Wednesday, August 6, 2008

The End of the Career

Hockey players bounce around from team-to-team ad infinitum. Which leaves you with a few very low-quality, dull years in random cities from various superstars. I'm not talking Ray Bourque going to Colorado to win a cup; I'm talking Brett Hull in Phoenix. So imagine this all-star starting lineup:

G: Ed Belfour (Florida)
D: Adam Foote (Columbus)
D: Bobby Orr (Chicago)
LW: Joe Neiuwendyk (Florida)
C: Brett Hull (Phoenix)
RW: Gordie Howe (Hartford)

Saturday, August 2, 2008


We're back here at Hockey Outsiders, with some updated information for our forthcoming defensive statistic:

The value is going to be "Shots Allowed." This isn't as easy as taking the team shots allowed and dividing by six of course, but requires calculating a good deal of information about each player based on +/-, power play performance, shots blocked, and the save percentage of the goalie behind him.

Right now, we're still crunching a pile of numbers, but the result should be in soon. I'm just hoping this isn't too biased against offensive defensemen.

Friday, July 18, 2008


The NHL schedule is out, and I want to point out a few things that bug me most of the time:

1.) Starting the season overseas. Honestly, I don't think this is a terrible idea in and of itself: the NHL has to keep spreading the world in Europe, and Stockholm and Prague are great hockey cities. And Stockholm - think about the crowds that would turn out to see the goalie that led Sweden to a gold medal in 2006!

...Too bad Lundqvist is playing in Prague. Oh well - you know, Jagr plays for the Rangers, they'll turn out to see him...ooh, right. Avangard Omsk. Well, I'm sure the Czechs will turn out to see journeymen like Petr Prucha and Michael Roszival and Vaclav Prospal as the Rangers play two games against the Lightning - winners of the draft lottery. Not sure why Elias and Hejduk weren't invited - Rangers-Devils would certainly be a more interesting game.

Sweden gets to see the Senators and the Penguins. So certainly, I can understand sending out Sid Crosby to rally up the faithful, and the Sens have Danny Alfredsson, so no complaints there. Alfie is the only Swede on either roster, however.

2.) Opening Night
The NFL starts the season with the Super Bowl champion getting a home game against a rival, all alone on national TV Thursday night. Giants and the Redskins. We're given the cup champions, versus an out-of-conference team that didn't make the playoffs and whose most notable offseason move is adding someone named Finger. Yes, the Red Wings get to hoist the banner in front of the Toronto Maple Leafs, not, say, the Blackhawks, the Stars, or the Avs. The NHL then treats us to fewer than THREE 10:00pm EST games on a Thursday night, featuring neither conference runner up, and forcing Bruins fans to start their season in a semi-meaningless game in Colorado.

Needless to say, you'll have to have Versus to get the Red Wings - Leafs game - NBC wouldn't touch it.

3.) The Cup Rematch

You'll have to wait until November, which isn't too bad, but then the game is somewhat randomly a Tuesday night affair on Veterans day, of all times.

Philly and Pittsburgh also don't get to settle that score in the first month of the season, despite being in the same division, and Dallas and Detroit don't meet up until November. These rivalries are fresh NOW, Bettman.

Thursday, July 17, 2008

A defensive stat

I've been struggling to come up with a stat that really measures defensive ability - there's a few helpful indicators (hits, shots blocked) but they're tricky to compare, and also to correlate. I'm trying to work something out with plus/minus right now, and I'm going to see if the first set of results passes the smell test.

The idea is this: Start with plus/minus. Subtract even-strength points, and you get something that's fairly close to how many goals were allowed when a given player was on the ice. The first problem is universal when you want a small number: my score in this stat is a perfect 0.

The next step is to normalize it - divide by game played, and divide by shifts/game (we don't want 7th defensemen ramping up their scores). This gets us the player's goals allowed/shift value, which tends to be a difficult-to-understand fraction, so I think it's worth scaling it up for a full season.

The next step is to normalize it for goalies, and maybe the total team offense.

Wednesday, July 16, 2008

The Year 2014

A far future time - a President Obama might start thinking about his legacy. The Nintendo Wii will be obsolete, replaced by a machine with sensors for your hands, arms, knees, and hips for a truly immersive environment in Dance Dance Revolution IX. And according to Capcom, a lone robot named Rock will have defeated six waves of evil Robot Masters under the control of the nefarious Dr. Wily.

The NHL will have taken a break for the Olympics in Sochi, Russia, four years after the games in Vancouver.

And nineteen players will still be bound by contracts they've already signed.

Some of these players will very likely still be in their primes: a 27-year-old Alex Ovechkin, Thomas Vanek at 29, Jason Spezza at 30. But then there's a 36-year-old Dan Boyle making nearly seven million dollars. Or Rick DiPietro at a reasonable $4M at age 32, but with a full seven years left on his contract.

Past performance is not an indicator of future success, so whatever these teams are hoping for their trigenarians in 2014, the odds are against them.

Using a sample of the top 100 scorers from the past 15 years or so, I charted aging players performance. And while it's easy to remember the later years of the Great One as being proof older players can perform, it's not true. Age 26 and 27 was distinctly the peak for this group of players - performance drops at a rate of about 5% per season after that. There's a bit of a cliff from 29 to 30, and collapse of 10% every year from 34 on.

Some players beat the odds, to be sure, but don't get fooled by early all star years that turn into a long, slow, wind of 50 point seasons.

Tuesday, July 15, 2008

More on the Mission

I realized I don't think I've spent any time explaining what I'm doing here yet:

The goal of the Hockey Outsiders is to make sense of NHL stats in a way that lets us know something more than "this guy scores a lot." The reason is that scoring goals isn't always that illuminating: is it a great individual talent? A product of excellent linemates? Weak schedule? Empty netters? At the same time, goaltending runs into similar problems: great goalie, or great defense?

The reason why this particularly matters is that the way NHL franchises build teams is changing - the past several summers, we've seen the price of top players explode through the roof, while the payoffs for their new teams are hit and miss. What we're looking for, eventually, is if there's some way to predict which players are worth spending money on every summer. And not only invididual names, but by position: should you load up on a few top forwards or roll four scoring lines? Do you want defensive defensemen or guys that chip in on the power play? Do you want to goalies to play 40 games each, or one guy who can shoulder a 70-75 game load before the playoffs even begin?

Additionally, NHL free agency is being drawn more and more towards long term contracts. REALLY long term contracts - if Rick DiPietro has a kid, they'll be playing together by the time DP is free of the Islanders. So a question that we'll answer in the coming days is - is that worth it? How long can you count on production from a player - both in terms of age and consistency?

And there's more - what are the best draft strategies to take? Are trades more valuable than free agents? Europeans vs. North Americans? And on and on.

We'll see what we can do. And of course, we'll study the biggest question: do the officials cheat in favor of Canadian teams?

Rule 3: Get a power play guy

Power plays count for a lot, especially in the playoffs. Power play percentage isn't nearly as important as playing 5 on 5 in the regular season, but it shoots up in the postseason, while 5 on 5 starts to even out.

At the same time, power play percentage, and the total number of power plays, is actually more influential on winning games and winning playoff games than goals themselves are. Goals scored is actually a relatively weak predictor, while special teams seems to count extra.

But here's the elite stat for special teams: 20% power play. 1 in 5. Only four teams reached the mark in 2007-8: Detroit, Pittsburgh, Montreal, and Philadelphia. Three of the league's Final Four - Western Conference runner-up Dallas put up an 18.1%.

However, 20% really seems to be the magic number, because things tail off below that. Tampa Bay records a 19.3% and Florida a 19.2%.

Saturday, July 12, 2008

Rule 2: No, really, defense is worth it

I took a look at some of the numbers on player contracts, thanks to the information provided by And this is the most shocking part:

There is a negative correlation between spending money on forwards and both winning games and winning playoff games.

It's not too strong, so it's probably just noise, but in any case, there is no positive link between shelling out the big bucks for your Danny Brieres and your Jaromir Jagrs (I'm looking at YOU, Avangard Omsk).

Defense still does win championships: there's a correlation of .32 between spending on defensemen and winning playoff games, and while goaltending is mostly random in the playoffs, I think that's mostly due to the underpaid Marc-Andre Fleury and Chris Osgood showng up in the finals. With a larger sample size, there is a distinct (.39) link between paying goalies and winning hockey games in the regular season.

However, that might be misleading - I have some surprising numbers to report soon about goaltending.

Friday, July 11, 2008

Team Stat Crunching

Okay, I've built a database for the 2007-8 season. Right now, I think my primary focus is going to be on whole teams and goalies, rather than breaking everything down for each skater. I'm working on a big database for them, but it's going to be a much longer, slower, process, and I want to get some information out there.

So, thirty teams. That's the easy part. Using the NHL's Stat Machine (not to be confused with the NHL Highlight Machine), I broke out the following statistics:

Wins, Losses, OTLs, Points, Goals For, Goals Against, 5 on 5 ratio, Power Play Scoring %, Penalty Killing %, Shots for, Shots against, Minor Penalties, Major Penalties, Total Penalties in Minutes, # of Power Plays, # of times Short Handed, Power Play Goals, Short Handed Goals, Short Handed Goals Against, Goal Differential, Total Cap Hit, Cap Hit for Forwards/Defensemen/Goalies, Face Off Percentage, and Playoff Wins.

So, there's two goals here: What lines up with Points (regular season success) and what lines up with Playoff Performance (playoff wins).

At this point, we've got a hot, steaming mess of a spreadsheet to look at. Let's see what comes out of the pile:


All right, it may be the New NHL, but defense still ruled the day in 2008. In almost every category, a team's defensive performance was more important than its offense - that is, it correlated more strongly to both points and playoff wins. Goals For led only to a 0.37 correlation coefficient for points and a 0.40 correlation for playoff wins, while Goals Against gave -0.78 and -0.54 respectively.

Math aside: What this means - Correlation is a number that shows how closely linked two different sets of numbers are - how often they either go up together or go down together, or even go in opposite directions at the same time. A perfect correlation is 1.0: Your age in years and your age in months have a correlation of 1.0. It means there are two numbers that are giving the same information, basically. At the same time, a correlation of -1.0 is also very meaningful. It means that as you accumulate more of one thing, you lose something else at the exact same rate. The correlation between the number of Pringles in the can and the number you've eaten is -1.0. Anything in between means that the numbers are more and more randomly linked, until you hit 0.0, which means that there is absolutely no connection between the two sets you're looking at.

So with that established, what does it mean? Well, -.78 and -.54 are both strong correlations for goals against - negative, here, because it means that teams that give up fewer goals win more games. It doesn't hurt that the three best defensive teams this year were Detroit, Anaheim, and San Jose, in that order. You have the Cup team/President's Trophy winner leading the way, followed by teams #2 and #4 in points. Tampa Bay is at the bottom, so that rush of offense they've added this year may not help them so much.

Detroit had a good offense too, though: #3 in goals scored. But they didn't have much company up there. Ottawa, Montreal, Buffalo, and Carolina round out the top five, with five playoff wins between them. Anaheim only beat out Columbus and the Islanders in scoring, yet ran up over 100 points, and San Jose got to 108 points with only pedestrian scoring.

As we expand the sample, we'll see if this was a one-year quirk, but there is strong evidence in other categories that defense wins across the board.

Major Stats

One problem that I thought I would have with the NHL is its relative paucity of statistics. Baseball is particularly rich, and football is just so complicated, there is an incredible amount to be measured. But in the NHL, we usually get little more than goals, assists, points, PIM, and +/- for skaters, then GAA and save percentage for goalies. The truth is, there's a lot more out there, and the NHL actually does an impressive job keeping those statistics. They just don't keep them in a very user friendly manner.

They call them "RTSS Stats," and they include things like hits and blocked shots and other little bits of minutia that, you know, seem pretty important. Also, the NHL keeps a remarkable amount of depth for splits - power plays, home and away, etc. There's a lot to chew on out there, but the first step is figuring out a way to make sense of it.

There are several major links I want to establish. The first of these is points. As in, two for wins, one for shootout/OT losses. What stats are the most strongly correlated with points - this seems to be the most obvious interest, and with an 82 game season, the correlation should be significant - not just lucky.

Secondly, there's the issue of the playoffs. The NHL is notable for shocking upsets from #8 seeds while President's Trophy winners flame out early. I want to see if there's something out there that grabs playoff wins better or worse.

Finally, there's consistency. Generally speaking, what stats really persist from year to year, and which ones are just a result of a few lucky bounces, some good linemates, or a hot streak. I'm particularly interested in goalies in this regard - are they a fabrication of defensive systems, or are they truly standouts on their own.

I've started scratching the surface here, and let's see what we've got.


Baseball has its Sabremetrics and the NFL has the brilliant work of, but at this point, I think there's a real gap for the NHL. What makes good teams so good? What really counts, and in an era of free agency, where should a team spend its money? What makes for a good defensive team? Is it better to load up on a top line or roll four scoring lines?

I don't know if there are clear answers to these questions, but I think that it merits pursuing. So on this blog, I hope to build a real statistical database to set up what makes the NHL tick, and what teasm should do with their salary cap dollars.

Welcome to the Hockey Outsiders.