The Novice's Guide to Advanced Stats Part IV: Hockey

Posted by Taylor Nigrelli on February 27, 2015 · 8 mins read

If you’re a fan of hockey, you may have heard the NHL recently added an “enhanced” stats section to its website. The new page features some of the more basic advanced stats that have been developed over the past decade.

This is not the culmination of a project, but rather phase one of a four-part plan. Future plans include creating visuals for stats, creating an archive on NHL stats dating back more than a century and creating a puck-tracking system similar to the camera-tracking systems the NBA and MLB have utilized.

These proposed innovations would revolutionize the way fans followed the sport. So many gray areas would be cleared up. Hockey could be understood at the level we’re able to study at basketball and baseball at this point.

But, what of casual fans or those who haven’t caught up with all the analytics hoopla just yet? Won’t they be left behind? Fear not, we’ve got your back.

Advanced hockey stats are in their infancy compared to other sports. While this means analytics gurus have a lot of work ahead of them, it makes it easier for the uninitiated to learn about the subject. Additionally, hockey analytics tend to be far simpler than other sports’. It’s mostly adding, subtracting, dividing and using reason and logic.

Here’s a few basic concepts to keep in mind as the NHL moves into a new statistical era:

Possession Numbers – This is important and simple: teams that attempt more shots than their opponents tend to do well. Shots on goal used to be the go-to stat to support this idea. However, considering all shot attempts allows for a larger sample size and acts as a good proxy for possession. Thus, Corsi and Fenwick were born in the mid-aughts. Corsi measures all shot attempts while Fenwick measures all shot attempts. Appropriately, the NHL renamed the two stats shot attempts and unblocked shot attempts when launching its new enhanced stats page (it’s generally good for stat’s names to describe what the stat measures – it lends credibility). Regardless of what you call these stats, they can be expressed in a variety of ways. They can be expressed as a raw number (ex: the Canucks out-attempted the Sabres 80-50 tonight), a percentage (ex: the Sabres attempt a league low 37 percent of shots), for an individual player (ex: the Bruins take more than 60 percent of shot attempts when Patrice Bergeron is one the ice) or as a relative number (Corsi Rel measures a player’s value by subtracting the team’s Corsi when he’s off the ice from the team’s corsi when he’s on the ice).

Additionally, it’s important to keep the game situation in mind when considering possession numbers.

Strength and Score Effects – Power plays and blowouts can skew possession numbers. Obviously teams on power play will dominate possession while teams on penalty kills will struggle. But it’s less obvious and less widely-known that teams with two-goal deficits tend to dominate possession. This happens for a few reasons; losing teams get desperate and start throwing everything at the net while winning teams tend to sit on their lead. Therefore, it’s important to use close-game, even-strength possession numbers. Goaltending and special teams can be evaluated separately.

Luck – Because of hockey’s low-scoring, high variance nature, luck is prominently involved over the course of games, months and entire seasons. Teams with good possession numbers tend to win more. If a team has good underlying numbers and isn’t winning, you can generally expect them to turn it around and vice-versa for a team with bad possession numbers that is winning.

But it’s not as black and white as it may seem. Sometimes, teams with good possession numbers aren’t top teams while some teams with mediocre possession numbers succeed. The two biggest factors that can either make up for bad possession numbers or ruin good possession numbers are shooting ability (either having more skilled players or having a good power play) and goaltending. Sometimes teams are better at one or both than the average team while other times a team can be riding an unsustainable streak of luck in shooting or save percentage.

The best way to measure luck is with a stat called PDO. PDO is just even strength save percentage plus shooting percentage. PDO should regulate to 100. Anything above that is above-average while anything below is below-average. Some teams have above-average goaltending or high-end skill and will naturally have a higher PDO. However, if a team doesn’t have either and still has a high PDO, regression will likely follow.

Deployment – Another important aspect of analytics to keep in mind: a player’s possession numbers can be affected by who he’s playing against most often. A center that’s consistently matched up with the other team’s scoring line will generally have lower possession numbers. Additionally, a guy that can line up against top competition and maintain good possession numbers is a valuable asset. It essentially means he doesn’t allow the other team’s best players to dominate play.

Summer of Analytics – For the most part, NHL teams ignored analytics or didn’t consider advanced metrics when making decisions (notable exception: Blackhawks GM Stan Bowman has consulted analytics since he took on the job in 2009). That changed big time this summer. The Toronto Maple Leafs, whose late season collapse in 2014 was predicted by just about everyone who pays mild attention to possession numbers, hired the guys who ran a popular hockey advanced stats website called “Extra Skater” along with assistant general manager Kyle Dubas has uses analytics in his player evaluation. Other notable hires in the summer included blogger/lawyer/stats guru Tyler Dellow to the Edmonton Oilers and former professional poker player Sunny Mehta to New Jersey. Other teams jumped into the fray with less notable hires as it seems having a numbers guy around the front office is quickly becoming a must.

Puck Tracking – As mentioned above, the NHL is looking into a way to track the pucks’ movements throughout the course of a game. This would be similar to the SportVu in basketball and Pitchf/x in baseball that I’ve finished the last two posts off with. Essentially, it would features a chip in the puck that tracks its every movement. This could create data that would blow Corsi, Fenwick and PDO out of the water and revolutionize the way the game is studied.

For now, just worry about possession numbers and identifying luck. And hit me up on Twitter @nigrelli93 if you have any questions or comments.