If you’re in the intersection of “analytically-inclined people” and “Devils fans” then you are likely very confused by some of this year’s early season results. That’s because, while there was a bit of turnover in the roster, a lot of players returned as well. And those returning players, well, they’re basically all doing the reverse of what you’d think. To demonstrate this we’re going to look at RAPM.
For those who don’t now anything about it, I’ll give a quick explanation of what your RAPM would be. To start, you need a base stat (GF, CF, xGF, etc.) — we’ll use xGF% which is is the percent of expected goals (think shots weighted for danger) your team is on the ice for. Then Rel_xGF%takes the games you’ve played and finds the difference between your team’s performance in that metric with and without you on the ice. This adjusts for the overall quality of your team. RelT does the same thing, except instead of comparing the teams performance with and without you, it takes every player’s performance with or without you and sums up all of those impacts (weighted for TOI spent together). This adjusts for the overall quality of your linemates. RAPM runs a regression that attempts to predict the xG differential of every shift using only contextual variables (Zone, Score, Venue, Back-to-backs, etc.) and the names of all the skaters on the ice. This adjusts for teammates, opponents, and context — it’s a supercharged “Rel” metric.
So when we look at RAPM xGF%, what we have is the isolated impact of a player on the team’s xG ratio. And the Devils results to this point in the season are just plain stupifying.
I mean, what the hell is this? I mean, it’s hockey, so there’s not a lot of knowns, but if you had to guess about the most likely analytical results for the year, towards the top of the list would probably be three things:
- Taylor Hall will be good — even when points weren’t coming, he’s been analytically amazing for his entire Devils tenure.
- Will Butcher will be either best or right behind newcomer Subban among Devils D — Butcher has led Devils defenders both of the last two seasons in the metric and is still fairly young.
- Andy Greene would continue to decline towards AHL-level performance — he went from +0.133 to +0.85 to -0.041 the past 3 seasons and was entering his age-37 season where things are not pretty for defenders.
So, as you can see for the chart we are 0 for 3 in those projections. And not only would we have been wrong, we would have been about as wrong as one could be. If you want a couple bonus ones: Nico = good, Zacha = bad was also a good bet coming into the season and those don’t appear true either. Now, this is a small sample and it’s only at even-strength, but even if we look at GAR (goals above replacement) instead, the story is the same. Look at the chart of these 5 players I mentioned over their careers.
Put simply, the good players (Hischier, Hall, Butcher) are all having the worst years of their career, and the bad ones (Greene, Zacha) are having the best years of their career. All of them, if viewed in isolation, would likely be considered “outliers” for the individual players. So what’s happening here? We can’t just have 5 different outliers happening simultaneously. I’m going to focus on the 3 examples I gave before because Nico hasn’t been that bad and Zacha is still very young, so improving isn’t that unusual. The other 3 make no sense thought so I’d like to dig in a little.
Hall is the only MVP in franchise history and he earned it. I mentioned in my offseason piece about what contract he should or should not be offered that, coming into the year, only McDavid, Stone, Barkov, and Stamkos had a higher GAR per game than Hall over the last 3 years. He’s an elite player who’s likely going to earn a boatload of money starting next season.
So what gives? His point production is there, sure, but the on-ice impact no longer is — and you need both to be worth the kind of money he’s likely seeking. Earlier this year, John took a deservedly critical look at Hall’s season thus far. The numbers are unambiguous — at even-strength Taylor Hall has had a severely negative impact on the pace of play, danger, and results — his CF%, xGF% and GF% impacts are all negative. He’s had a hard time settling the puck, he’s not making good choice on the pass, and he’s had no aggression off the puck. The last one, you can see directly in his career-low takeaway rate, but also in the results — he has the worst RAPM xGA/60 on the team. This means that isolating for teammates, opposition and context; the Devils’ defensive results are worst with Hall on the ice.
Why is this happening? I came up with 3 possible reasons, though feel free to add your own. Reason one is the least pernicious: he spent a long time away from the game and some of the finer points are still settling in. I don’t know how likely I think this is. I’m not sure if it should take 20+ games to re-adjust after a ~9 month absence, but if it does take that long, I’d expect that we’d gradually see improvement, and I don’t think that’s really happened so far.
The 2nd reason is still very reasonable, but a little more worrisome: He’s 28 years old and he’s been injured quite a bit. Aging and injury have cut short plenty of players’ careers and Hall has missed 138 games due to injury over his 9-year career so it’s not unreasonable.
The 3rd is the galaxy brain take: Hall has given up on this team. I don’t believe this is the case, but I know many people believe with his pending UFA status and the way he’s acted towards the fanbase verbally and physically, he’s on his way out. If his heart’s not in it his game will suffer. Again, I don’t believe this is the case — but some do, and that belief is not entirely without evidence.
Over the past two seasons, Will Butcher has been to the Devils defense what Taylor Hall has been to the offense, analytically speaking. He’s been the most unflaggingly positive impact player the past two seasons, in fairly convincing fashion. If you graph every defender’s season-long RAPM xG +/- over the Devils previous two seasons, Will Butcher is actually both of the top 2 impacts.
At no point has Butcher’s impact seemed in doubt. I called him underutilized at the end of his rookie season. He got the extra utilization ultimately, and I assured that he’d done well with it. And then in the offseason, I advocated for giving him a 6-year deal. Did he do particularly well with Ben Lovejoy? Yes — they were one of the best pairings over the two years they played together. But according to MoneyPuck’s line tool he was also positive with Carrick, Vatanen, and Severson. The only person he couldn’t drag into the black was Steve Santini — but I don’t think Hafþór Júlíus Björnsson could’ve dragged Steve Santini anywhere so that’s not on Butcher.
What’s happened this season? I have two theories. Theory one basically boils down to this: Matt Tennyson just awful. Now, the RAPMs and GAR methods haven’t figured this out yet, but I think they’re probably punishing Will Butcher for their time together more than they should. The way RAPM models work, they will attempt to distribute credit in a way that maximizes the fit to the results of the shifts this season. But if the model “knew” how good Butcher had been before this season, I think it would look much less graciously on Tennyson. They’re not public yet due to small sample, but if we were able to view multi-year RAPMs heading into this year, I bet that model would do a better job of recognizing the guilty party. The other possibility is injury. Butcher went on IR for a little while after blocking a shot in the game against the Rangers. According to NaturalStatTrick’s game logs Before that game, his xGF+/- Rel was +0.08 whereas it’s been -1.35 since.
It’s worth noting that these two reasons may not be unrelated — upon returning from injury, his most frequent partner was Matt Tennyson. It remains to be seen if being apart from Tennyson will cure his ills — he’s only been positive with Severson so far — but if he’s healthy he should regress upward towards his career norms — the prior was too strongly established for this to be the real him in my opinion.
I think if you want the full story about just how much Andy Greene has improved, you should look no further than John’s piece.
I think that people started to wonder what Andy had left in the tank starting back with Greene-Lovejoy was a thing. I believed then, and believe now, that this was a premature assessment. However, last year, the decline seemed real. By any metric, Greene was not an effective back. After 6 consecutive years of being negative (good) or neutral in xGA60 Rel, last year he had the highest (worst) result of his career by FAR. Then this year he’s better than he’s every been in the metric — though, I’d say, suspiciously and unsustainably so.
It’s also not just a product of his partner. Though he and Sami Vatanen were great together (+13 xGF% Rel), he’s been a heavy positive with his two other main partners as well — Subban (+15.3%) and Severson (+13.7%).
The only real reason I can think of for this uptick is that he’s finally being used appropriately, which is to say that he’s always being given a puck-mover on the other side (as opposed to Santini/Lovejoy), and he’s no longer consistently and historically buried in his own zone — he’s taking far fewer defensive zone starts per 60 than any time in his Hynes career. The NHL-median in DZS/60 for a defender is around 8.34 and in the first 4 years under Hynes his DZS/60 stats were and 12.56, 13.02, 11.92, 11.04 — this was the heaviest DZ usage in the NHL. This year, he’s being given a cool 8.79.
While I do think that he’s going to come crashing down to earth, because these numbers are simply unsustainable, it’s also possible that he may have another couple years if his workload is lightened. For a 37-year-old that’s pretty remarkable.
I don’t think there’s any one feature to explain why such weird things are happening to the Devils analytically at the moment. Some is because Hynes is clearly still experimenting with his new toys until he finds something that works — and those things had worked for some better than others. As you can see from my explanations above, it’s possible a fair amount of this is all because of interesting player-specific storylines regarding injuries, recoveries, linemates, and usage. And, the elephant in the room of explanations, is that it’s still early in the season. This is hockey. Weird things happen. Sometimes they don’t even level out in the course of a season, let alone a dozen games. I wouldn’t dwell on these things, but they’ll be interesting statistical storylines to follow moving forward.
What do you guys think are explanations for the bonkers goings-on that we’re observing. Is there something I missed? Do you think these things will revert to the expectations, or continue as they are right now? Thanks for reading and leave your thoughts below!