Prop LabSpeed, faceoffs and lines
NHL DFS Stacking Works. “Chemistry” Is Mostly Noise.
We measured how every pair of NHL teammates scored together across nine seasons. Stacking linemates is supported by the data for a simple, structural reason. The idea that two specific players have a special connection barely survives a second season.
By Edgehalla ResearchPublished 6 min read
Why stacking is a thing
In daily fantasy (DFS), a stack is a group of players from the same team, usually the same line. If the line has a big night, every piece of the stack scores. You need that kind of upside to win a large contest.
Stacking only works if the players’ points move together. So we measured it. For every pair of NHL teammates with at least 50 minutes together at 5-on-5, in every season since 2017-18, we computed the game-by-game correlation of their DraftKings points.
How much do teammates’ DraftKings points move together?
Typical game-by-game correlation by pair type
Forward linemates correlate about 0.27 to 0.30; non-linemate teammates only 0.02 to 0.03.
Forward linemates sit at .27–.30 in every season we checked. That is about ten times the baseline for teammates who do not share a line. The link is real and stable.
The reason is boring. Hockey goals usually come with one or two assists, and the people most likely to get them are the scorer’s linemates. Power-play units add to it. When the line scores, several stack members get paid at once.
Is there special chemistry on top of that?
The popular idea goes further. Some pairs supposedly “just click,” so their points move together even more than normal linemates. If that were true, a pair’s extra correlation would show up again the next season.
We tested it across eight season-to-season comparisons. For each pair, we took its correlation minus the typical value for its type, then checked whether that gap repeated the following season.
Does a pair’s “extra chemistry” repeat?
Year-over-year correlation of a pair’s DK correlation minus its pair-type average
A pair’s extra correlation barely repeats: r 0.05 to 0.12 depending on pair type.
It barely does. The year-over-year correlation was .07 for forward pairs, .05 for forward–defense pairs and .12 for defense pairs. Even the forward pairs with the highest chemistry scores (top quarter) correlated only .30 the next season, against .26 for the bottom quarter.
The cleanest test is prediction. We tried to forecast each pair’s correlation next season three ways: its own raw number, a shrunk version of it, or just the league average for its pair type.
Forecasting next season’s pair correlation
Mean squared error (lower is better)
The pair’s own raw number was the worst forecast; the league average for its type did as well as anything.
- Pair’s raw number
- Shrunk estimate
- Pair-type average
Shrinking a pair’s raw number cut the forecast error by about 45%. And the shrunk estimate was no better than ignoring the pair entirely and using the average for its type. In plain terms: knowing which two players they are adds almost nothing.
What about playing better together?
Fantasy correlation is one question. Whether a pair creates more chances together than apart is another. Here the signal is a bit stronger, though still modest.
A pair’s shrunk “with you vs without you” lift in expected goals for per 60 minutes repeated year to year at r .14–.28. Its lift in expected-goals share repeated at r .08–.21. So some pairs do drive more offence together, but the effect is small and noisy. Lines also change often, which shortens how long any of it is useful.
How to build NHL DFS stacks
- Start from the line. Confirm who is actually together tonight on line combinations today. A broken line breaks the correlation.
- Stack forwards who both score. The value of a stack is shared upside, so combined scoring matters more than any pair’s past correlation. The top 2025-26 stacks by that measure were MacKinnon + Necas, Draisaitl + McDavid and Kucherov + Hagel.
- Add a defenseman for the power play, not for even strength. Forward–defense correlation is only .07–.10 overall, so a shared power-play unit is the usual reason to add one.
- Do not pay up for “chemistry.” A pair that ran hot together last season is not more likely to do it again than a normal pair of linemates.
Related: the faceoff matchups article covers the other team-play “matchup” people overrate, and game script myths covers garbage-time scoring.
Frequently asked questions
Should you stack linemates in NHL DFS?
Yes. Forward linemates’ DraftKings points correlated .27–.30 in every season from 2017-18 to 2025-26, against .02–.03 for teammates who are not linemates. When one scores, the other often gets an assist.
Do some NHL linemates have special chemistry?
Not in a way you can bet on. A pair’s correlation beyond the normal linemate level repeated the next season at only r .05–.12, and forecasting from the pair’s own history was no better than using the league average for its pair type.
What is a normal DraftKings correlation between NHL linemates?
About .27 to .30 for two forwards on the same line. Forward–defense pairs run .07–.10 and defense partners .05–.12. Teammates who do not share a line run .02–.03.
Should you stack a defenseman with forwards in NHL DFS?
Only with a reason, usually a shared power-play unit. At even strength, forward–defense pairs correlate .07–.10, much less than forward linemates at .27–.30.
Do some NHL pairs create more chances together?
A little. A pair’s shrunk with-vs-without lift in expected goals for per 60 minutes repeated year to year at r .14–.28. That is a real but modest signal, and line changes limit how long it lasts.
More in Speed, faceoffs and lines
Sources and method
- NHL stats (NHL.com)
- NHL API reference (community documentation)
- Empirical Bayes shrinkage
- Responsible Gambling Council (Canada)
- ConnexOntario (1-866-531-2600)
- National Council on Problem Gambling (1-800-GAMBLER)
Research and analysis, not betting advice. Past results do not guarantee future returns; bet only what you can afford to lose.