A Beginner’s Introduction to Hockey Stats

By Tyler Kerdman (@kerdy19)

As I’ve continued to write more articles, I’ve had more friends contact me asking how they could learn more about advanced stats in hockey and why they’re important. There are many fantastic websites and contributors of new (Scott Cullen, Corsica Hockey, Travis Yost, Mimico Hero, War on Ice, etc…) and of old (Tyler Dellow, Eric Tulsky, Vic Ferrari, etc…) that produce fantastic content, but I’ve been hard-pressed to find one article that is an easy introduction not just to the stats, but why they’re important.

The point of this piece is going to be to introduce anyone you want to why, in my opinion, the new-age statistics are important and some ones that I use on a frequent basis. Hopefully ,by the end of it you realize these ‘advanced stats’ are really not so advanced and are both logical and easy to understand. Let’s begin:

*All Stats from Corsica.Hockey, NHL.com, and hockey-reference.com

Shot Attempt-Based Metrics

Why are they Important?

Have you ever watched a game of hockey and a player winds up for a wide open slap shot in the slot and fires it off the post? In that instance, everyone’s first inclination is, ‘9 times out of 10 that puck finds the back of the net’. The opposite can occur, too. Sometimes a defenceman can launch a shot on net, it hits 4 bodies, then the post, goes off the goaltender’s back, and trickles past the goal line. That may only occur 1 time out of every 20 attempts, but it does happen.

The example above is attempting to illustrate that goals can be very random and occur in a variety of ways. In the first situation, the defense is entirely at fault for letting the player find room in the slot. In the second situation, the players may have done everything they could to prevent a scoring chance but luck turned against them and the puck went in. Therefore, over a large sample of time it is irresponsible to only look at on-ice goals to determine a player’s offensive or defensive capabilities. If so, the player who scores with his shinpad is deemed more valuable in that game than the player who may have created 6 scoring chances but ran into a hot goalie.

Looking at shot-attempts rather than goals widens the sample size and removes unknowns such as puck-luck or goalie quality from the equation. The correlation between shot attempts and shots is incredibly high, and more shots lead to more goals. Allowing less shot attempts against results in less shots against, resulting in less goals against.

Ultimately, looking at shot attempts instead of just goals is focusing more on the process than the result. No team or rational fan ignores their team’s goal differential and hopes that their shot attempt differential is top tier. Rather, it allows for us to say, “that team may not be scoring but they’re controlling play, so maybe goals starting coming soon” or vice-versa. Shot-attempt based statistics show teams and players that control play and which ones are often chasing the game.

What Stats Show This?

Corsi is the mother of all ‘advanced’ statistics. This intricate formula determines a team’s propensity to control puck possession, as measured through shot attempts. The formula for Corsi is as follows:

5v5 Corsi = Shot Attempts For (Corsi For) – Shot Attempts Against (Corsi Against)

Okay, I lied about it being difficult. Corsi is literally just a shot attempt differential. I am adamant that the ‘referendum’ on Corsi wouldn’t exist between stat believers and non-believers if it was simply referred to as shot attempt differential. Often it is expressed as a percentage:

5v5 Corsi For % = (Corsi For) / (Corsi For + Corsi Against)

The relationship between a strong CF% and winning is evident. In 2015-16, the R2 (correlation) coefficient between CF% and points was 0.24, which is a very strong indicator for a statistic not directly related to wins (for reference, R2 between Faceoff % and points is 0.04). What is evident is that greater Corsi differential proves to show stronger teams over large samples. In the 2016 playoffs, all 4 remaining teams were in the top 10 for Corsi For%. The cup winners from 2010 to 2015 have ranked 1st, 14th, 2nd, 4th, 1st, and 2nd in this statistic, respectively.

You may also see people use Fenwick. Fenwick simply takes Corsi (all shot attempts) and subtracts blocked shots. It is used less-frequently than Corsi, but produces very similar results.

Corsi and Fenwick can be used to measure possession of an entire team or possession of an individual player. Corsi and Fenwick can also be observed in an offensive or defensive context only. Looking at Corsi For will show a player’s ability to drive offensive possession, while Corsi Against will outline a player’s ability to suppress the other team’s offense.

‘Even Strength Per 60’

Why are they Important?

Remember when Jonathan Drouin got sent down to the American Hockey League and he demanded a trade? Simply put, Drouin believed that he was treated unfairly and was more valuable to the team than they realized. Looking at his base stats, this is difficult to validate. 4 goals and 10 points in 21 games is nothing terrible, but is not jaw-dropping for a player who is considered to be an offensive dynamo. Looking further into his play, it can be determined he only received 14 minutes and 27 seconds of ice time per game, less than one-quarter of the game. So while he only recorded 10 points in 21 games, maybe his production was only low because of a lack of opportunity.

This is why Per 60 stats exist. For any statistic you can think of, Per 60 normalizes a base result to a count for every 60 minutes of playing in order to account for discrepancies in ice time. So a guy like Ryan Suter may have a stronger base Corsi For than Colton Parayko because he plays so many more minutes, but per 60 minutes of ice time, Parayko drives more offensive possession. Per 60 statistics can be very important because it could identify players that are deserving of more ice time or expose players who are playing too much.

Relating back to Drouin, his points per 60 in his 21 games was 1.98, 2nd on the Lightning. While the sample is small, it shows that he was very productive given his very limited ice time. As evidenced by his spectacular performance in the Lightning’s playoff run (13 points in 16 games), his points per 60 may have been very indicative of his true value instead of his base counts which were impacted by the coach’s ice-time decisions.

What Stats Show This?

Almost any statistic that is base (meaning it’s simply a singular value) can be converted to a ‘per 60 minutes’ stat. This includes Corsi For Per 60 (CF60), Corsi Against Per 60 (CA60), Points Per 60, Goals Per 60, or Primary Points Per 60 (Primary Points = Goals + 1st Assists). All of these statistics are calculated the same way:

5v5 ‘Y’/60 = 5v5 Base Statistic Y/5v5 Time On Ice * 60

Per 60 can be more valuable than base counts for individual players, but can be used in certain instances for team statistics in different contexts. For example, a team could look at their Corsi For per 60 when leading vs. when tied. Even though that team could have played 300 more minutes tied than with the lead during the season, the per 60 element allows the team to normalize the results and compare them on a level playing field.

Team Relative Shot Attempt-Based Metrics

Why are they Important?

I ended off by saying that Corsi and Fenwick can be used to measure the possession of an individual player, which is done all the time. Whether it be the Corsi in a game or over an entire season, an individual skater’s possession impact can be measured.

However, that poses a problem. Let’s say that I look at the CF% of Jordan Nolan, depth forward for the LA Kings. This year, Nolan produced a CF% of 50.12, meaning for every 100 shot attempts he was on the ice for, 50.12 were directed at the opposition’s net. This means that he was neither a liability nor a possession driver, which is very fine. Now, I look at Nathan Mackinnon, the star center for the Avalanche. Mackinnon’s CF% this year was at 46.46, much worse than Nolan’s. Without context, it would be concluded that Nolan is a stronger possession driver than Mackinnon.

This is where team-relative shot-attempt metrics come into play. Nolan plays for the Kings, who finished 1st in team CF% at 56.38. Mackinnon, meanwhile, played for Colorado, who finished last in the league with a 44.20 CF%. Whether it be team talent or coaching style, LA clearly is a much stronger team at controlling possession than Colorado. Therefore, to measure an individual player’s possession impact and remove team quality/coaching quality from the equation, it is more accurate to see how the team’s possession changes when they are on the ice vs. when they are on the bench. This is the purpose of Relative Statistics: to account for these variables and show a player’s true value independent of team.

What Stats Show This?

Relative statistics can be used for many different stats, including CF, CA, CF%, Goals For, Goals Against, or Fenwick-related counts. While they are all calculated in the same way, we will look only at CF% Relative for now:

Corsi For % Relative (CF% Rel) = CF% Player A – CF% of team when Player A off-ice

Let’s refer to the players above. Nathan Mackinnon had a CF% Relative of 3.65, first on the Avalanche. This means, despite his individual CF% of 46.46, the Avalanche’s possession was 3.65% higher when he was on the ice. Meanwhile, Jordan Nolan’s CF% Relative was -7.92%, meaning the Kings’ possession was almost 8% greater when Nolan was on the bench. This shows, despite the difference in CF%, Nolan’s was inflated due to being on a great team and Mackinnon’s was diminished for the opposite reason.

A Caveat about Relative Statistics

An important thing to remember about relative stats is that they can sometimes be slightly skewed based on team talent. With Mackinnon and Nolan, it makes logical sense that Mackinnon, the 1st overall draft pick and future superstar, has a much higher CF% Relative than Jordan Nolan. However, Sidney Crosby had a CF% Rel of 2.61 while Tyler Kennedy on New Jersey had a CF% Rel of 5.55. Clearly, people who rely only on statistics would argue to the grave that Kennedy > Crosby, right?

Not entirely. Relative statistics can sometimes overinflate a player’s value because of team strength. Tyler Kennedy is not one of the 30 most valuable players in the league, and the fact that the Devils are poor possession drivers as a team makes Kennedy’s style of play stand out. However, looking at the team strength and per 60 measurements will account for this flaw.

Puck Luck (PDO)

Remember when the Toronto Maple Leafs made the playoffs in 2013? They made it to game 7 against the Eastern Conference Champions and lost a nail-biter (and nothing else bad happened and they didn’t blow any leads). What about when the Colorado Avalanche finished 3rd in the league in 2013-14? In the following years, both teams finished bottom 10 in the NHL. Contrary to some of the talking heads, this wasn’t due to a lack of compete or heart from the players. These teams were introduced to the scary world of PDO.

PDO stands for absolutely nothing and was invented by a man named Brian King. PDO is simply 5v5 shooting percentage + 5v5 save percentage. Its importance is that it is a very strong indicator of luck. The average PDO in the league hovers around 100.

The thought process is this: As a team, it is very difficult to maintain a high shooting percentage. In 2015/16, the median team shooting percentage at even strength was 7.57% and 16 teams finished with a shooting percentage between 7 and 8%. So when a team like the New York Rangers, who just 2 seasons ago had a team shooting percentage of 6.68%, is the only team to have a shooting percentage of 9%, it appears to be unsustainable. The same can be said about save percentage. At even strength, if the median sv% is around 92.2%, a team that records a save percentage of 93.5 or higher seems to be lucky (unless they have a goalie like Lundqvist who has proven to consistently perform at that high a level).

Relating it back to the Leafs, their shooting % in 2013 was 10.67% and their save percentage was 92.33, resulting in a PDO of 102.99. The next year, their shooting % fell to 8.48 and their PDO dropped to 100.21. Without a strong possession game, their game became dependent on scoring at absurdly high rates and relying on their goalie to save games. This doesn’t mean that PDO is bad: goaltending and shooting luck are sometimes needed to win a Stanley Cup. However, if that is the only basis of a team’s game, it proves to be a problem and many teams have fallen victim to believing their team is better than it actually is because of PDO effects.

PDO can also be used to measure players. There are only a very small handful of players that can sustain a shooting % that is much greater than the average in the NHL, which hovers around 8%. Of players who have played over 400 games in the last 6 seasons (average of 66 per game), only 34 have maintained a shooting % of 12% or greater. Only Steven Stamkos has maintained a shooting % of over 17%. Therefore, when a player like Zach Smith scores 25 goals but has a shooting % higher than 20 (with a historical sh% of 7), it feels like that is unsustainable. Looking at an individual’s PDO can explain why a player’s performance suddenly skyrocketed or plummeted, and can provide reasoning for whether that performance is sustainable.

What’s the Point

Hockey is a truly unique sport. Unlike basketball, baseball, or football, there aren’t really ‘possessions’. No team is really on offence or defence most of the time, and it means that it is really difficult to determine just how strong a player is. Therefore, there are small events that a player or a team performs that, in a vacuum, don’t appear to be important. Winning a seal-out along the board, a successful zone entry, or a smart pass down low to continue the cycle don’t appear to be that significant. However, over a large sample, all of these little things add up.

The advanced statistics that have started to take form in hockey aren’t really that advanced. What they attempt to do is provide the team or a fan with more information that may be difficult to decipher just by watching the game. By broadening the scope from just goals to all shot attempts, looking at player’s possession impacts and score impacts relative to their ice time or their teammates, and understanding the impact luck plays in the sport, simply provides more information and context.

Ultimately, hockey stats attempt to provide more worth to the process than the current system of evaluation does. Goals are great and are clearly needed to win games. Possession statistics look at what drives the goals instead of simply looking at goal differential and assessing team or player value. They don’t ever claim to be all-telling, but it’s just knowledge. With more knowledge, hopefully comes a greater understanding.


1 Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s