How much is a missing NHL star worth?

A star scratched at the last minute moves the betting line. Our Elo model does not move. We measured how much a missing player has actually been worth.

Research as of Nov 20, 2026.

The short answer. In our 2021-26 data a usual regular missing is worth about 2 points of win chance, a player at our star cutoff about 4, and the very biggest stars (Connor McDavid, about 8.6) 6 to 9. Those sizes come from the same seasons we fit them on. When we tested them on two seasons we had never used (2018-19 and 2019-20), we could not detect any improvement. So the honest summary is: lineups probably matter a little, we have not proven how much, and late scratches and some starting goalies are only confirmed near puck drop.

1. The Makar night

On Nov 19, 2026, Colorado's defenceman Cale Makar was out. The betting line on Colorado moved from -154 to -130 during the day. Our Elo did not move at all: it still had Colorado at 65.7%, because it is built from results and has no idea who is in the lineup. The same night Ottawa's line moved from -189 to -150 on a backup goalie and two depth scratches.

That is one night, an anecdote and not data. It does raise the question this article answers with data: how many points of win chance is a missing player actually worth? For the wider comparison of lines and our model, see why NHL betting lines move and our Elo model doesn't.

2. A usual regular out: about 2 points

First study (Nov 11). A “usual regular” is a team's top six forwards or top four defencemen by ice time this season. Over 2023-24 to 2025-26, when one was missing from the lineup, a team missing one did worse than our Elo expected, by about 2 points of win chance per regular out. For comparison, we found about 5 points per starter out in the NBA. Hockey spreads its minutes over 18 skaters, so one missing player matters less.

How often does it happen? In those seasons a team had no regular out in about 43% of games, one out in about 37%, two in about 15%, and three or more in about 5%.

The effect fits in-sample, and across five seasons (2021-26) it helped in four when each was predicted from the others. But it made one season (2025-26) worse, and no single pre-planned test cleared the bar. Our verdict was “underpowered”: probably a small real effect, not proven.

3. A star out: about 4 points at the cutoff, 6 to 9 for the biggest

A fair objection: a 4th-line scratch and a superstar are not the same. So the second study (Nov 19) weighted each absent player by how many points he produces above a call-up or depth player. In the 2021-26 data that put a player right at our star cutoff (the top 5% of regulars by value) at about 27 Elo points (about 4 points of win chance at a coin flip), and a plain regular at about 13.5 Elo (about 2 points).

Player type (2021-26, in-sample)Elo pointsWin chance
Plain regular~13.5~2 points
Player at our star cutoff (top 5% by value)~27~4 points
Connor McDavid60~8.6 points

McDavid (60 Elo), Kucherov (57) and MacKinnon (52) are among the five largest values in those seasons, so they are the top of the scale, not typical stars: 6 to 9 points of win chance for the very biggest. A star was missing in roughly 4 to 9% of team-games.

One test of the headline case: favourites missing a star while the underdog was at full strength won 60.7% of 349 games in the fit seasons against 63.6% our Elo stated. In the untouched seasons the same group won 66.0% of 97 against 61.7% stated, the opposite direction. Both samples are small, and together they say nothing clear.

4. The honest test: two seasons we had never used

Everything above is in-sample. The real test was pre-registered: fit the size on 2021-26, then see whether adding it to Elo improves predictions on 2018-19 and 2019-20 (2,182 games). It did not. The change in accuracy was essentially zero (log loss +0.00006, margin of error about ±0.002): nothing detectable. The two test seasons also pulled in opposite directions.

That is not proof that absences do not matter. The margin of error still includes the in-sample effect, and those two seasons are noisier for Elo. It means the effect was not big enough to show up there.

Two more things we checked. First, weighting by player value did not beat the plain count of regulars out (the count was slightly better on the untouched seasons). So a clever formula did not buy us anything. Second, absences that were just a few games old looked stronger than long-term ones, which fits the idea that Elo adapts to a long injury and misses a fresh one. It is a hypothesis: the evidence came from the same seasons the idea came from.

5. When are NHL lineups actually known?

All of these numbers use who actually dressed, which is an upper bound for any live version. In real life NHL lineups are mostly confirmed close to puck drop: morning skates, warm-ups and late scratches. Goalies are often announced the morning of, sometimes later. Lines can react to a late scratch in minutes; a model that is built from scores cannot, and does not claim to.

6. What we do about it

Nothing, for now. Our Elo is frozen and does not see lineups, because we found no pre-registered test where adding them clearly helped. We show the NHL injury report and the confirmed goalies on the starting goalies page so you can see who is in. We are not running a separate test of the fresh-absence idea: on the untouched seasons it did no better than simply counting regulars out. Our comparison with closing lines (linked above) is where absences are tracked now. Related: a backup goalie starting cost about 1.5 points on our fairest measure, not clearly different from zero (see what actually predicts NHL games), and the record of the model itself is on NHL moneyline and puck line.

None of this is a reason to back or fade a team. It is a measurement of how much a missing player has historically moved results.

Frequently asked questions

How much does a missing star change an NHL game?

In our 2021-26 data, about 4 points of win chance for a player at our star cutoff and 6 to 9 points for the very biggest names, with Connor McDavid at about 8.6. These are in-sample sizes. Two untouched seasons, 2018-19 and 2019-20, did not confirm them.

Does a missing regular matter as much in hockey as in basketball?

Less. Each usual top-six forward or top-four defenceman who sat was worth about 2 points of win chance in our data, against about 5 points for an NBA starter. A hockey team spreads ice time over 18 skaters, so no single player carries as much.

Why does your NHL model not adjust for injuries?

We tested it. Weighting each absence by player value did not beat simply counting regulars out, and neither made the predictions measurably better on two seasons we had never used. That means no detectable gain, not that absences do not matter. Late scratches and some goalies are also confirmed only near puck drop.

The week ahead, once a week

Last week's graded record — wins and losses — and the plays for the week ahead. One email, no daily noise, unsubscribe in a click.

Educational content only. Research results are historical, from our own data, and describe what happened, not what will happen. Must be 21+ and in a jurisdiction where sports betting is legal. If gambling is a problem for you or someone you know, call 1-800-GAMBLER or visit ncpgambling.org.