Hockey Stats 101: Learning the basics

This is a crash course in hockey analytics. It won’t be exploring particularly complex topics, but it will provide you with a solid foundation that will make following the game — and finding value in the betting market — easier. Here are some statistics that help us predict goals.


Corsi (a dumb name for a statistic) is the foundation on which hockey analytics were built. While it’s sometimes referred to as an advanced statistic, there’s nothing advanced about it. At its core, Corsi is just plus-minus for shot attempts. It’s useful because shot-attempt differential actually does a better job of predicting future goal differential than does past goal differential. 

In other words, if halfway through a season a team has a shot differential of plus-75 but has been outscored by five goals, research has shown that it is more likely to have a positive goal differential in the second half. Scoring goals requires skill, but it’s also heavily influenced by luck. Routinely outshooting your opponents, however, is a more repeatable skill, and one that can be relied on. 

Fenwick (another dumb name for a statistic) is nearly identical in definition except for one difference. Corsi counts shots that hit the net, missed shots and blocked shots. Fenwick does not count blocked shots. Corsi is generally more predictive than Fenwick, but the latter is still useful when analyzing certain situations.


The previous metrics have flaws that might have become apparent when you read their definitions. When it comes to shot attempts, a shot from the far offensive zone that somehow makes its way to the goaltender for an easy save is counted the same as a shot from five feet in front of the net that would require great skill — or luck — to stop. These are the issues that expected-goals models aim to address. 

Expected-goals models, like the one found at the stats website Evolving Hockey, weigh each unblocked shot attempt based on a number of variables to determine the likelihood of a shot finding the back of the net. Shot location is a big one, but shot type and angle are also taken into account. Some models even factor in things like rebounds and whether a shot was off a rush. 

The debate still rages on whether expected-goals models are more predictive of future goals than Corsi. Using both metrics is common, as Corsi sort of tells us who had the territorial advantage while expected goals lets us know if they also won the shot-quality battle. Most expected goals do not take individual shooter talent into account.


A WAR (Wins Above Replacement) model attempts to assign a total value for every player, which represents how much that player contributed to his team in a single number. Evolving Hockey’s GAR is a metric that assigns a value to players based on how much value they have contributed compared with that of what we would expect from a replacement-level player. WAR and GAR are interchangeable as goals are just the currency used to obtain wins. A win is worth about 5.5 goals.


Whether talking about Corsi, Fenwick or expected goals, these metrics are best expressed in ratio (percentage) or rate (per 60 minutes). For example, if a team registered six shot attempts and allowed four, we would say it owned 60% of the shot share.

Rate stats allow us to measure efficiency. Goals per 60 minutes, for example, is pretty self-explanatory. It tells us how many goals a team scores per 60 minutes of play, which could come in handy when pricing games and period totals. 


Five players and a goalie take the ice for each team at the start of a game. But because of penalties, teams sometimes play up or down a man. Teams can play at even strength (5-on-5, 4-on-4, 3-on-3), on the power play (5-on-4, 5-on-3, 4-on-3) or on the penalty kill (4-on-5, 3-on-5, 3-on-4). The strategy will change depending on the situation, however, which is why it’s important to isolate each area of the game and analyze it on its own.

Most of the time, we should be talking about 5-on-5 play because it is the most frequent situation. Another reason for mainly focusing on 5-on-5 is that both teams have a chance to drive play. At both the team and player level, it’s important to separate performance analysis based on individual situations. However, one exception to the rule exists. Sports bettors reference expected goals during all situations to judge whether the teams they bet on should’ve won or not. 


It’s easy to look at the box score and jump to the conclusion that whoever had more shots on goal was the better team, but that isn’t necessarily the case. Score effects are a well-researched phenomenon across all sports, especially in the NHL. Teams with a lead, particularly late in games, will often sacrifice offensive opportunities in an effort to protect a lead. Teams that are trailing will see this opening and somewhat abandon their defensive structure to try to tie the game.

Consider the following scenarios:

• Team A was outshot 40-20 and outchanced 15-7. Team A won the game 1-0, scoring in the final couple of minutes.

• Team B was outshot 40-20 and outchanced 15-7. Team B won the game 4-3 but led 4-0 after the first period. 

At first blush, these performances may seem similar. The goal, shot and scoring-chance differentials were the same, but Team A won a close game, meaning both teams were incentivized to drive play and create scoring opportunities. 

On the other hand, Team B was in a game in which it dominated the first period, jumping to a sizable lead. Because of strategic and behavioral tendencies, however, Team B sat on the lead and let the opponent get back into the game. It’s because of score effects that score adjustments must be made. 

Score adjustments are simply a method of weighting events. Teams get more credit for generating shots and chances with a lead and less credit for doing so when trailing. Failing to understand score effects can lead inexperienced hockey bettors to believe that a trailing team offers in-game value simply because it is outshooting and outchancing the team with the lead. But even the worst teams typically outshoot their opponents when they’re trailing.


For decades, the two most commonly cited statistics when talking about goaltenders have been goals against average and save percentage. Goals against average (GAA) is the number of goals a goaltender allows per 60 minutes of playing time. It is calculated by taking the number of goals against, multiplying that by 60 (minutes) and then dividing by the number of minutes played. 

GAA is a team statistic, though, and should not be applied to an individual goaltender because it is altered by factors completely out of a goaltender’s control. They don’t control how many shots they face or where the shots are coming from. They definitely can’t control how many penalties their team takes. All of these things influence a goaltender’s GAA, so let’s throw it out.

Save percentage (Sv%) is pretty self-explanatory. It’s flawed, but it’s definitely the better of the two. It is calculated by dividing the number of saves by the total number of shots on goal. Although goaltenders can’t control how many shots they face, at least save percentage tells you, unlike goals against average.

Goals Saved Above Average (GSAA) tells us how good or bad a goaltender is relative to league average, but since shot quality isn’t taken into account, Evolving Hockey’s Goals Saved Above Expected (GSAx) is better. GSAx is calculated by taking the number of goals a goaltender allows and subtracting it from their expected goals against. Wins are not a goaltender stat.

back to news

Live On Air

Streaming Now: The Greg Peterson Experience

play Watch Live radio Listen Live


A Numbers Game: Be careful betting baseball games right now, if they don’t involve teams still trying to make the playoffs. Teams that have clinched or teams that are completely out of it will be approaching games far differently. View more tips.

Ian Cameron: USC at Colorado - OVER (73.5). View more picks.