Friday, November 27, 2009


Points v. Scoring

I'm having some problems with my site at the moment - I've got my tech guy (or, if my mother is reading this, my "brother") on it but I had a quick post that I wanted to make, so I thought I'd avail myself of Matt and Andy's hospitality and post it here.

There's been some discussion at Lowetide's site about the EV PTS of the various centremen on the Oilers. Brule is doing quite well with 13 so far, followed by Horcoff and Gagner with 7 apiece and Cogliano with 6. Leaving aside, for a moment, the question of whether or not it's fair to call Brule and Cogliano NHL centres, there's a point about points that I thought was worth making.

One of the goals of a hockey team is to score as many goals as it can. The other goal is to allow as few goals as possible. Goals are the currency of the game. Once a goal is scored, we retrospectively allocate credit for it by assigning the players involved. As David Staples has pointed out when responding to the (vociferous) criticism of his errors system, we only allocate credit for the goals are scored and then draw our inferences about offensive ability or performance from that. It strikes me as fair to say, when looking at only a single season, a person could well be misled.

I figured I'd take a look at this in a couple of different ways. I scraped for the 5v5 information for forwards who played at least 40 games in 2007-08 and 2008-09, which gave me a list of 302 guys. I then took a look to see whether or not points/goal had any repeatability. It certainly doesn't seem too - see the chart at left, although there's a caveat to that, in that it seems to me that a lot of the players who are recognized as "star" players seem to have some repeatability in this department. I suspect a lot of the real goons fall into the same category.

For example, Crosby's been in on 92%, 86% and 84% of Penguins even strength goals during the past three seasons. This makes a certain amount of intuitive sense to me - guys like Crosby, who control the game and have the play flow through them, likely drive the offence to a greater degree than your average player. If you could somehow drop him into an even better league than the NHL, in which he was an average player, presumably his "share" of the offence would fall.

Extrapolating from this idea a bit, I thought I'd take a stab at classifying offensive performance into four groups from 2007-08 based on two values: on-ice shooting percentage and points/goal. The average shooting percentage for my group of players was 8.35% in 2007-08 and they averaged points on 69% of 5v5 goals scored while they were on the ice. I classified the players into four groups +S%/+pts/gl, +S%/-pts/gl, -S%/+pts/gl and -S%/-pts/gl. I'm using "+" to mean above average there and "-" to mean below average. There's a graph at left setting this out.

My theory, to the extent that I have one, is this: offensive numbers drive salaries in the NHL. Most players put up a significant portion of their offence at even strength, which makes sense, as that's where the game is (largely) played. If salary is driven, to an extent, by the events of the most recent season, a general manager trying to make the best use of his salary cap dollars should try and avoid signing players to contracts who are coming off of seasons in which they fell into the +S%/+pts/gl grouping, unless those players have a rather strong record of belonging there. For the record - twenty guys fell into that group in both 2007-08 and 2008-09 - the list is:

Sidney Crosby
Evgeni Malkin
Henrik Sedin
Jason Pominville
Mike Ribeiro
Mike Cammalleri
David Booth
Andy McDonald
Marc Savard
Ales Hemsky
Niklas Hagman
Alex Ovechkin
Pavol Demitra
Joe Thornton
Vincent Lecavalier
Ilya Kovalchuk
Daniel Paille
Jarome Iginla

If you were to screen that list for the guys who don't play very much, you'd end up with a pretty solid list of hockey players, guys about whom the general consensus would be that they drive offensive results.

Here's the key chart - it compares the offensive performance of my four groups between 2007-08 and 2008-09. There's something interesting at work here, I think. Note the difference in point production for the first two groups in 2007-08 and 2008-09 - almost half a point per 60. Most people I know would consider that to be significant. The difference in goal production though, is miniscule - an extra goal every 20 hours, an amount of time virtually no forward will play in the course of a season. In 2008-09, things turned around - the second group, which had the lower pts/gl ratio in 2007-08 outscored the group with the higher ratio that year and saw an extra 5v5 goal scored for every ten hours or so that they were on the ice.

If goals scored was what you cared about, you'd do as well picking guys who had above average on ice shooting percentages on the basis of their on-ice GF as you would their points. The difference in the lower two groups persisted - the -/- group actually fared worse in on-ice goals in 2008-09 than they did in 2007-08, although they closed the gap in terms of points a bit.

What can you take from this? Well, if I was a GM, I'd tread carefully when signing guys coming off seasons in the +/+ quadrant, particularly if they wanted +/+ dollars. Unsurprisingly, there are a couple of examples of these guys in Kevin Lowe's current crop of contracts, most notably in the form of Shawn Horcoff and Robert Nilsson. Patrick O'Sullivan got paid coming off of a +/+ quadrant season as well. I've argued extensively elsewhere that Horcoff is not that badly overpaid but it seems reasonable to expect that Lowe should have known that he was looking to sign Horcoff at the worst time possible from the perspective of maximizing the value from his contract.

The broader question I think, is to do with how we should weight and think about points. Much was made of the offensive talents of the Oilers defenders last year. It was sold as a strength of the team, while the forwards were perceived as disappointments. Tom Gilbert, Sheldon Souray and Denis Grebeshkov were all over 1.18 ESP/60, fantastic numbers for defenders and likely above the 90th percentile. Lubomir Visnovsky was at 0.93 ESP/60, also an excellent number and likely above the 80th percentile for defencemen. While Grebeshkov and Gilbert had impressive GF/60 numbers though (3.26 and 3.02, respectively), Souray and Visnovsky were nothing special (2.50 and 2.64, respectively). The question that comes to mind - and I don't claim to have an answer for this - is whether they really had fantastic seasons at ES or whether the peculiarities of how we credit players for offence somehow assigned too much to those guys, leading us to be impressed by numbers that don't tell us anything.

To bring this back to my initial point, the wonderful thing about baseball statistics is that the components of runs - walks, singles, doubles, triples and homers - are easily tracked. It's harder to do in hockey. We trust that individual statistics accurately reflect the offensive contributions made by a given player but I'm not entirely certain that they do, or, to be more precise, that they can be trusted without slicing the numbers and looking at them in different ways.


Good work on showing that a player's GF and points ratio generally regress to the mean.

However, I'm not sure why you assume that GF are driving points, rather than the opposite. Instead of saying "he gets points on an unsustainably high percentage of goals for which he is on the ice", couldn't the statement "his team scores an unsustainably low number of goals in which he is not involved while he is on the ice" be equally or more correct? I don't think your work here has addressed this issue.

There may also be differences within points - goals vs first assists vs second assists.

Your discussion of defenders is interesting, as skilled defenders face a clear tradeoff where they can trade defence for more offence, maybe even more than forwards.

Brian Campbell is an outlier amongst defenders in EV Pts/GF ratio - he has had 0.50 pts on every goal for in his career. Nobody else is close since 1995 - Phil Housley is second with 0.42. Even going back to 1968, only Bobby Orr can match him. Obviously his EV Pts/GF ratio is unsustainably high - he's not Bobby Orr. It's equally obvious that he's trading defence for offence more than most defenders. What ratio should we expect him to regress to? Paul Coffey's 0.44? Sandis Ozolinsh's 0.39? Certainly not the league average of 0.30, or Ken Daneyko's 0.15.

Oops, I botched Campbell's numbers in the last post - his Pts/GF ratio is in fact 0.41, so more Ozolinsh than Orr. But my point still stands that there is considerably more variation with this ratio for defenders than for forwards.

Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?