A reserve forward plays eleven minutes, misses both his shots, and finishes the night as the best number on the sheet. Every recap in the building notices. By the second paragraph he is the reason the bench won the game, and the sentence has a number attached to it, so nobody questions it.
The number is real. The reading is not. He was on the floor for one stretch in the second quarter when the starters on the other side were resting, and for four minutes at the end of a game that had stopped being competitive. Both stretches went well. Neither had much to do with him.
This is the single most frequently misused figure in a post-game write-up, and the reason is that it looks like an individual statistic and is not one. It is a team statistic with one player’s name on it, and over a single evening the team part swamps the player part almost completely.
What to check before quoting the number
- Minutes. Under fifteen and the figure is mostly noise. Under ten and it is close to meaningless.
- Who else was out there. Five names produce the number. One of them gets the credit.
- When those minutes happened. A margin built after the result was settled counts the same as one built in the third quarter.
- The opponents on the floor. Bench-heavy stretches inflate reserves and deflate starters, in both directions and by a lot.
- The shooting variance inside the stint. Two opponent three-pointers in ninety seconds move the figure more than anything the player did.
- Whether a better number exists. If lineup data is available, use it. If it is not, say less.
What plus-minus in basketball actually counts
The definition is trivial and the trouble starts immediately after it. The figure is the team’s scoring margin during the time a player was on the court. Nothing else enters it. Not shots, not rebounds, not defensive assignments, not whether the player touched the ball.
That construction has one great virtue: it captures contributions no other box score line records. A player who sets legal screens, occupies two defenders, communicates a switch and never appears in a highlight will show up here and nowhere else. Over a long enough run, that is genuinely valuable information, and it is why the figure exists.
Its defect is the same property viewed from the other side. Because it records everything that happened while a player stood on the floor, it records four other players, the five opponents, the referees, the shot luck and the score situation. Over a season those inputs partly cancel. Over one evening they do not cancel at all.
The error, stated plainly
The mistake is not using the figure. It is treating a single game’s version of it as an evaluation of a person. Almost every misuse in a recap reduces to that one substitution, and it usually happens in a sentence that begins with the words the team was.
Consider the arithmetic on its own terms. In a stint of ten minutes, a professional team will run through perhaps twenty possessions at each end. A run of made and missed shots inside forty possessions is a small sample by any standard, and the swing produced by two or three of them is larger than any plausible individual effect over that span.
So the figure moves for reasons that have nothing to do with the named player, and it moves a great deal. Nothing about this is controversial among people who build these numbers. It becomes controversial only when it reaches print, where a differential is read as a verdict.
Lineup context: you are measuring five people
The clearest way to see the problem is to notice that in any given stint, five players share one number. If the starting five outscore the opponent by eight during a six-minute run, all five carry the same plus eight, and the write-up then chooses which of them to credit based on what it already believed.
That choice is where most bad analysis enters. A player who shares his minutes almost entirely with the best two players on the roster inherits their results. A player whose minutes are deliberately staggered so he anchors bench units inherits theirs. Neither pattern is anything he decided.
Coaching staffs know this, which is why front offices ask a different question when they are deciding what a roster is actually worth. They want to know how the team performed with a player and without him across many games, with the partners held roughly constant. That question is answerable. The one-game version of it is not.
Garbage time is a different game on the same scoreboard
Once a margin passes the point where the trailing team stops competing normally, the sport changes. Defensive intensity drops, rotations stop mattering, and both benches play. The scoreboard keeps working the same way, and the differential keeps accumulating as if nothing had changed.
This produces two systematic distortions in the same box score. Reserves who only play in that period collect margin that was available to anybody standing there. And starters who were pulled early are protected from a stretch that would have gone against them, or denied one that would have gone their way.
There is no agreed threshold for where this begins, which is part of the difficulty. Public datasets use different rules, some based on margin and time remaining, others on which players are on the floor. A recap does not need a formal definition. It needs to look at the clock and the margin during the stint before quoting the total, which takes about fifteen seconds.
Minutes are the denominator nobody checks
A differential quoted without minutes is close to unreadable, and it is quoted without minutes constantly. Plus twelve in thirty-four minutes and plus twelve in nine minutes are not comparable quantities, and the second one is far more likely to be an accident.
Per-minute versions exist and they help a little, but they introduce their own trap. Dividing a noisy number by a small number produces a larger, noisier number that looks more precise. A reserve with seven minutes and a plus six becomes a spectacular rate figure, and the rate figure is worse evidence than the raw one, not better.
The practical rule most analysts settle on is a floor rather than a formula. Below a certain number of minutes the figure gets reported as context or not at all. Where exactly that floor sits is a judgement call, but the habit of checking before quoting is the part that matters.
Substitution patterns are a decision, not a measurement
Rotations are built around matchups, foul trouble, rest and the shape of the schedule. That means a player’s minutes are assigned rather than random, and a differential built out of assigned minutes inherits the assignment.
Two examples are enough. A defensive specialist sent out specifically to guard the opponent’s best scorer will spend most of his night in the hardest possible circumstances and will look worse for it. A backup who plays only against second units will look better than he is, every single night, and the effect compounds across a season rather than washing out.
Foul trouble adds a further layer that nobody controls. A starter who picks up two early fouls sits through a stretch he would ordinarily have played, and whatever happens during that stretch attaches to somebody else. The number is describing a coaching decision at least as much as it describes anybody’s play.
On/off, net rating and adjusted versions: what each fixes
Several better instruments exist and it is worth knowing what each of them repairs, because none of them repairs everything and a recap that names the wrong one is no more accurate than one that names none.
Net rating converts margin into points per hundred possessions, which removes the pace problem. A game with many possessions produces larger raw margins for everybody, and net rating stops that from being mistaken for quality. It does not touch the lineup problem at all.
On/off differentials compare team performance with a player on the floor against the same team without him, which addresses part of the context question. Adjusted and regularised versions go further, using many games to estimate each player’s contribution while accounting for teammates and opponents, and they come with wide uncertainty ranges that are usually left out when the numbers get quoted. All of them need substantial samples, which is the recurring theme in this whole area of measurement.
What survives one game and what does not
| Figure | Reliable after one game | Main distortion | Safe use in a recap |
|---|---|---|---|
| Raw plus-minus | No | Lineup context and garbage time | Only with minutes and stint timing attached |
| Net rating | Rarely | Same as above, pace removed | Team level, not individual |
| On/off differential | No | Tiny off-court sample in one game | Leave it out entirely |
| Points, rebounds, assists | Yes, as description | Says nothing about efficiency | Report as what happened |
| True shooting percentage | Partly | Small attempt counts | With attempts stated |
| Turnovers and fouls | Yes | Referee variation | Straightforward counts |
The pattern in that table is consistent. Counting statistics describe an evening honestly because they are descriptions. Differential statistics attempt an inference, and inference needs a sample that one night does not provide.
A reading order for a single box score
- Start with the score and the shape of the game. When did the margin move, and did it ever come back? Everything else gets interpreted against that.
- Read minutes before anything else in the player rows. Minutes tell you who the coach trusted, which is a real signal that requires no adjustment.
- Read shooting with attempts attached. Nine points on six shots and nine on eighteen are opposite evenings.
- Read turnovers against usage. A high-usage player with four turnovers is a different story from a low-usage player with the same four.
- Only then look at the differential, and look at it as a question rather than an answer. If it is strange, find out which stint produced it.
- Check the stint against the clock. If the margin was built after the game was decided, the figure has already told you what it is worth.
Three sentences to cut from a recap
The first is the bare quotation: a name, a differential and nothing else. It reads as evidence and contains none, and it is the version that circulates furthest because it fits in a single line.
The second is the causal upgrade, where a positive figure becomes the reason for a result. The team outscored the opponent while he played is a description; he changed the game is a claim about mechanism, and the differential cannot support a mechanism.
The third is the comparison between two players on opposite teams, which stacks every problem at once. Different teammates, different opponents, different minutes, different score situations. That sentence appears in an enormous share of the arguments that dominate a week, and it is usually the weakest sentence in the argument.
The honest version of the same paragraph
None of this means the figure should be thrown out. It means it should be reported with the two facts that make it readable, and framed as a question rather than a conclusion.
A workable sentence names the stint rather than the total: the margin moved during a particular stretch, these five were on the floor, and here is what changed on the court during it. That is checkable, it is more interesting than a number, and it survives someone rewatching the game.
Where the whole roster is affected by an unusual environment, the same discipline applies with more force. Seasons played in altered conditions, of the sort covered in the reporting on relocated and compressed schedules, distort margins across every lineup at once, and single-game differentials from those stretches deserve the least confidence of all.
Frequently Asked Questions
How many games does the figure need to be useful?
More than most people assume, and the honest answer is that it depends on how the player’s minutes are distributed. A player with stable partners and consistent opponents stabilises faster than one whose role moves around. The practical guidance is that a single game is never enough and that a few games are not much better.
Is the figure worthless in hockey and football too?
The problems are the same wherever the statistic is used, and in low-scoring sports they are worse, because a small number of goals produces enormous swings over a shift or a spell. The hockey version has been studied for a long time and the conclusion is the same: it needs volume and adjustment.
What about a very large single-game number?
A large figure usually means a stint went unusually well or badly, which is worth investigating and is not worth quoting on its own. Extreme values are more likely to be produced by variance than by performance, which is the opposite of the instinct most readers have.
Do coaches look at it after a game?
Staffs generally look at lineup-level data rather than individual differentials, and they look at it alongside film rather than instead of it. The question they are answering is which combinations worked, which is what the underlying data supports.
Is there a quick way to sanity-check one before quoting it?
Find the two longest stints the player had and check the score and the clock during each. If either falls inside a decided game, the total is already compromised. If both are competitive and short, the number is still noisy, and the honest move is to describe the stretch rather than cite the figure.
The next time a differential appears in a recap, look for the two numbers that should be standing next to it. If the minutes and the timing of the stint are not there, the sentence has not been checked, and there is a good chance the writer would not want it to be.
