Notes
Evergreen note: “Analytics” are not mere numbers. They’re models: ways of bridging practical (what happens on ice) and theoretical knowledge (what will or can happen based on prior events). They help define the conditions of performance over a period of time rather than merely reference a series of outcomes the way plus-minus does. The climate versus the weather; defining the signatory rather than just the signature. These are not meant to be a right or wrong way to look at a hockey game; they’re meant to be useful in understanding the conditions of hockey itself.
The Stat: Synthetic Goals (sG)
The question sG attempts to answer: What is the scheduled value of a player, from shift to shift, once we isolate their impact from outside influences?
How it answers that question: sG is somewhere between an expected goal model, and a WAR stat, as I understand it. But first let’s start by discussing Micah Blake McCurdy’s expected goal model. His xG model measures a player’s shot differential, and the likelihood of those shots becoming goals (or not) based on where the shot was taken if we assume league-average goaltending. Here we have the following variables:
even strength offense
even strength defense
power play offense
short handed defense
penalties taken/penalties drawn
teammates
competition
score effects
zone usage
coaching
home-ice deployment
player tendencies (new)
blocked shots (new)
time since player started the shift (new)
a goaltender’s tendency to produce rebounds (new)
You may be wondering what makes some of these factors relevant. Here’s why: (I’m only gonna go over a few)
Re: penalties drawn/taken. Drawing or taking a penalty increases the likelihood of a goal being scored from eight percent (even strength odds) to 24 percent (odds of goal being scored in a special teams game state).
Re: Home-ice deployment. A team’s chances of scoring a goal in the second period, on home ice, goes up by almost 17 percentage points. Conversely, teams are much better defensively in the third period on the road, receiving a 7.9 percent boost in decreasing goal probability, assuming league-average goaltending.
Re: score effects. Trailing teams are more likely to score. Specifically, the odds of scoring go up 5.5 percentage points if down by three, 7.1 percentage points if down by two, and 4.5 percentage points if up by one.
Re: Rebounds. A rebound increases the likelihood of a goal being scored, but that increase is higher for when a goalie must stop a rebound by a different shooter.
The point in calculating all of this and including so many numbers is to cut down on bias and get down to a player’s true value. After all, don’t we want to know if a player is getting an unfair edge by only playing more in trailing situations, or have numbers inflated by the power play, or because they have better teammates? Think of xG as the ultimate in plus/minus, only instead of adding up goals, and judging the total difference, we’re looking at shots each shift, and trying to isolate that player’s impact from what they contribute on their own versus what may be artificially contributing to said impact.
“Ok but what does that have to do with sG?” Synthetic goals take everything we explained above, and are then put into a separate math blender. This new blender is everything from the above model, but evaluated against one thousand minutes of EV time, 200 minutes of PP time, and 200 minutes of PK time. The point here is to not only give us a better sense of scheduled or future value, but also to give us a single-value stat.
“Ok how does this help me understand hockey better?” You’re a good student, and thank you for asking. To me the crucial insight that “analytics” offer is being able to resolve contradictions. Consider Matt Duchene. Good player, right? Older, sure, but he scored 65 points in the 2023-2024 regular season. How does a 65-point player rate as a -3.7 in sG?
(See attachment below)
Remember that separate math blender we mentioned? That’s because, again, we’re trying to assess their intrinsic value, from shift to shift. But first let’s unpack this graph. Red means more shots than league average, blue means fewer. So it’s always good to see red in the offensive section, and blue in the defensive section.
Top left: Duchene decreases the chances of his teams scoring a goal by four percentage points (the -4) at even strength.
Top right: He’s a hard positive on the power play, increasing the team’s chances of scoring by eight percentage points (+8)
Bottom left: Opponents are seven percentage points (+7) more likely to score when Duchene is on the ice
Bottom right: Duchene had neither a positive or negative impact in his very very few shorthanded minutes.
Middle numbers: Finishing refers to a specific assessment of a player’s impact on offense per shot). Setting refers to Duchene’s player passing, etc.
All of it, again based on isolating Duchene’s impact through all of the above, adds up to a player who was a net negative per sG.
“He scored 65 points, and was a +15 in the regular season. This model must be wrong.” Again, models only respect the questions trying to be answered. If we go back to the original question in the context of trying to figure out Duchene’s isolated impact now and in the immediate future, we want to discover something useful.
In which case, I think we do. Micah’s model doesn’t say Duchene couldn’t produce, only that we could expect his impact to be a net negative outside of his production. These models rely on more and more data in order to become more and more descriptive. Duchene’s season on the whole, I think illustrates these insights, since once Dallas was isolated from weaker competition, and once the game became more of an even strength battle, players like Duchene didn’t make the impact that his regular season production would lead you to believe.
This explainer only scratches the surface. For more reading, I recommend going straight to the source.
References:
hockeyviz.com/txt/sG (HockeyViz)
hockeyviz.com/txt/magnus7EV (HockeyViz)
dcastillo.substack.com/p/in-laymens-ter… (Stars Stack)
