Breakfast Salvos: Stray observations on the Minnesota win, but mostly about Miro Heiskanen's rating in The Athletic
This morning is a veritable charcutier board of hockey discussion.
I didn’t realize Miro Heiskanen would make such a point of being the story of the night, so we’ll talk about him (a lot) in a bit.
Minnesota didn’t exactly ice an NHL-caliber roster. Conversely, Dallas had some of their best veterans ready to put in some reps. Heiskanen, Kyle MacDonald, Mason Marchment, Jamie Benn, and Emilio Pettersen were the goal scorers in a perfunctory 5-2 victory at American Airlines Center.
I was happy to see Pettersen garner some attention and momentary acclaim. Originally from Calgary’s system, he came to Texas in the Riley Damiani trade and his game was noticeable almost immediately. He’s shockingly dynamic, which you wouldn’t suspect judging by his meager point totals. Despite profiling like something of a wheelhouse passer ala the Roger Corman1 version of Jonathan Drouin or something, he’s got grit and bite to his game, which might explain why his production is always capped.
All the same, he was the most noticeable forward prospect versus the Wild. Of course, I’d be remiss if I didn’t talk about Antonio Stranges. He was noticeable too in all the ways that Stranges is, for better and for worse, noticeable. If you missed my analysis of him last month, now’s the time to catch up. If you don’t care, that’s fine too. It only takes one clip to extrapolate all that’s needed to understand his game.
Excellent stickwork, puck control, and positioning — all undone in an instant by, well…stickwork, puck control, and positioning.
I don’t bring this shift up to pick on Stranges. He had a beautiful rush earlier in the game. But as one of hockey’s truly great weirdos, he’s worth a highlight for more than just his moves.
The last topic I’ll mention before we dig deep into the Heiskanen debate is Matt Dumba. He still has moments where the lost step or two in his skating is visible, and he gets beat. But the power part of his game looks like vintage Dumba. He even blew up an old friend in victory green: Devin Shore. This is the kind of play that’ll win fans over. But it’s also exactly what Jim Nill brought him in to be. How that manifests in actual on-ice effectiveness remains to be seen, but if Heiskanen-Dumba becomes a thing, I’m on board.
And speaking of…[takes a deep breath].
Heiskanen well below Fox, Makar, and Hughes?!
Dom Luszczyszyn posted his 2024-2025 player tiers in The Athletic. In it, the top 150 players were grouped into five groups: MVP cornerstones (Tier 1), franchise pillars (Tier 2), All-Stars (Tier 3), bona fide stars (Tier 4), and elite support (Tier 5). Right away the comments blew up.
You know how this goes: your favorite player wasn’t ranked higher than the guy you don’t watch as much, and so you starting get mad at Dom about it. Obviously, that doesn’t describe all fans, but it probably describes more fans than they’d like to admit. However, it certainly doesn’t describe Garrett Hohl, or Micah Blake McCurdy; all-stars in the analytics community. And all three had quite the lively, if somewhat heated, debate. Those weren’t redditors, two Modelos deep, going after an author. This was philosopher Henri Bergson giving Albert Einstein a check left hook in 1922.
I’m taking a few liberties with the drama that wasn’t really there, but I really enjoy seeing high level debate, even if words like “sensibly distributed residuals” go over my head entirely. What I appreciate about moments like this is how much it illustrates a base-level misunderstanding of “analytics” by combative casual fans and lazy broadcasters, namely this: analytics are not trying to replace what you already know—or what you think you know. They’re trying to categorize what we can’t readily arrange.
I can’t speak from the perspective of a mathematician, or a data scientist. But I can certainly speak for my gut; my intuition; the little hairs on the back of my neck. And my guts says. “Heiskanen in the second tier with Cale Makar, Adam Fox, and Quinn Hughes in the MVP cornerstone group? [Expletive deleted] outta here!”
So I decided to peak into the rabbit hole, starting with Micah’s synthetic goal model, which now has a comparison tool. Here’s how Heiskanen (the blue line) compares to the Tier 1 defenders.
I reacted to Micah’s debut of his sG model last year, which I’d recommend reading if you have an appetite for this kind of stuff. sG is not just xG with a new coat of paint. The difference is very subtle, but very distinct.
Again, always try to look at these as a player’s broad value outside of their production. Production is not irrelevant, especially with players like Makar and Hughes who excel at it, but to the extent that we have a better glimpse into a player’s shift-to-shift net rating, Heiskanen has grown from a good defender, into a great one, and now he’s in that rarified air of being transcendent. Purely going by sG, Heiskanen belongs in the tier Dom didn’t have Heiskanen in2.
However, this is not an attack on Dom’s model, nor am I using sG because it suits ‘my wish’ for Heiskanen to be the best. It’s easy to describe fans as systematically blind to certain features of a player for so many obvious reasons. But what if there is a systematic blindness to certain features of a position on the part of the models evaluating them?
A digression: about that weird Makar trajectory
I really want to stress why it’s so important to distinguish between production and performance, especially for defensemen. Not only because I think it’s relevant to the discussion in general, but because the 2023-2024 season between Heiskanen and Makar was such a perfect case study in polar opposites. Production is not always misleading, but it can be. Makar had a career high in points this year with 90, 34 more than Heiskanen. That’s incredible. How about his performance outside of his production?
Hot garbage.
This is the kind of profile that might look good on a veteran AHLer making the jump into the big leagues, but not a Norris winner. RAPM is one of the better stats for categorizing performance (insofar as we can), so I consider this information relevant. Makar’s sub-standard shift-to-shift impact is a big reason why his connection with Nathan MacKinnon wasn’t near as dominant; a point well-documented in The Athletic’s playoff preview of the Colorado vs. Winnipeg series. Now, there was a good reason for this. Makar took a nasty-looking spill from Kyle Okposo early in the season, and it wasn’t that long ago he took a blindside hit straight to the head from Jeff Carter, either, so it’s widely known that he was less than 100 percent in multiple areas for most of the season.
Heiskanen vs. The World
For years, I never actually felt like Heiskanen was truly the best defensemen in the league, even if making that determination will always be kind of nebulous to begin with. To me, Fox and peak Makar were slightly better defensemen for different reasons, with Heiskanen and Charlie McAvoy neck and neck for different reasons ever so slightly below them.
However, as Heiskanen has changed, so has my estimation. I’d also argue that we’re still at a crossroads in evaluating defensemen the way they deserve. Corey Sznajder’s Defenseman Compass really feels like the beginning of something broader, and more essential to getting at the essence of a defender’s three-zone responsibility, and thus, their value in relation to one another.
None of this means that I think less of how other models view Heiskanen, whether it’s Dom’s Net Rating, Micah’s sG, or Josh and Luke Younggren’s RAPM. In fact, as you can see, I’ve leveraged some of the other models to argue against Dom’s estimation. But I also don’t think Dom’s model is wrong, either.
This is not a call to modelmakers to create better ones just because Heiskanen “exposes a contradiction.” Contradictions can, and will come up; there’s that old data cliche about how all models are wrong, some are useful. Rather, models that rely so heavily on shots and shot quality may not necessarily capture the ceiling on Heiskanen’s value when his game is so based around elevating the floor.
But here’s the thing: that floor is starting to reach the ceiling. Thanks to Corey Sznajder’s tracking data, we know something critical about Heiskanen’s passing game and how it compares to his peers—a feature every model except Micah’s sG project doesn’t account for. Thanks to Louis Boulet’s work, we now have a visual and a percentile through which to evaluate him compared to his peers.
That looks top tier to me. And offense has never been his strong suit. Or so the sentiment has always gone.
About last night
It’s only a preseason game, but seeing Heiskanen on his strongside reminds me of how long we’ve had this debate, and why it’ll remain relevant as long as he isn’t playing there full time. It’s kind of lazy, but I don’t think it’s irrelevant to ask: how would Fox, Makar, and Hughes look playing on their weakside?
One thing’s for sure. Heiskanen can’t do stuff like this on his weakside.
It’s the thing I can’t get out of my head when I see the data calling into question Heiskanen’s inclusion into the highest of tiers; not because I’m a Stars fan, or because he’s one of my favorite defenders to watch — but because I feel like there’s something missing in the calculus of defensemen. Again, it’s just a feeling.
But seeing Heiskanen on his strongside provided that little extra clarity. As does knowing that all those Tier 1 defenders have had strong defensive partners throughout their career (Lindgren with Fox, Toews with Makar, and while Hughes is the exception, as Vancouver has saddled him with also-rans through his career, that is no longer the case), while Heiskanen has been weighed down rather than built up.
But let’s assume the data is correct, and Heiskanen is in a Tier just below Fox, Makar, and Hughes. Is anyone gonna care if he wins Dallas a Cup?3
I’m not a Marvel person. In fact, I consider it my moral duty to oppose Marvel films. But I remember seeing the trailer for Roger Corman’s Fantastic Four back when it was in production, and had never been so amped as a kid.
For more on Dom’s model, I’ve written an explainer card here.
Kind of a stupid question, actually. This is my job, so yes, I will actually!
I knew that Stranges clip before I even played it. I watched that exact moment last night and said to myself "look at him go" in the same way I would if I was watching my half wit dog set loose in a dog park for the first time to unleash his ultimate spaz out. Stranges is entertainment
I read the same thread you did, and I'll go further than you regarding different catch-all numbers. I think Dom's GSVA is a tier below (lol) Micah's sG and Evolving-Hockey's xGAR as far as putting a number to a player's *talent*. In fact, if I were giving them tiers, I'd put sG in 1A, xGAR in 1B, GAR in 2A and GSVA in 2B.
The reason is really simple, GSVA is very heavily points-dependent. If I understand it correctly, each player has a game score for each game* and Dom adds up the game score for each player on each team for the full season, then feeds it into a regression for goal differential to get how much each player's game score contributed to the whole, then he scales it for the number of goals scored for the season. Heiskanen's point totals dropped last year, which hurts him in GSVA especially, but I'd point to the Stars power play trotting out Benn and Pavelski all year long as the reason for that.**
There's a little cognitive dissonance here, because Dom's team projections are consistently the best in the public sphere and the only model that consistently beats betting odds, but I think GSVA is picking up some sort of team effect and not necessarily assigning value proportionally to talent.
* https://hockey-graphs.com/2016/07/13/measuring-single-game-productivity-an-introduction-to-game-score/ this is an old version, and he's updated it with xG and QoT/QoC adjustments, but points is still the biggest driver.
** sidenote: I feel like Robertson and Hintz were similarly undervalued in the exercise because their point totals dropped last year, but that was mostly a consequence of Pavelski aging.