For those who have been following their growth, it's been evident for a while that hockey's advanced stats are here to stay. They've provided us with a new way to watch and analyze the game. They've quantified ideas that previously couldn't be quantified. And they've proven to be a better predictor of future success than anything else that's available.
If anyone still needed to be convinced that they're here to stay, this summer should do the trick. In June, the Devils hired former stats blogger Sunny Mehta as their new director of analytics and the Bruins promoted Ryan Nadeau to director of hockey operations/analytics. In mid-July, Eric Tulsky -- who has consistently pushed hockey analytics forward through his blogging for Broad Street Hockey and SB Nation -- announced he had been hired part-time by a still-unnamed NHL team.
A week later, the Maple Leafs hired Kyle Dubas as their new assistant general manager. The 28-year-old has been called the Theo Epstein of hockey, and he spent the previous three seasons using an analytical approach to take the Ontario Hockey League's Sault Ste. Marie Greyhounds from cellar dwellers to division champions.
The latest development in what some are calling the Summer of Stats came Tuesday when Bob McKenzie broke the news that Tyler Dellow -- who, like Tulsky, has made a name for himself as one of the best and most forward-thinking stats bloggers around -- had been hired by an NHL club, later reported to be the Oilers.
Clearly, the NHL is embracing analytics more and more. Teams are going straight to the source and hiring those who are doing the most with these stats. This doesn't mean that teams are going to stop scouting or stop watching games (as an aside, the idea that stats people "don't watch the game" is the dumbest straw man argument going). It just means they're realizing that advanced metrics can supplement what they're already doing.
So, given all that, now seems like as good a time as any for an advanced stats primer. It's never too late to jump on the bandwagon and learn new things about a game you love.
It starts with Corsi
Corsi is the advanced stat you'll hear about the most. The first thing you should know about Corsi is that it's actually not very advanced at all. The second thing you should know is that it's not an acronym; it's named after Jim Corsi, the goalie coach who first started tracking it.
Corsi is a plus/minus stat that measures all shot attempts (shots on goal, shots wide, shots blocked and goals) when a given team or player is on the ice. It is used as a proxy for puck possession, which is the best way to figure out which teams or players are controlling play and creating scoring chances. You can't score a goal without shooting the puck, and you can't shoot the puck without first possessing it.
Corsi can be expressed as a differential per 60 minutes or as a percentage. It is most often used as a 5-on-5 stat because that's how the majority of games are played and that's when teams are on even footing.
For example, the Bruins had a Corsi differential of plus-8.3 last season, meaning that for every 60 minutes they played at 5-on-5, they attempted 8.3 shots more than they allowed. If you prefer the percentage (personally, I do), they had a Corsi-for percentage of 53.8 percent. That ranked fourth in the NHL, marking the third straight year in which the Bruins have been a top-five possession team. The Stanley Cup champion Kings led the NHL for the second straight year at 56.8 percent, while the Maple Leafs finished last for the second straight year at 42.8 percent (watch for that to change as Dubas gets involved).
At an individual level, Patrice Bergeron led the NHL with a Corsi-for percentage of 61.2 percent, marking the third straight year in which he's finished in the top 10. Brad Marchand and Reilly Smith also ranked in the top 10 this past season at 59.8 percent and 58.5 percent, respectively. The only Bruins to play at least half the team's games and finish with a Corsi percentage under 50 percent were Daniel Paille (48.0), Shawn Thornton (47.7) and Gregory Campbell (45.6).
All of these stats can be found on ExtraSkater.com, which has become the best place to get advanced hockey stats.
You might also see Corsi used for "5-on-5 close" situations, meaning when it is a tied or one-goal game. This eliminates score effects -- a team that is up by multiple goals is more likely to play conservative and give up more shots, while a team that is trailing by multiple goals is more likely to play desperate and shoot the puck more.
You could see Fenwick used as well. Fenwick is the same as Corsi, but without blocked shots. Some prefer Fenwick for larger sample sizes because it correlates more closely to scoring chances, but we'll just stick with Corsi here since it's more widely used.
CorsiRel is just as important
Corsi by itself is a great way to figure out which teams are the best (or worst) possession teams. Outside of using close situations instead of all 5-on-5 situations, there isn't really a whole lot more you want to see.
But Corsi by itself only tells part of the story when it comes to individual players. As you might imagine, players on good possession teams tend to have a better Corsi than players on bad possession teams. This doesn't necessarily mean that every player on the good team is a better possession player than every player on the bad team. We need to adjust for their different circumstances.
Enter Relative Corsi, or CorsiRel. CorsiRel measures a player's Corsi when he is on the ice vs. the team's Corsi when he is not. A player with a positive CorsiRel helps his team's possession, while a player with a negative CorsiRel hurts it.
Bergeron, for example, had a CorsiRel percentage of plus-9.7 percent, meaning when Bergeron was on the ice, the Bruins' Corsi-for percentage was 9.7 points better than when he wasn't (so the Bruins had a 51.5 percent Corsi when he wasn't on the ice). That number was good for second in the NHL, which makes Bergeron look even better. It tells us that he didn't just post a great Corsi because he plays on a great possession team; he posted a great Corsi because he was a huge factor in driving possession. (The best CorsiRel player was Calgary defenseman Mark Giordano at plus-10.3 percent. His straight Corsi-for percentage of 53.1 percent ranked just 91st in the NHL, but CorsiRel accounts for the fact that Giordano played on a bad possession team.)
At the other end of the spectrum, Paille, Thornton and Campbell all ranked in the bottom 12 in the league in CorsiRel. In a reverse of Giordano's case, you might look at Campbell ranking 70th-worst in straight Corsi and think, "Well it's not good, but it's not terrible either." But when you account for the fact that he plays on a good possession team and then see that he ranks third-worst in CorsiRel at minus-10.3 percent, you might want the Bruins to explore some other options for fourth-line center.
Like Corsi, all these CorsiRel numbers can be found at ExtraSkater.com.
Zone starts matter, too
CorsiRel starts to give us a clearer picture of who drives possession and who holds his team back. But we can get an even better picture by looking at some other factors, one of which is zone starts.
Zone starts tell us whether a player is being used in more offensive situations or more defensive ones by calculating how many shifts a player starts in the offensive zone vs. how many he starts in the defensive zone. This is important because a player who starts mostly in the offensive zone will have an easier time generating shot attempts than one who starts mostly in the defensive zone. And obviously the one who starts mostly in the defensive zone is more likely to be on the ice for shot attempts against.
As a statistic, zone starts are expressed as offensive zone start percentage, or just zone start percentage for short. Sticking with the Bergeron example, he had a zone start percentage of 46.0 percent last season. That means that of all Bergeron's non-neutral zone starts, 46 percent came in the offensive zone and 54 percent came in the defensive zone. Bergeron was used in more defensive situations than offensive situations, which makes his Corsi and CorsiRel look even more impressive. Not only did he drive possession, but he did it without the benefit of being used in an offensive role.
Chris Kelly (44.4 percent) and Zdeno Chara (48.1 percent) join Bergeron as the three Bruins who had the toughest zone starts. Conversely, Torey Krug (66.2 percent), Shawn Thornton (65.4 percent) and Kevan Miller (61.0 percent) had the most favorable zone starts. This tells us that Krug and Miller, the Bruins' third defense pairing for much of the season, weren't really trusted enough to be used in a lot of defensive situations. It also tells us that Thornton's bad CorsiRel is even worse than we initially thought. Not only did he hurt possession, but he did it while being used in situations that should have helped his possession numbers.
The graphic below gives us a good visual of where individual Bruins started their shifts. Zone start percentage is shown on the X axis.
So what's that Y axis?
Great question, reader. That would be quality of competition (or QoC), another important factor to consider when comparing players. It measures exactly what it sounds like -- the quality of competition that a given player faces. Basically, is this player facing an opponent's best players, worst players or something in between?
ExtraSkater.com measures QoC as a percentage using opponents' time on ice, with the theory being that an opponent's best players are the ones who play the most. You might see this written out as TOI% QoC. BehindTheNet.ca (another great stats site) measures it as the average CorsiRel of the competition that player faces. This would be written out as CorsiRel QoC. So, Bergeron has a QoC of 29.5 percent using time on ice (as seen above), or plus-0.934 using CorsiRel.
These two calculations spit out similar results in terms of how players compare to each other, but personally I like the CorsiRel QoC more. I find it easier to put in perspective. A player with a positive CorsiRel QoC faces tougher than average competition, while one with a negative CorsiRel QoC faces easier than average competition. (For what it's worth, 28.5 percent is the dividing line using time on ice percentage.)
Similar to zone starts, QoC tells us who is being used in tough situations and who's being used in more favorable ones. It should be harder to put up a great Corsi against an opponent's best players, and easier against its worst. On the graph above, the higher up the Y axis you are, the tougher the competition you face.
Usage charts like this are the best way to visualize a player's impact on possession. The color of the circle is Corsi. Blue is positive, red is negative. The darker the blue, the better. The darker the red, the worse. As you can see, and as we mentioned before, the Bruins had just three negative possession players last season. (If you want to see lots of red, check out this Maple Leafs graph.) The size of the circle is ice time -- big circle equals big ice time, small circle equals small ice time.
Players in the top left quadrant are the ones with the toughest usage. They're facing a high quality of competition while starting more shifts in the defensive zone than the offensive zone. These players are often classified as playing "shutdown" roles. The bottom right quadrant, on the other hand, features players with the easiest deployment. They're facing low quality of competition while starting in the offensive zone more than the defensive zone. These players are often classified as being "sheltered."
So, being blue in the top left corner -- like Bergeron and Chara -- is the mark of a great player who really drives possession. Being red in the bottom right -- like Paille and Thornton -- is the mark of a player who isn't bringing much to the table in terms of possession. Being the darkest red spot on the chart -- like Campbell -- is also bad, regardless of usage.
HockeyAbstract.com adds the nice wrinkle of being able to color the bubbles with CorsiRel rather than just Corsi. When you do this, Miller (CorsiRel of minus-5.5 percent) joins the Merlot Line as a player who doesn't look too good. Marchand, Smith, Dougie Hamilton and Johnny Boychuk stand out as players who are in a really good spot.
If you know how to read these usage charts, you'll be in pretty good shape in most conversations about advanced stats. ExtraSkater.com allows you to map any combination of players by going to one player's page (here's Chara's) and using the "Compare" button. The chart below shows Chara compared to Duncan Keith and Shea Weber, this year's other two Norris Trophy finalists.
WOWY, this is cool
The last thing we'll touch on, for now, is so-called "with or without you" stats, or WOWYs. These tell you how certain players play together vs. how they play when they're separated. It gives you an idea of who's lifting their linemates' or partner's possession numbers, who's bringing them down, and how any two players mesh.
The best place to get these is Stats.HockeyAnalysis.com. Here is a look at Bergeron's WOWYs for last season. The numbers you want to pay attention to are the three CF% (Corsi-for percentage) numbers. The first is Bergeron and that player together, the second is Bergeron apart from that player, and the third is that player apart from Bergeron.
As you can see, everyone who played at least 35 minutes with Bergeron was better with him than without him. Many saw their Corsi jump by 10 points or more with him. This is just further confirmation that Bergeron is a possession monster who makes everyone around him better. While Marchand and Smith are both fine players in their own right, there's no denying that the biggest reason they ranked in the top 10 in Corsi this past season is because they played on a line with Bergeron.
Loui Eriksson is another player who lifted the possession numbers of pretty much everyone he played with, while Campbell and Miller dragged down pretty much everyone they played with.
When added to what we see with our eyes (seriously, we stats people really do watch games), everything mentioned above can help give us a really well-rounded picture of any player's impact on the game and his team.
But there's still so much more that can be done, and that's probably the most exciting part of hockey's advanced stats. The NHL is starting to test tracking systems like the ones used in the NBA. These could tell us, among other things, which players are covering the most space, which forwards are creating separation before shooting, which forwards are getting to the front of the net the most, which defensemen are most effective at taking space away from attacking forwards, and which defensemen are best at starting breakouts.
Some bloggers have started to do some of this by tracking zone entries and exits (Tulsky was a pioneer in this area, and Corey Sznajder is doing some of the best work now), but it's a very time-consuming process. Zone entries can tell us which forwards are the best at carrying the puck into the offensive zone, which is important because carry-ins are more likely to lead to scoring chances than dump-ins. At the other end of the ice, zone entry stats can tell us which defensemen are the best at breaking up opponents' entries, and zone exit stats can tell us who's driving breakouts.
ExtraSkater.com has recently started tracking things like shots on net percentage (which players are best at getting shots through), pass/shot ratio (how often a player passes vs. shoots), and setup passes (how many passes from a player lead directly to a shot attempt). There is evidence that those last two are very flawed stats, though. Still, they're admirable ideas that could become very useful with a little more work.
Hockey-Reference.com (with point shares) and HockeyProspectus.com (with goals versus threshold, or GVT) have both attempted to create an all-encompassing stat like baseball's WAR. Both are flawed for different reasons, but again, the effort is commendable.
We'll continue to learn more about all these stats and how to use them. Eventually more and more of them will be used by NHL teams, and eventually more and more of them will make their way toward the mainstream. As that happens, hockey will continue down the paths that baseball and basketball have already paved, with analytics continually teaching us new things and challenging us to look at the game in different ways.
UPDATE: As of September 3 -- less than a month after this was written -- some of this article is already out of date. Most significantly, ExtraSkater.com no longer exists. Its founder, Darryl Metcalf, was hired by the Maple Leafs and the site was shut down. Most of the stats from Extra Skater can still be found elsewhere, but they're not as easy to read and visualize on other sites. Also, we now have a site that tracks shot location: war-on-ice.com.