Box score to plus-minus: what each generation of metric added

What the box score records

A traditional box score counts discrete events attributable to an individual, which makes it verifiable, reproducible and easy to argue from. It captures scoring, rebounding and passing that led directly to a score, and it misses almost everything that has no discrete owner. Defensive contribution is particularly poorly served, since preventing an attempt from being taken produces no recorded event at all.

Spacing, screening and off-ball movement create value for others and appear in the record only through somebody else line. A ranking built on the box score therefore systematically favours roles whose contribution happens to be countable. That favouritism is invisible in the numbers themselves, which is precisely what makes it so durable in public argument about who mattered most.

Efficiency and the pace correction

Raw totals depend on how many possessions a side plays, which varies enormously between teams and between eras of tactical fashion. Dividing by possessions produces per-possession figures that are comparable across those differences and reveals that some large totals were mostly opportunity. Shooting efficiency measures that weight attempts by their expected return correct a similar problem, since not all attempts are worth the same.

These adjustments were a genuine advance, because they separated volume from quality without requiring any new data collection. They did not address the deeper problem, which is that the box score cannot see contributions that produce no event.

Plus-minus and the noise it carries

Measuring the difference in outcome while a competitor is present captures everything they contribute, including the parts nothing else records. The cost is that it also captures everything their teammates and opponents contribute during the same stretch, which is a great deal of noise. Separating an individual effect requires many possessions in varied combinations, and rotation patterns supply far less variation than the method needs.

Regularised versions pull estimates towards the average to control that noise, which stabilises the numbers and introduces a dependence on prior assumptions. The result is a measure that is conceptually right and practically uncertain, which is the opposite failure mode from the box score.

Tracking data and the behavioural turn

Position tracking records where everybody was at every moment, which allows behaviour to be described rather than inferred from outcomes. It makes previously invisible contributions measurable, including the space created by a threat and the routes closed off by good positioning. The difficulty is that behaviour is not value, and converting a description of movement into a contribution requires a model with its own assumptions.

Those models are usually proprietary, which means public rankings built on them cannot be checked by the people reading them. Verifiability was the great strength of the box score, and each advance in sophistication has traded some of it away.

Using the generations together

The measures disagree, and the pattern of disagreement is informative because each is blind to something the others can see. A competitor rated highly by impact measures and modestly by box score measures is contributing through channels the count cannot capture. The reverse pattern suggests accumulation in a favourable role without a corresponding effect on what actually happens while they play.

Neither pattern settles a ranking, and both narrow the explanation to something specific enough to investigate. Reporting several measures with their disagreements visible is more honest than selecting whichever supports a preferred conclusion. It also gives the reader a way to notice when a confident ranking depends entirely on one measure that happens to flatter its subject.