Through the Fog of Jesperi Kotkaniemi’s Star-Crossed Season, his Peripherals are Glowing

By: Gordon Liang

There is no bigger casual in the sport of hockey than the guy who judges players solely off how many points he has. In doing so, he is selling himself on at least one of three lies:

1. You can’t be producing if you’re not scoring.

2. Two players with equivalent goals and assists are equivalent offensively.

3. Every player is given an equal opportunity to score points.

The first lie is pretty easy to discern. Think of the numerous times in a Boston Bruins game where a team has accumulated so much time in their offensive zone that the broadcast starts showing you how long it’s been since the puck entered the zone. Think of how often a goal actually came out of it. Or whenever you see an odd-man rush where the goalie makes an uncanny save. That’s production. A point won’t go on your stat sheet but that’s production. You did everything right but so did the goalie.

The second lie is pretty obvious as well. Dakota Joshua and Rob Fabbri.

Nobody has been on the wrong side of this more than former third overall pick Jesperi Kotkaniemi. After a booming start for Kotkaniemi, scoring 13 points in the first 14 games of the season, it looked like the 23-year old was finally seeing the breakout he’s been waiting for. However, after hitting a chain of scoreless droughts which saw him scoreless for 21 days, bringing him to the third line, and then another 17 day drought before a 37 day drought when he was finally demoted to the fourth line where he spent essentially the rest of the season, even taking some time off as the Canes’ healthy extra.

Figure 1: Jesperi Kotkaniemi accumulated 15 points from the start of the season (10/11/2023) to the end of November. Then followed up with three points in the next two months.

Kotkaniemi finished his season with 27 points (12 goals/ 11 primary assists/ 4 secondary assists) in 1027:16 minutes on ice. But that doesn’t tell the whole story. Removing secondary assists– because those are often stat pads– Kotkaniemi’s point/60-minutes rate increases from 0.99 points/60 to 1.34 points/60 which is a testament to his growth as a playmaker and how he’s been more involved in his limited time on the ice this season.

If you’ve watched the Hurricanes this season and focused in on Kotkaniemi’s shifts since his demotion to the fourth line, you’ll probably notice a theme of generating offensive zone time but alas a lack of scoring. This is the nature of playing in the fourth line. In an article written by Ryan Henkel, Kotkaniemi noted the differences in his role as a second line center and as a fourth line center.

Kotkaniemi told Henkel, "When you play less, you kind of play a different way. Making sure that the other team's not putting the puck into your net when you're out there and just trying to get a breather for the big boys. I think that's the biggest thing with being on the third or fourth line.”

It’s not as simple as reducing your time on ice (TOI). Playing in the bottom six places a bigger emphasis on staying out of your own zone and playing solid defense until the offensive weapons arrive. Sure, scoring would be nice and you’d want a $4 Million center to do so but to expect it from the fourth line is unreasonable as it’s not his role.

[NOTE: This next part gets pretty technical and might be hard to follow for people who are less involved in data science. You can skip to the next highlighted bracket without losing much of the content if you feel overwhelmed with the jargon]

Therein lies the third lie. To understand Kotkaniemi’s season from the context of his role, we need to adjust his ice time by the line he was playing on. To accomplish this task, we used shift data from NHL’s API and formed a column for every second of the game. A player would receive a 1 if they were on the ice at that specific second and a 0 otherwise. This allowed us to use K-Means Clustering to interpolate forward lines and defenseman pairings. Each cluster (a line or pair) was then ranked by total time on ice that game.

From there we’re able to separate a player’s time on ice by his line or pairing and fit two separate linear regression models on goals and points according to his adjusted time on ice like so:

Where our error term for both models ideally have constant variances (Spoiler: they do not). We trained our model on a pseudorandom subset of players (separated by season so Connor McDavid in 2020-21 is a different sample from Connor McDavid in 2021-22) from the 2015-16 season to the 2022-23 season. Then we validated our model on the other subset that was excluded from our training.

To avoid confusion, “Expected Points” and “Expected Goals” are used interchangeably with “Weighted TOI (wTOI)” below but not with each other because our TOI is essentially weighted by line/pairing and then multiplied by a constant to get expected goals or expected points. However, the weighing differs based on which statistic we’re trying to predict.

Figure 2: The goals model against our validation data showed a linear relationship between expected goals (TOI weighted by line) and actual goals scored. An intercept term wasn’t fit because a player that doesn’t play should score no goals. RMSE = 4.33

Figure 3: The points model against our validation data showed a linear relationship between expected points (TOI weighted by line) and actual points total. An intercept term wasn’t fit because a player that doesn’t play should score no points. RMSE = 9.38

As alluded earlier and evident in the figures above (and below), the variance of our error terms aren’t constant. In other words, as wTOI increases, so will the uncertainty of the model. Which generally isn’t ideal but, in a way, is intuitive. The more a player is on the ice in higher scoring situations, the more opportunity he’ll have to break out of the norm in either a good (Auston Matthews) or bad (Johnny Gaudreau) way.

Figures 4: Both models experienced heteroskedasticity. The standard deviation of goal residuals was around .227*expected goals + 0.782. The standard deviation of points residuals was around 0.187*expected points + 1.478

Great! So we’ve shown that more TOI leads to more goals and points… but I could’ve told you that before writing all this code. The point of all this was that our TOI was weighted by lines. How does playing on the first line differ from playing on the fourth line? How does playing as a forward differ from playing as a defenseman. To further analyze this, we incorporated a pseudorandom bootstrapping method and ran 1000 linear regressions based on our bootstraps to develop a 95% confidence interval for the coefficients of each feature.

Figure 5: At each bootstrap, we saved the coefficients attached to each feature (TOI by line) and plotted them above (Left: coefficients with respect to goals scored, Right: coefficients with respect to points scored). 95% Confidence Intervals for goals scored by minutes on a line:F1 TOI [0.0184,0.0189], F2 TOI [0.0141,0.0145], F3 TOI [0.0058,0.0065], F4 TOI [0.006,0.0063], D1 TOI [0.0046,0.0047], D2 TOI [0.0023,0.0027], D3 TOI [0.0013,0.0015]. 95% Confidence Intervals for points by minutes on a line: F1 TOI [0.046,0.047], F2 TOI [0.033,0.034], F3 TOI [0.013,0.014], F4 TOI [0.014,0.014], D1 TOI [0.021,0.021], D2 TOI [0.011,0.012], D3 TOI [0.008,0.008].

Most of the interpretations are intuitive. Defensemen tend to score less goals as they’re usually chilling at the point and preventing zone exits. That’s evident in all of their goal peaks being on the left while the forwards are on the right. What’s less intuitive is the stark difference between the top six’s coefficients and the bottom six’s. It’s so substantial that the bottom six are between the defenseman pairings. And mind you, this is for every minute on the ice. 

[TECHNICAL PORTION OVER]

With context to his line, Kotkaniemi’s results are right on par with what someone should expect with bottom six TOI. Judging solely from his wTOI, we’d expect his goals and points total to be at around 11.457 and 26.158 respectively. His actual totals are 12 and 27 respectively.

Figure 6: Jesperi Kotkaniemi played 61:38 minutes on the first line, 347:41 on the second line, 228:38 on the third line and 389:19 on the fourth line. We'd expect the average player with this ice time to score around 11.457 goals. Kotkaniemi scored 12 goals.

Figure 7: Jesperi Kotkaniemi played 61:38 minutes on the first line, 347:41 on the second line, 228:38 on the third line and 389:19 on the fourth line. We'd expect the average player with this ice time to score around 26.158 points. Kotkaniemi scored 27 points.

The difference in production between the top six and bottom six goes far beyond getting less time. It’s also the way that time is used. You can give a fourth liner more ice time but if he’s still playing with the objective of setting up the first line, he’s probably not going to see that twine move anytime soon. So it’s unfair to compare points total across players with different roles due to the difference in objective between them.

In the role he was playing, however, he excelled. According to NHL Edge, when Kotkaniemi was on the ice in even strength this season, the puck was in the offensive zone 45.8% of the time, putting him in the 96th percentile. In a line that isn’t focused on generating offense, he excelled at it anyways.

According to Natural Stat Trick, Jesperi Kotkaniemi’s 3.16 5-on-5 Expected Goals For (xGF) per 60 minutes ranked 26th highest amongst the 428 skaters with 800 or more 5-on-5 minutes on ice this season and 22nd highest amongst the 258 forwards in that group. xGF is a metric that considers the chances generated when a player is on the ice and estimates how many goals scored they’d expect based on those chances.

However, his actual Goals For (GF) tells a different tale. He ended his season with a xGF-GF per 60 of 0.71 which ranked 20th highest amongst skaters and 15th highest amongst forwards showing a lack of finish or simple bad luck whether it be a goalie just making an insane save or perhaps a case of overpassing on odd-man rushes like we saw when Teuvo Teravainen was on his line.

From his increase in primary assists to his 90th percentile xGF/60 amongst forwards, the metrics undeniably portray Kotkaniemi as an elite playmaker. If nothing else, he’s grown into a player that’ll generate scoring chances in a way that not many players can. In a season that saw him go from the first line when Sebastian Aho was injured to a healthy scratch and with the numerous unconverted chances, it’s honestly remarkable that his goals and points were still on par with his wTOI.

At the end of the day, the NHL is a results-based business. The Carolina Hurricanes know as well as anyone that sometimes you can do everything right and it doesn’t matter if you don’t come home with a win. They saw it first-hand in the Eastern Conference Finals last year. Kotkaniemi isn’t a perfect player by any means and he’s still got room to grow at 23. And no one would fault the Hurricanes if they err on the side of trading him this offseason in hopes to clear up cap space for free agents like Seth Jarvis, Brady Sjkei and Jake Guentzel amongst other players whom this organization holds dearly to the heart. But if they do, I’d expect all 31 general managers to jump at the chance to acquire the playmaker that Kotkaniemi is while also gaining extra value in the return to take on his contract.