There’s been some discourse in the Statsies about expected goals and high-danger chances. It seems odd sometimes that “gimme” goals aren’t being chalked up as high-danger, or that somehow the Canucks manage to rack up much better expected goals than their opponents only to be slapped with a 5-goal loss.

Advanced stats aren’t the end all be all of assessing hockey games. Yes, from an empirical point of view, they provide something that can be quantified and measured, an invaluable asset for teams making decisions and fans to better understand everything from set plays to defensive adjustments. At the end of the day, however, they offer a different perspective than meets the eye, which isn’t necessarily better or worse — simply different.

As such, it is important to understand what exactly expected goals and high-danger chances mean, because what they offer is additional value to what can be seen by the eye. It shouldn’t be analytics vs the eye test, but analytics + eye test that can best guide an informed opinion. And, to truly gain proper perspective, comes with understanding what exactly is being measured.

The title of this advanced stat is probably a little misleading. Expected goals aren’t “expected”, per se, but rather the probability of which any given shot will lead to a goal. A better name to think of this is shot quality. What this advanced stat aims to capture is that not all shots are equal. Some are more likely to become goals, while others are less likely to find the back of the net. Unfortunately, to have a pure shot quality stat would be asking for hockey to stop being hockey. The element of randomness within the sport makes every single shot unique. What if the puck is rolling? What if there’s a divot on the ice as the player is about to shoot? It’s trying to chase down something that can’t be fully controlled, unlike a shot in basketball which is much less random. As such, expected goals provides the best measure of how good a shot was.

Teams and fans should not view an expected goal as “expected.” A team generating high quality chances, more often than not, will have a high xGF and xGF% share, but will usually not come close to what their actual goal total is. But why is that?

Measuring expected goals takes historical data and puts it through a formula. NHL stat tracking will record unblocked shots (Fenwick) along with tons of other variables on a singular shot event, such as who the shooter was, where the shot happened, when the shot was taken, what the shot was, the list goes on. What publically available xG models do, is that they take these stats and plug them through their own algorithm, which trains their models to calculate the probability of a goal from specific areas of the ice.

There is a ton of information that goes into these models. For Natural Stat Trick, the website that we use here at CanucksArmy to get most of our stats, they do not provide the variables that they use (as are their IP rights). However, we can assume that the list is similar to that found on MoneyPuck, which lists the following:

Shot Distance From Net
Time Since Last Game Event
Shot Type (Slap, Wrist, Backhand, etc)
Speed From Previous Event
Shot Angle
East-West Location on Ice of Last Event Before the Shot
If Rebound, difference in shot angle divided by time since last shot
Last Event That Happened Before the Shot (Faceoff, Hit, etc)
Other team’s # of skaters on ice
East-West Location on Ice of Shot
Man Advantage Situation
Time since current Powerplay started
Distance From Previous Event
North-South Location on Ice of Shot
Shooting on Empty Net

The weighing and value of each of these categories are different between models but remain generally consistent in terms of what they’re trying to achieve. They will create their own probability model of the likeliest event to happen in that specific situation. That is how expected goals are calculated and produced.

That’s because it’s pretty tough to say that a single chance is a guaranteed goal. Infamous empty net misses include Patrik Stefan and Craig Smith, whose contributions include driving down the historical success rate of empty net goals from in the crease. Obviously, I’m highlighting two exceptional examples, but the point is that no scoring chance should be viewed as a guaranteed goal. That is another thing that expected goals tries to capture, that even though the probability of a shot will increase, it is highly unlikely that any shot no matter how high quality will definitely become a goal.

It is simply not possible in a game of hockey to absolutely score on a “gimmie” chance. What if the puck skips on the player and he hits the post? What if he tips it off the bar? What if he has 6×4 but comes up zero by rifling the puck wide? All of these factor into why there are no certain goals. And, with a league average shooting percentage that hovers around 10% a year, it makes sense for even high-danger chances to result in a fractional xG.

Comparing an absolute event (goal) to a continuous event (expected goals) will always result in discrepancies. It doesn’t mean that this stat is useless. Rather, it gives a look at what kind of opportunities a team is creating, how much they are creating relative to their opponent, and, in individual players’ cases, how lucky (or unlucky) they have been and what the trends say about their future.

Higher xGF will mean that a team is getting high-quality chances. Most of their opportunities are more likely to become goals, and through accumulating that number, we can get a number of how many goals they are likely to score assuming league-average shooting and league-average goaltending. This is what makes xGF% more valuable, as it gauges which team is producing the better opportunities and is more likely to score. It can lead to further calculations, such as the differential between the two teams in their xGF, or their rates relative to each other.

For players, expected goals can help determine if they are earning their goals or simply have gotten lucky. For instance, if a player has scored 20 goals but is only recording 13.4 expected goals, then there’s probably some luck involved in the goals that he has been scoring. Conversely, if a player has only scored 5 goals but has recorded 10.2 xGF, then we can make the inference that he’s probably been getting unlucky and a correction to the mean is coming. This is assuming league-average finishing ability too — which is definitely not the case for some players, and is also why the eye test is also important with what the stats provide.

Firstly, let’s go over the definition of a high-danger chance. According to Natural Stat Trick, it goes a little like this:

Attempts from the yellow areas are assigned a value of 1, attempts from the red areas are assigned a value of 2, and attempts in the green area are assigned a value of 3.

Add 1 to this value if the attempt is considered a rush shot or a rebound. A rebound is any attempt made within 3 seconds of another blocked, missed or saved attempt without a stoppage in play in between. A rush shot is any attempt within 4 seconds of any event in the neutral or defensive zone without a stoppage in play in between (originally defined by David Johnson on the now-offline Hockey Analysis, and modified to 4 seconds by War-on-Ice).

Decrease this value by 1 if it was a blocked shot.

Any attempt with a score of 2 or higher is considered a scoring chance.

From the above definition, it gives a very stringent picture of what a high-danger chance looks like. And for good reason — if all good scoring chances were high-danger, then none of them would be high-danger. Inherently, shots that register high in xG will more likely be high-danger chances, as evidenced by the uptick in both stats on a power play.

It’s why fans are probably surprised to see the Canucks not give up as many high-danger chances as they think they should be. Yes, they do give up a lot of high-danger chances, but those two-on-ones from the blue line in? Those probably fall under middle-danger. This could be for a wide variety of reasons – taking more than four seconds to get the shot off, having the initial chance blocked, shooting from at the top of the faceoff circle, the list goes on. They look easy to score, but under the definitions of what a high-danger chance is, they probably aren’t falling into that category.

Just because a chance looks easy and/or is finished off easier does not mean that it is guaranteed to go in. There’s a lot of factors and variables to consider, specific definitions for the advanced stats because rigour is needed for the best possible numbers-based analysis of what is happening on the ice. Fans might have a perception of what counts or does not count, but for the data sets that are being produced, there’s a clear black-or-white categorization that allows for quantitative data to be collected.

I do hope that this has been helpful in understanding what goes on behind the stats.