Advanced Stats Overview

 

 

 

Graphs and numbers, tables and weird acronyms: advanced stats are assumed to be the province of the nerd and bane to the traditional hockey observer.

I consider the dichotomy a false one. I don’t begrude the typical skeptism I’ve encountered in the past when I’ve used and discussed "new metrics" here and elsewhere. Hockey is a fast, chaotic game and the efforts to properly capture it quantitatively have been halting and of dubious value. However, in this post I will attempt to detail and clarify some of the newer tools that have come to light over the past few years, many of which have proven useful to me in my amateur analysis and, in some ways, changed the way I’ve viewed the game. There’s room for both traditional observation and new stats in hockey analysis.

This is my attempt to demystify some of this stuff for those who are skeptical or unfamiliar. For each stat, I discuss what it is, why it’s relevant, where it can be found on the ‘net as well as what "context" in which to couch it. On the latter point, I think some of the reason that new stats are dismissed out of hand is because of their unfamilirity – people don’t have an established framework through which to view them, so they can’t really put their finger on what a given measurement means in absolute terms. For example, your average hockey fan understands that 20 goals in a season is pretty good, 30 is very good, 40 is exceptional and 50 is elite. They also know similar things about points totals, plus/minus, etc., but not about stuff like corsi.

So, if you’ve ever had any questions about what the heck I’m talking about sometimes, or if you’re just generally curious about this stuff but haven’t had the time or opportunity to investigate, please read on.

All shots at the net (shots+blocked shots+missed shots) for and against while a player is on the ice (usually at even strength).

Named after Buffalo coach Jim Corsi, who originally developed the stat as a method to track goaltender performance.

Corsi or "possession" is the prism through which a lot of modern analysis is currently viewed. A proxy for zone time, corsi measures where the puck is when a player is on the ice at ES: a negative value (or ratio below 50%) indicates the puck is spending more time in the defensive end. A positive value means more offensive zone time.

All other things being equal, good NHL players drive possession and have a positive corsi value. The higher the corsi rate; the more time a player is spending in the offensive zone; the more dominant he is. This is what we mean when we say a guy "drives the play north" – he’s effective at moving the play from the defensive end to the offensive end.

Coaches frequently make personnel decisions based on possession and players ability to drive it. Match-ups and placement in the line-up often march in lock-step with how much time a player is spending in the offensive or defensive zone. If he struggles, he’ll be moved down the line-up to see easier match-ups and vice versa.

Corsi correlates strongly with scoring chance differential at even strength over a large enough sample. For the 72 games I counted last year, the corsi to scoring chance correlation was +0.64 for the Flames. Others have found correlations closer to 0.9. 

A huge portion of NHL players linger around a 50% corsi ratio (or +-0 differential), because of parity and the ability of coaches to try to limit the damage by managing their roster. In absolute terms, the range for corsi/60 minutes of ice is about +/- 20 with the very best players at the top end of the range (Datsyuk, Ovechkin, etc.) and the worst at the bottom (goons, rookies, etc.). As a rule of thumb, any /60 corsi rate around +10 or above is in "excellent" territory, while anyone around -10 or lower is getting into "terrible" territory. Over the course of a season, anyone in around 55% ratio is at the top end of things, while a 45% ratio is pretty poor.

Keep in mind, it’s always a good idea to rate a player in terms of his team, primarily because team effects can go a long way to determing a guys results (in all things, not just advanced stats). Therefore, a player with a +1.0/60 corsi rate might not be impressive in absolute terms, but is probably noteworthy if the rest of his club is underwater.

We’ll get to the other factors that moderate corsi later in this article.

www.behindthenet.ca is the clearing house of advanced stats. Click the season in question and then "basic +/- scoring statistics (5-on-5)". From there, you can choose teams, player position and stat type from various drop down menus. For corsi, click "corsi" from the "show stats report" menu and then find the "corsi on" column. This is the players corsi/60 differential while he’s on the ice.

A ratio that measures the amount of offensive zone draws to defensive zone draws a player is on the ice for. Usually only used at even strength.

Zone starts is one of those factors that moderates possession. As best as we can tell, an offensive zone draw is worth about +0.8 in terms of corsi, while a defensive zone draw is worth the opposite. This isn’t a big deal if the number of offensive and defensive zone face-offs is roughly equal for a player. On the other hand, some guys get buried by their coaches (shut-down defensemen, checking centers) while others are privileged (scoring wingers, third pairing defenders). In some cases, coaches will give a player a higher ratio because he’s considered more dangerous and therefore has a better chance of taking advantage of starting in the offensive zone. On the other hand, a coach can shelter a player by giving him a higher ratio of draws that start away from his own net.

Being a percentage point below or above 50% probably isn’t a big deal, but once a guy drifts to the +/- 4 or 5 % in a zone start ratio, the influence on his corsi rates can be pretty significant. There’s a few shut-down defenders and centers that get into 40% or below territory while some coaches will privilege guys to the tune of 60% o-zone draws.

Again, visit behind the net, click 5-on-5 stats and select zone start from the pull down menu. The relevent column is the first one marked "OPCT".

A method to quantify the level of competition a skater faces. We all know that not all match-ups and assignments are created equal in the NHL. Behind the net’s Gabriel Desjardins has come up with a couple of methods to aggregate each skaters quality of competition. The first relies on summing, averaging and weighting the even-strength scoring rates of the skaters a player sees over the course of the season. The second is the same, except with corsi rates rather than scoring rates. The first gives you an idea of Qual Comp in terms of scoring, the second in terms of possession. Since the two often travel hand-in-hand they are usually in agreement, although there is some variation. 

Qual Comp gives us some idea of the difficulty or ease of a players ice time and allows us to add more context to his other results, such as scoring and, again, corsi. For example, a player who faces the toughest competition on his team and still has a respectable corsi rate is probably an excellent-to-elite player. If he doesn’t start 60% of his shifts in the offensive that is (and maybe even then).

Qual comp isn’t a stat that can be readily compared across teams, so it’s best utilized as a rank/order tool inside individual clubs. As such, the raw numbers are often largely irrelevant compared to where a guy ranks relative to his teammates.

Go to behind the net, click 5-on-5 and you’ll find both the standard quality of competition and corsi quality of competition. The first is the "QUALTEAM" column while the second it labelled "CORSI QoC"

the sum of on-ice save percentage and shooting percentage at even strength.

The stat was named after the alias of a frequent blog commenter who suggested it’s existence/relevance.

One of the most counter-intuitive aspects of hockey is the degree to which "variance" or "luck" effects results. Most savvy or experienced hockey observers can watch a game or two and identify times where a puck was lucky to bounce in the net or off a post. The problem is, the human brain is actually pretty lousy at determining such things over a few events or games, so the influence of fortune tends to be forgotten or minimized.

What’s been learned recently (thanks to the high parity/competitiveness of the league as well quality of coaching and goaltending in the NHL) is that each player (and team’s) PDO tends to regress towards the league-wide mean of 100 over the long-term. Meaning, if a club is experiencing success because they are "riding percentages" (ie, their PDO is well above 100) it means it is probably due more to variance (luck) than true ability. It’s the same for players, who are far more capable of driving possession (corsi) than they are percentages (PDO) at the individual level. Again, a guy who has a nice stats line thanks to a PDO of 105 is more likely than not in for a correction over the long-term.

In the short term, PDO’s swing widly, but once the sample gets large enough, the typical range for PDO for both skaters and teams is about 95-105. Guys in the upper range are probably going to fall back down to earth. Guys at the lower range are probably going to rebound.

In the behind the net "corsi" drop-down option you’ll see a column marked "PDO". Beside it you’ll notice the seperate on-ice SV% and SH% for each player as well.

A bit of a project started a couple of years ago by an Oilers blogger, scoring chances are shots at the net from a defined (ie; scoring) area on the ice for and against.

Clearly, not all shots at a goal are created equal. Shots from the blueline, along the boards or from the corners are of relatively poor quality because of the angle to the net, the chances for interception of blocking and/or increased time for a goalie to react. 

The scoring chance project allows us to not only count team-wide totals for and against for each contest, but determine differentials for each skater while he’s on the ice. We can then correlate scoring chance rates with other metrics to see how they relate.

We mostly look at ES stats since that’s where a majority of the game is played, but chances at all game states are recorded.

A rule of thumb that has arisen since the project began is: each game has about 35 chances total between both teams, give or take. Some abberrations and blow-outs occur, but that’s about the unofficial mean. A very good night for a skater is usually in the +10 area, while a terrible night is at the other end (similar to corsi).

As mentioned corsi and scoring chance differential tend to agree with each other over the long-term, although they can be out of sync during individual contests. Like corsi, however, scoring chances are obviously moderated by the same contextual issues: zone start, quality of competition, etc.

As of right now, there’s no single site to review scoring chance data. Various bloggers tracked SC for teams last year. Their sites are listed here (left hand sideboard). I counted the Flames last season and the results are scattered around my personal site. This season, you’ll find the chance data for Calgary right here at Flamesnation.

There’s no question that hockey can be enjoyed for it’s speed, excitment and majesty without these sorts of tools and metrics. However, when it comes to analysis, things like this lend context to the regular counting numbers (or surface stats) one gets on the back of hockey cards. On their own, they are of limited utility. But, considered in concert with each other, as well as counting numbers and observation, these measurements tell us who a guy is playing against; where he frequently starts his shifts; where the puck spends it’s time when he’s on the ice; what his scoring chance differential is and the degree to which variance is influencing his results, giving us a much richer picture of his true abilities.

In addition, considering a players performance in light of the various moderating influences can help evaluate guys one may not get to see very frequently. After all, even dedicated scouts can’t view every game by every player over the course of a season, let alone fans.

By way of example, Patrice Bergeron scored 19 goals and 52 points for the Boston Bruins last year. Respectable, though hardly exciting right?

Here are the circumstances of his results:

Corsi: +15.70/60 (best amongst regular skaters on the team)

Zone start: 48.1% (below 50% and one of the lowest on the club)

Qual comp: +0.149 (tops on the team), +0.214 corsi qualcomp (3rd amongst forwards)

PDO: 97.4

So…with these tools out our disposal, we know Bergeron was starting more often in his own end, playing against other teams top lines and still managed to have the best possession rate on the club. He also had one of the lowest PDO’s on the Bruins, meaning the bounces weren’t kind to him and he’s in line for a rebound. If you’re a fantasy hockey guy, this is a player you’d look for iin the mid-rounds – there’s a huge chance he’ll up his scoring this year.

What all this points to is that Bergeron is an excellent hockey player – much, much better than what is originally suggested by his 19 goals and 52 points. If you were wondering why Steve Yzerman saw fit to include Bergeron on the Olympic squad…now you know.

So hopefully this overview gives an idea of what these new stats are all about and why many of us find them useful these days. This primer wasn’t exhaustive and we’re still likely in the formative years of really understanding what determines success at a team or individual level, but the metrics discussed here are the building blocks of such an endeavor.