Where does Trevor Zegras belong? Part 3: the answer is that there isn't one
The performance split, pre-registered before it was run — and the arithmetic showing that 60% of the NHL's forwards fit inside our margin of error. One season cannot answer this, for anyone.
Part 1 worked out which games Trevor Zegras played centre and which he played wing. Part 2 showed the job barely changes between the two — same ice time, same teammates, same opponents, same situations.
This is the part everyone actually wanted: is he better at one?
The answer is that we can't tell, and — more usefully — nobody can, from one season. That isn't a dodge. It's a number, and it's the most interesting thing in this whole series.
We wrote down the answer conditions first
Role changes are a response to how a player is going. Zegras's own season proves it: he was promoted to centre in January and demoted back within a month. So anyone who goes looking for a centre-versus-wing performance gap, with six metrics and a free hand, will find one.
So the analysis was pre-registered before the comparison was run. The script has two modes: one computes the noise using shuffled role labels and never touches the real split; the other runs the actual test. Only the first had been run when the decision rules were written down — which metric counts as primary, what a null would be allowed to mean, and what would happen if something came back significant.
The noise model matters too. His roles come in long blocks — wing from mid-October to late January, centre from March onward. Shuffling individual games would shatter those blocks and invent a world far quieter than the real one, so instead the blocks themselves get shuffled: same spells, same lengths, same number of games in each role, reordered across the season. Ten thousand times.
The finding: the margin of error is wider than the league
Running that shuffle answers a question you can ask before you know the result: how big would a difference have to be before we could tell it apart from luck?
For on-ice expected-goals share — the primary metric — the answer is 7.9 percentage points. Here is what that looks like next to every NHL forward who played 300 minutes at even strength last season:
To register as a result, moving Zegras from wing to centre would have to change his on-ice share by more than the gap between a fourth-liner and a first-liner. Position changes don't do that. Nothing does that. The study was incapable of finding a believable effect before it was run.
We checked whether measuring his own shooting instead of his line's results would help, since individual numbers carry less of other people's noise. It's worse — a forward takes few enough shots that his personal rate is even shakier over 30 games. The best of six metrics still needed an effect the size of a 1.3 standard-deviation jump in the league's talent distribution.
The result, since we promised to publish it
The primary metric came back flat — on-ice xGF% +1.5 points at centre, p = 0.74. Sitting comfortably in the middle of the noise. Under the registered rules that is the verdict, and the verdict is "we could not detect a difference" — which is not the same sentence as "there is no difference", and the chart above is why.
The trap we wrote down in advance, springing on schedule
One metric did come back significant: on-ice expected goals for per 60 fell by 0.53 at centre, p = 0.027. In a less careful post, that's the headline. "Zegras generates fewer chances at centre."
Three reasons it isn't:
- Six metrics were tested. Even if absolutely nothing were going on, there's a 26% chance of at least one landing under 0.05 — better than one in four. Getting exactly one hit is the single most likely outcome of a null world. Correcting for six looks, the bar moves to 0.008, and 0.027 doesn't clear it.
- It was pre-registered as non-verdict-bearing. The primary metric was named in advance precisely so a secondary couldn't be promoted after the fact. This is that rule doing its job.
- Both ends moved together. Chances for fell 0.53 and chances against fell 0.63. That isn't playing better or worse — it's playing slower. Fewer events at both ends, which is why the share barely moved. And the most likely cause is sitting in Part 2: his linemates changed completely. On the wing he played three quarters of his minutes with Christian Dvorak; at centre, with Owen Tippett. Different linemates, different pace.
This is the entire argument for writing your rules down first. With the same data and no pre-registration, "Zegras suppresses chances at centre (p = 0.027)" is a defensible-looking sentence. It is also almost certainly wrong, and we'd have had no principled way to talk ourselves out of it after the fact.
How much would be enough
The noise shrinks with the square root of the games, which is a punishing exchange rate:
- Adding his 2024-25 season in Anaheim — where he also played both roles, split by an injury — buys about 57 more games and takes the best metric from 1.33 to roughly 1.02 standard deviations. Still not enough.
- Getting to a genuinely useful resolution needs roughly seven times the games: around 570 split between the two roles. That's not a season. That's most of a career of being moved back and forth.
So this isn't a Zegras problem. Single-season, single-player role splits are not answerable — not by us, and not by anyone working from the same public event data. If you see a confident take on where a player is better based on one year, the take is noise wearing a number.
Where that leaves the question
Three posts in, here's the honest ledger. We know exactly which games he played each role, published game by game. We know the job barely changes between them, apart from who stands next to him. And we know that whether he's better at one is beyond what a season can show — with the arithmetic to say how far beyond.
The eye test isn't going to be settled by this data set. What it can do is tell you when somebody's numbers aren't settling it either.
The per-game table from Part 1 is still open for corrections, and the offer stands: if you watched one of the eight disputed games and we called it wrong, we'd like to know.