonymous Karina said...

So true... Anders Eriksson couldn't even crack the Leafs' defense when he was in the system... I have no idea what Sutter saw in him!

 
At 3:23 PM, October 30, 2007, Anonymous twain said...

Yah, but Mirtle, then it is a barely meaningful question dressed up as a serious one--because we all know people will read this and think "defensive prowess/liability", and that your question (who was on the ice for the most goals against) can't really get at that.

One of the most notable variables here is save percentage, of course, and that is unaccounted for by these stats.

In short, Mirtle, I think your last quip is insincere.

 
At 4:07 PM, October 30, 2007, Blogger James Mirtle said...

I respectfully disagree.

I'm not exactly comparing the 80th defender to the 91st. There's a huge separation here between the top and bottom, one that can't be accounted for, or eliminated, but factoring in teammates, goaltenders or the like.

This data tells us the kind of year Kevin Bieksa is having, and the kind of year Rusty Klesla is having, and no amount of rejigging will change that.

Derek Morris, the Coyotes have been telling us, is having a great year, but have we seen a single statistic this season telling us so?

Is it perfect? Heck, no. But I find it useful (keeping in mind the limitations) and others do, too.

 
At 4:37 PM, October 30, 2007, Blogger J. Michael Neal said...

Is it perfect? Heck, no. But I find it useful (keeping in mind the limitations) and others do, too.

Yes, but how useful is it? We already knew that Derek Morris was doing pretty well. That's indicated by +4 on a very bad team, using the old, crude plus/minus system. This stat isn't telling us very much that looking at less complicated stats with a critical eye wasn't telling us. In roughly the same way, we can tell that Kevin Bieksa certainly isn't having a good year.

Given its limitations, this stat isn't really improving our knowledge. It simply gives the illusion of improving our knowledge.

 
At 4:43 PM, October 30, 2007, Anonymous twain said...

I kind of agree with all that--but I think that it leads you beyond your last stated question, insofar as you are now introducing evidence to suggest "what kind of year players are having".

My point is simply that you and the statheads _are_ asking the same question, and thus more sophisticated approaches give better answers to the same question. But this data is still useful, to a point, in answering that question, if one is willing to bring some obvious interpretive tools to bear on the raw data, and to concede it useful from distinguishing molehills from mountains, but nothing more fine tuned than that.

Anyways, I digress, because I am, well, starting to make a mountain out of a molehill.

 
At 4:48 PM, October 30, 2007, Blogger James Mirtle said...

But this data is still useful, to a point, in answering that question, if one is willing to bring some obvious interpretive tools to bear on the raw data, and to concede it useful from distinguishing molehills from mountains, but nothing more fine tuned than that.

That's all I'm doing guys; not reinventing the wheel here. I wrote this post in 10-15 minutes.

If someone else wants to do a more detailed analysis, have at it. I'm happy with what's here.

 
At 7:29 PM, October 30, 2007, Blogger Hawerchuk said...

Respectfully, I must disagree with J. Michael Neal. The point of these stats is to give a coarse estimate of a player's performance, the contribution of his teammates, and the level of competition he faces. We also want to keep the overall metrics as simple as possible.

Once you introduce recursion or power rankings into the mix, you lose the relationship between the data you collect and the final statistic.

If you wanted a better way to measure a player's performance, I wouldn't suggest for a second that more computation is the way to go.

As with baseball, football and basketball, the key would be to increase your observations of the actual game. We would want to know who has the puck at all times, where it is, and lots of information about passing and positioning.

Recursion's not going to get you there. BTW, if you'd like to run an alternate algorithm on the dataset, I'd be happy to send it to you.

 
At 7:36 PM, October 30, 2007, Anonymous Rod said...

Something is a bit off with the numbers though.

Smid's an easy one to look at given that he's played only three games:
- 8.01 GAON/60, 14.98 minutes per game

That works out to six goals against (8.01 / 60 * 14.98 * 3)...but Smid's been on the ice for four goals against at 5 on 5, plus two more while on the PK.

Perhaps penalty killing goals against aren't getting filtered out of the numbers.

Then again, I'm probably doing something completely wrong. If I use the same process, Smid's GFON/60 makes even less sense:
(9.34 / 60 * 14.98 * 3) = 7

But, Smid's been on the ice for four goals-for. All four @ 5 on 5, but no PP goals or otherwise. Not sure how seven fits into the picture.

Most logical explanation...I'm doing this wrong. :)

 
At 7:47 PM, October 30, 2007, Blogger J. Michael Neal said...

Respectfully, I must disagree with J. Michael Neal. The point of these stats is to give a coarse estimate of a player's performance, the contribution of his teammates, and the level of competition he faces. We also want to keep the overall metrics as simple as possible.

Sure. Is this giving us a better coarse estimate than even simpler stats would give us? My speculation is that simply taking (Goals while on ice/Team's total goals allowed)/(Average time on ice/60) would give us almost as good an estimate.

 
At 8:59 PM, October 30, 2007, Blogger voxel said...

Smid's has not been on a goal against while on the PK.

For all Smid ES events(including 4-on-4):

4 ESGF
4 ESGA

while playing a total of 50.25 ES minutes.

The BehindTheNet numbers filter out 4-on-4 and when goalies are pulled I believe.

In terms of raw ES GA events

Klesla 3
Morris 3
Bieska 11

As you can see a single game (i.e Canucks vs. Philly) can have adverse effects on the stats.

As for Brent Burns... he's still a STUD. This is the problem with using GA to rate a defensemen... because when Burns in on ice - offense happens.

 
At 9:02 PM, October 30, 2007, Blogger James Mirtle said...

Yeah, Burns still has had a breakout year — he's just got some room to improve, is all.

 
At 10:57 PM, October 30, 2007, Anonymous Rod said...

voxel: OK, I got the PK side wrong...not sure what I was looking at. Unfortunately that probably distracted you from the point I was trying to make.

Both you and I said Smid's been on the ice for:
- 4 five-on-five goals scored for
- 4 five-on-five goals scored against

Well, BtN numbers indicate six goals against, and seven goals for...unless I messed up the math. Meaning there something going wrong in the calculation of the GFON/60 and GAON/60 numbers.

From another angle, if Smid has been on the ice for the same number of goals scored 'for' and 'against' when 5on5, that should result in identical GFON/60 and GAON/60...right?

BtN:
9.34 GFON/60
8.01 GAON/60

Something's not quite right. Like I said before, most logical explanation...I'm doing this wrong, or interpreting things incorrectly. :)

Further clarification appreciated. Thanks.

 
At 5:30 AM, October 31, 2007, Anonymous Rod said...

Looks like I was right that something was amiss. BtN's numbers for Smid now read (still after three games):
GFON/60: 5.34
GAON/60: 5.34

As expected, the two are identical.

 
At 12:41 PM, October 31, 2007, Blogger Hawerchuk said...

Please lodge all complaints with the NHL. The first time I grabbed the game data, Smid was apparently on the ice for the three shootout goals.

Grab it again, poof, he's not on the ice anymore. I fixed everything so that nobody can be on the ice for shootout goals...

 
At 12:57 PM, October 31, 2007, Blogger Hawerchuk said...

Sure. Is this giving us a better coarse estimate than even simpler stats would give us? My speculation is that simply taking (Goals while on ice/Team's total goals allowed)/(Average time on ice/60) would give us almost as good an estimate.

You: GAON/(GAON+GAOFF)/TOI.

BtN: GAON/TOI - GAOFF/TOFF.

I'm not sure there's much difference in the significance of the two statistics.

But the point of the second one is pretty basic: to show the difference in a team's +/- when a guy is on the ice, and when he's off the ice.

It's expressed in goals per game, which is something most people will wrap their heads around without complaining about math.

The other point of this site is that I am not trying to do something secret and proprietary and make money off this like hockeyanalysis or thehockeyrating (etc) do. Rod is able to check my numbers and make sure they make sense, and if somebody wants to come up with improved statistics, I'm happy to zip up the entire database and send it to them.

 
At 3:32 PM, October 31, 2007, Anonymous Rod said...

Good work finding the problem! Did the league do that for every shoot-out? Unreal.

 
At 6:27 PM, October 31, 2007, Blogger Hawerchuk said...

The NHL feed isn't the most reliable thing in the world. Game 20107 has no data (CAR-PHI), for example, and if the past is any guide, it will never get fixed. Last year, the Rangers summaries almost never showed the goalie playing in the 3rd period. etc, etc...I just expect them to make mistakes...

 

Post a Comment

<< Home

Links to this post:

Create a Link


.

Free Page Rank Checker
eXTReMe Tracker