Plus-minus is the worst widely used stat in hockey
Plus-minus survives on the back of two things. The first is that the math behind it sounds reasonable — count the goals scored for your team while you were on the ice, subtract the goals scored against, end up with a single number that's supposed to summarize whether you were a net positive when you played. The second is that the stat has been on hockey cards since the 1960s, and stats that have been on hockey cards for sixty years don't disappear quickly.
The math doesn't actually work. A single season of plus-minus has so much noise relative to signal that you can confidently identify only the very best and very worst players on a roster, and even those identifications are mostly driven by who their linemates were. As a verdict on individual skill, the column is almost meaningless. As a verdict on which line a player belonged to, it's only slightly more informative. As a way to compare players across teams, it actively misleads.
What the stat ignores
Plus-minus credits a player for being on the ice when a goal is scored. It does not credit them for what they did to cause it, what they did to prevent it, or what else was happening on the ice. A fourth-line winger who was on the ice for an empty-net goal his team scored gets a +1. The same winger getting hemmed in for forty seconds before his team got the clear and scored against the run of play also gets a +1. Neither situation tells you anything about the winger's skill. Both produce the same column entry.
The stat also doesn't account for ice time. A player who plays twenty-five minutes a night against top opposition will accumulate plus-minus events at a much higher rate than a fourth-liner getting eight minutes a night, even if their per- minute impact is identical. The raw column flatters high-minute players and penalizes them in equal measure depending on team context. You cannot read a player's plus-minus without also reading their team's win-loss record and their ice time, at which point the stat has reduced to "this is a player on a good team who played a lot."
Why the linemate problem is fatal
The deepest issue is that plus-minus is fundamentally a team stat masquerading as an individual one. The five skaters on the ice for any goal all get credited with the same plus or minus, regardless of which one was actually responsible. If you skate next to a top scorer, you get his pluses. If you skate next to a defensive liability, you eat his minuses. Two players with identical actual skill levels can have wildly different plus-minus numbers depending on which linemates they spent the most time with.
This is not a small effect. Public analyses of plus-minus across full seasons consistently find that linemate quality explains as much or more of the variance in individual plus-minus as the player's own contribution does. You can manipulate any player's plus-minus number by 15 to 25 points across a season just by moving them up or down the depth chart. The number is responsive to the lineup card more than to anything the player does.
The empty-net problem
The single most absurd feature of plus-minus is the empty-net adjustment, which does not exist. Empty-net goals — scored when the trailing team has pulled their goalie for an extra attacker — count the same as any other even-strength goal in the column. This means a player can boost their plus-minus substantially over a season just by being on the ice for empty-net goals scored by their team. Players on good teams that lead games more often accumulate more empty-net pluses; players on bad teams that trail more often eat more empty-net minuses.
The size of this effect varies year to year, but for the league's better teams it can account for 5 to 10 plus-points per top-line forward across a season. Stripping out the empty-net inflations doesn't fix the stat, but it does change the rank order at the top of the league materially. Most public plus-minus reports don't strip them out.
What to use instead
The honest answer is that no single number cleanly summarizes individual hockey impact. The best public attempts are the family of on-ice rate stats — expected goals percentage, Corsi-for percentage, and the weighted-shot metrics that have replaced raw plus-minus in front offices. These stats normalize for ice time, separate offense and defense, and adjust for the strength of the on-ice situation.
Expected goals percentage, on a per-player basis, is roughly what fraction of the expected-goals total of a player's on-ice minutes were generated by their team versus their opponent. A player whose on-ice xG% sits at 55% or higher was on the ice for more high-quality chances than they gave up, normalized for sample size. League average is 50%. Anything above 55% across a full season is meaningful. The stat is not perfect — it inherits the linemate-quality problem — but it is much more informative than raw plus-minus because it's measured per minute and over many more events.
For the more advanced public sources, "relative to teammates" adjustments try to partly fix the linemate problem by comparing a player's on-ice rate stats to their team's rate stats when they're off the ice. A player with a 55% xG% on a team that runs 52% off-ice is doing better than expected; a player with the same 55% on a team that runs 58% off-ice is actually a drag. The "relative" adjustment captures something plus-minus cannot.
The professional use of plus-minus
Front offices have not stopped looking at the stat entirely; they look at adjusted versions of it. RAPM — regularized adjusted plus-minus — is the basketball-style attempt to actually solve the linemate problem by running a regression across many seasons and many lineups. The output is a number that estimates each player's true individual impact while accounting for every teammate and opponent they played with. RAPM-style metrics for hockey exist in front offices and in a few public datasets. They produce a meaningfully different rank order from raw plus-minus, and they have predictive value that raw plus-minus does not.
The point is that the math underneath plus-minus, properly regularized, is salvageable. It's the raw column on the broadcast graphic that's broken. The version that takes a full year to compute, requires real statistical machinery, and produces a result no fan would recognize is the version the professionals actually use.
The takeaway
When you see a player's plus-minus, especially over a single season, treat it as a starting point for a question, not an answer. If a player has +30 across a season, ask which line he played on — chances are he was on the top line of a good team and ate the pluses his linemates generated. If a player has -20, ask the same question in reverse. Then look at his on-ice xG percentage and his relative-to-teammate numbers, both of which are publicly available and both of which will give you a much sharper read on whether the player is actually any good.
Plus-minus is to hockey what wins are to pitching: an old, simple, intuitive number that the sport hasn't quite retired despite knowing for decades that it shouldn't be used as a primary verdict. The retirement is coming, the same way it came for batting average and passer rating. It's just going to take another decade or two for the broadcast graphics to catch up.