Wednesday, February 06, 2008

Tough Minutes in Motown

There are a lot of good writers and commentors on the hockey blogs that I read, and more and more of them are referencing Desjardins' Quality of Competition numbers to support their positions. This seems reasonable to me, as I think that the context in which a player's goals, assists and +/- are registered makes a big difference to his value to the team.

A couple of days ago Tyler made a compelling argument for the Oilers being in a better position to improve at 5v5 than many of the other also-rans in the league, based on the fact that the Oiler 5v5 results-getters are the same guys that are staring down the other team's best players in many games. In short, it's easier to rebuild your forward depth if you don't need them to play really tough minutes without bleeding. I'm sure he's right.

The merits of Desjardins' methodology have been discussed at length before, in threads at this site and elsewhere. In short, if you stand too close it looks like a mess. Look at any one game and you'll see a lot of the wrong guys getting 'tough minutes' credit from him. But if you stand back, you can see that there is a clear picture there, and it makes sense. Give it a few weeks and, for whatever reason, his numbers have drifted to closely match the opinions of the hockey fans who pay attention to context in a hockey game.

What I hope to do in this post is remove a bit of the magic from this stuff, and check Desjardins' results against other sensible measures. It's not easy to describe 'tough minutes', but I think generally we would all agree that we're looking at position and opposition. If you're playing a lot against Joe Thornton, clearly that makes it harder to get good results, especially 5v5 +/- results. On the other hand, if you're on the ice for way more offensive zone draws than ones in your own zone, and your coach is throwing you over the boards against tired players a lot, and sending you out as the puck is heading through the neutral zone towards the other team's net ... well you'll probably put up good numbers at 5v5, you would be hard pressed to do otherwise. This even if you've logged your fair share of Thornton-minutes.

So, for starters, a quick look at Detroit. Babcock runs a rational bench, and a fairly consistent game plan, plus most of us are familiar with that team without being emotionally attached. So it's a good start.

If you click on the image it should enlarge. The first column is a tough minutes measure derived from the shots-against that happened while a player was on the ice at 5v5, and the 5v5 +/- of the players that were on the ice against them when these shots happened. No ice time used at all. It's just a matter of summing up the GF/(GF + GA) of every player who was on the ice for the other team when they got a shot on net against you. Simple as that. We'll call that the "shots metric". And the correlation with Desjardins (.92) is extremely strong. Really it's just Maltby and Draper moving up a bunch, everyone else is static. Possibly Babcock is relying on them more in defensive situations than Desjardins' numbers allow for.

The third column is the simple sum of two things:

1. The percentage of the faceoffs that happened in the defensive zone when a player was on the ice. Just that.


2. The average time-on-ice per game (EV + PP) of the opposition that a player faced. I excluded defencemen from that, otherwise I presume that it would be a dog's breakfast. No other stats used.

Equal weighting to both (1.) and (2.). Obviously (1.) is more in line with the shots metric, and (2.) is more in line with Desjardins. Give them equal weight and they end up with an overwhelming correlation to both (.93 and .91 respectively), even though all of these measures are built from different stats, the end results are linked together with a strong thread of reason. It's just hockey after all.

There are hundreds of ways to skin this cat, and frankly none of those listed above are great IMO. But they all start in different places, that's the point really. These seemed relatively simple, easy and independent, that's why I used them. The results are compelling and obvious think, so I didn't bother farting about, refining things, or looking other directions.

A mathy aside: I've multiplied everything by constants to get numbers in the same range that are easy on the eyes, this doesn't have a significant effect on the correlations. I also tacked an exponent onto Desjardin's numbers to spread them out a bit, because for whatever reasons they always want to bunch together more than the others (x110 and ^1,2 iirc). Again, this doesn't affect the correlation, just the presentation. It brings the numbers into the same range for the eyes to make comparison.

I'll get back to this later, for now I hope this is at least a point of discussion, and gives some confidence to the people who have been using Desjardins' QUALCOMP numbers. Of course comparing team to team with these is pointless, they are only relative to teammates. And some coaches run the bench more on position and tiredness of opposition than by the player matchup. Still, for 27 teams in the league, Desjardins' QUALCOMP arrows seem to be about the right size and are almost always pointed in the right direction, at least compared to these two other measures. EDM, CAR and especially PHI are the clear exceptions.


Next up, Columbus. Hitchcock is always an interesting guy.


Blogger Lowetide said...

Vic, one thing I'd like to have is previous seasons, say 2000-01 through today. It would be an exceptional tool in seeing how a coach handles his bench season over season and things he changes as he becomes more established as an NHL coach.

A lot of what MacT is doing this season could safely be described as "juggling like the devil" but much of it is dictated by the roster.

However, his Power V. Power matchups this season are curious imo and I'd like to see how often he's done it over the decade.

2/06/2008 6:44 pm  
Blogger Vic Ferrari said...

I'd like to go back a lot further, Lowetide.

One thing I hope to establish with this post, and probably a couple more, is the credibility of the QUALCOMP numbers that BehindTheNet makes available to everybody. Not that anything is perfect, but that there is a terrific amount of truth in them for most teams.

Have I started to achieve that with this post LT? Too numberish?

Remember that "results against the Big 8" (or whatever number it was) thing that me and mudcrutch kicked around for a while last year? Basically it was looking at how many of a players EV+'s and EV-'s came against the elite players in the West. On the surface the results looked really reasonable.

We could take that a little further, add more players into the "A" group, and make a "B" group as well. Then I could write a script to generate those results against that crowd, create a brand spanking new 'tough minutes metric' and we could compare against QUALCOMP to see if there was a strong correlation. To establish it's credibility.

From there I could go back to 99/00 with ease. And if we can get the score sheets there is no reason we couldn't go back to whenever they first recorded who was on the ice for every goal (1970?). The HAG folks or someone you know from the hockey history group might have that information somewhere.

I'm not much into the "best player ever" types of conversations, but i would like to see the raw 5v5 EV+ and EV- numbers for some of these famous names (Orr, Perreault, Sittler, Park, Gainey, LaFleur, of course the Isles and Oilers dynasty teams as well).

Like who was Huddy playing with and against when he managed to go +60, or something similar, in his rookie season. Common sense says vs depth and with Gretzky, but I'd like to know.

2/07/2008 1:06 am  
Blogger Lowetide said...

Actually I'm pretty excited about this exercise you're doing here. We've learned that Desjardins is best used when comparing a player to his teammates but after that it's just lying there and we're taking it as a concrete line in the sand. Well, Darrell Evans was one of the finest batters in MLB history but he hit .240 a LOT and was often traded for a load of hay because MLB teams didn't know his value.

So this is going to tell us who the Darrell Evans are, and also expose the Jesus Alou's.

There are all kinds of things I'd like to know:

1. How long before most 18-year olds who do play in the NHL become useful?

2. Is there an actual reason why all of the players from my childhood retired between 1979 and 1981 or did NHL teams just decide to stop paying the veterans?

3. Is it possible that a "career value" defender like Ray Bourque should be considered more valuable than a "peak value" defender like Bobby Orr just on the strength of so many consistent seasons?

The fly in the ointment on all of this has always been (imo) getting the "grades" for each season. That Reif book tried years ago but the math is flawed (how often do really smart people zig when they should zag? I've come to believe it's more than 50% which would be the logical answer).

2/07/2008 7:35 am  
Blogger Showerhead said...

One thing I hope to establish with this post, and probably a couple more, is the credibility of the QUALCOMP numbers that BehindTheNet makes available to everybody. Not that anything is perfect, but that there is a terrific amount of truth in them for most teams.

Have I started to achieve that with this post LT? Too numberish?

To me, you accomplished exactly what you set out to do. The happy part of it of course being that you've done this work to answer a question that had recently come into my mind. Basically, when behindthenet's numbers become pervasive to the point where more than a few HFB folk reference them then I begin to wonder what their true value is a lot more curiously than when people with common sense threw them around ;)

2/07/2008 11:09 pm  

Post a Comment

<< Home