Discover more from Oil In Goal
Quantifying the Impact of Goaltending on the Oilers
Measuring how much Skinner and Campbell have impacted the standings
“In hockey, goaltending is 75% of the game. Unless it’s bad goaltending. Then it’s 100 percent of the game because you’re going to lose” -Gene Ubriaco. No team knows this better than the Oilers, who have suffered through multiple seasons of inconsistent goaltending. The signing of Jack Campbell was supposed to shore up the position, while local product Stuart Skinner rode shotgun. The 58 games so far have not panned out this way, with Skinner making 31 appearances to Campbell’s 30. Wins are not a goalie stat, as run support is half the game that can’t be affected from the far crease, but Campbell sports a 17-8-4 record compared to Skinner’s 14-11-4. A quick glance at any stats would show that Skinner has been the better puck-stopper of the two this year, but how has it affected the Oilers’ bottom line in the standings?
Generally, when dealing with statistics, binning results is frowned upon, as the timing of positive or negative events is more or less random so there is greater importance on the entirety of the results. This presents a challenge when looking at the impact on a game-by-game basis, as each game is a discrete sample size, with fixed results. No matter how great or poor a performance is, it is localized to 60 minutes (plus overtime) and can have a maximum impact of two points in the standings. Additionally, since the goalie has no impact on his team’s finishing and little if any impact on his defensive environment, the skaters’ impact is best measured by looking at the Goals For-Expected Goals Against. These results also filter out events on an empty net, since that net would not occur or be weighted the way they are without the slim leads that result in empty-net situations. Each performance from a goalie ultimately falls into one of 6 categories, as defined below.
The team should project to collect two points when they have a positive GF-xGA, and would typically lose when that difference is negative. The difference between how many points are actually collected and that projected point total is the goaltenders’ standing points impact.
When isolating for a single same performance, sample sizes become extremely small which makes it tough to skew randomness from results. The other thing that can happen to a goaltender’s defensive environment in a single game is the quality of chances can skew toward lower or higher danger chances. Public expected goal models are based on the entirety of events for a season, and end up being very accurate over the aggregate. However, in single same samples, there can be a higher ratio of high or low-danger changes compared to normal, which will make the goaltender’s workload more difficult or easier respectively.
To deal with this, I employed a strategy much like MoneyPuck’s Deserve To Win O’Meter, but applied it to goalies. For each individual and independent expected goal event, I’ve “flipped” a weighted coin of whether it’s a goal or not, with the xG value representing the likelihood that it’s a goal. This is repeated 1000 times to create a simulated sample size. The actual goals against totals are compared to the simulated sample size to come up with a goaltender’s performance percentile against his expected environment.
Based on the game breakdowns, I could disagree with some of the specific results that this process yields on a game-by-game basis, but I generally feel that these stats paint an accurate picture.
Here’s how the Oilers’ goalies have fared so far.
Below is Jack Campbell’s game log for the season so far.
Needless to say, it hasn’t been a good season for the 31-year-old. He’s struggled heavily through large portions of the season. Campbell has almost twice as many negative performances as positive ones. He’s also been the benefactor of strong goal support, taking away from the goal differential margin just as many times as he’s added to it.
Of particular note is Campbell’s streak of 12 straight starts with a point. During this streak, he has not had a game where he has single-handedly gained the Oilers’ points. In fact, his play has cost the Oilers 3 points over that stretch, as the model attributes the 3 recent OTL to Campbell.
I’m not sure what more you could ask from Stuart Skinner. As a rookie, he’s been able to stabilize the crease while Campbell went through his struggles. He’s been relatively consistent and still had some high-peak games, stealing 5 wins to date. The first-time all-star has twice as many positive performances as negative ones, which is reliability that has often been missing from Edmonton’s crease.
Where Skinner’s record is hurt comes down to minimized losses, which are losses where he was a positive contributor. So far this year, Skinner has 9 of this type of result, which is more than games where he’s added to a win. Simply put, Skinner has not received adequate run support, which has resulted in the team wasting a noninsignificant number of quality starts from the 24-year-old.
It can be tough to parse out goaltender performance from team performance in results. Media can create narratives, fans can have biases, and playing styles can deceive our eyes. Luckily there is data to accurately parse out how much the goaltender is responsible for each result.
Based on the data, Stuart Skinner is having a very solid year. Not only is his aggregate performance been extremely solid as a rookie, but he’s come up big for the Oilers in moments where they needed it, resulting in an extra 8 points being gained in the standings, even despite what might appear to be a middling overall record. On the other hand, Jack Campbell has been along for the ride while the offense scored goals. His good games have mostly come while the team has been outplaying the opposition, and he has wins in a large number of net negative performances.
Despite some specific start decisions or what some parts of the media might tell you, in a meritocracy, Stuart Skinner is the Oilers’ #1 goalie.
You can find me on Twitter @OilinGoal or on substack
All data used in this article was derived from Play-By-Play Queries from Evolving Hockey