How the NBA Measures Player Impact

Six Ways to Measure a Basketball Player
Using Plus-Minus:
One Walk Through.

Plus/minus started as a single number on a hockey scoresheet. Six fixes later, it is the foundation of how the NBA argues about who is best. Walk through every metric in chronological order, anchored in the actual 2024-25 numbers of Most Valuable Player Shai Gilgeous-Alexander.

For each of the six metrics, this page answers four questions in plain language and a visual: What does it measure? How is it calculated? What does the number mean? What does it do well? Built as a teaching example of how AI can turn a dense statistical concept into an interactive explainer.

Six metrics Anchored in Shai's Most Valuable Player season Updated May 2026
By Dr. Kevin Luczynski A behavioral scientist who watches NBA basketball (especially playoffs) and coaches youth basketball. universalbcs.com

A summary of what is coming below

Plus/minus started as one number on a hockey scoresheet: was your team winning while you were on the floor? Six refinements later, modern impact metrics try to isolate what each player actually contributed. The best public metrics still disagree about who deserved the 2024-25 Most Valuable Player award. This page walks through every refinement in chronological order, anchored throughout in Shai Gilgeous-Alexander's verified MVP season.

Words to know (for the casual NBA fan)
Possession
One team's turn with the ball — from getting it (off a rebound, after a made basket, or off an inbound) to scoring, missing, turning it over, or getting fouled. A typical NBA game has roughly 100 possessions per team. Pace is the count of possessions per 48 minutes.
Box score
The official statistical summary of a single game: points, rebounds, assists, steals, blocks, turnovers, fouls, minutes, shooting percentages. "Box-score stats" means numbers a reader can pull straight from this sheet, with no advanced math.
Per 100 possessions
A way of measuring efficiency that strips out pace. A team that scores 110 points in 100 possessions is playing equally well as a team that scores 121 in 110 possessions — both are 1.1 points per possession. Per-100 numbers let a slow team and a fast team be compared on the same scale.
Per 36 minutes
What a player would produce if scaled to 36 minutes of play (roughly a starter's typical workload). Lets the reader compare a starter who plays 35 minutes a night to a bench player who plays 20, on the same time-on-floor scale.
Stat sites cited here
Basketball-Reference is the encyclopedia of basketball statistics (every stat for every player since 1947, free). Cleaning the Glass is the same data with garbage time stripped out (mostly paid). Dunks & Threes publishes Estimated Plus-Minus, a modern impact metric that incorporates Regularized Adjusted Plus/Minus in its methodology. The Ringer is a sports website that covers the NBA editorially.
G-League
The NBA's developmental league — the minor-league system. A "G-League call-up" is the kind of replacement-level player a team would sign on a short contract to fill out the bench.
Metric 01

Raw Plus / Minus

In one sentence Was the team winning while the player was on the floor? That number is the player's plus/minus.
✓ Best at

Telling you exactly what the scoreboard did during a player's minutes. Simple, universal, no statistician needed.

✗ Still cannot do

Tell you whether the player actually contributed to those points or just stood there while teammates did the work.

The simplest version. Every point a team scores while a player is on the floor adds to that player's number. Every point the other team scores subtracts from it. Nothing else counts — not shot quality, not defense, not effort. Press the Play segment button below to begin.

Home
24
1st Quarter
2:00
Away
22
Play-by-Play
Press Play to begin.
Segment Start
Home 24 · Away 22
Segment Differential
—
Away · Visitors
Home
Metric 02

On / Off

In one sentence Is the team better when the player is on the court than when the player is on the bench? That gap is the on/off.
✓ Best at

Catching a player whose team falls apart without him, or a player whose team is fine either way. The first metric that asks "what changes when this guy sits?"

✗ Still cannot do

Separate the player from the lineup. The number depends entirely on who replaces him on the bench. Strong backup means small on/off; weak backup means huge on/off.

On/Off compares the team's performance when a player is on the court to its performance when that player is on the bench. It was the first attempt to fix raw plus/minus's presence-versus-contribution problem by accounting for the rest of the lineup. The fix is real. The new flaw: the answer depends entirely on who the player gets swapped out for.

When Shai Gilgeous-Alexander is on the court
2,599 minutes · 2024-25 regular season
2
Shai ★
+
Wing
+
Wing
+
Big
+
Big
Oklahoma City Offensive Rating 124.4 · Opponent Offensive Rating 107.5
Team net per 100 possessions +16.9
When Shai Gilgeous-Alexander is on the bench
1,345 minutes · Cason Wallace primary replacement
22
Wallace
+
Wing
+
Wing
+
Big
+
Big
Oklahoma City Offensive Rating 113.9 · Opponent Offensive Rating 108.2
Team net per 100 possessions +5.7
Shai Gilgeous-Alexander's 2024-25 On/Off Calculation
Team net WITH Shai +16.9
−
Team net WITHOUT Shai +5.7
=
Shai's On/Off +11.2
Oklahoma City scored 10.5 more points per 100 possessions and gave up 0.7 fewer when Shai Gilgeous-Alexander played. The team was still positive without him because his bench replacements (Wallace, Wiggins, Caruso) were genuinely good. Most stars on most rosters do not have that quality of replacement player. Source: Basketball-Reference on/off splits.
What On/Off does well

It accounts for the rest of the lineup. Two players with the same raw plus/minus can have very different On/Off numbers — one keeps the team afloat when he sits, the other does not. This is the first metric that tries to ask the right question: how much of this is the player versus the four teammates around him?

Try It Yourself · Build Your Own On/Off
Try this first
  1. Leave the WITH slider at +16.9 (Shai's actual on-court Net Rating).
  2. Drag the WITHOUT slider down toward zero, then back up past +10.
  3. Watch the differential below. When Shai's replacements are weak, his On/Off balloons. When they are strong, his On/Off shrinks toward zero — even though Shai himself did not change.
The Player's On/Off Differential
+11.2
−30−150+15+30

Drag either slider. The differential is simply WITH minus WITHOUT. Notice how a player's number depends entirely on how good the team is around him and how good his replacements are. Drop the OFF slider toward zero and his on/off inflates; raise it toward his ON value and his on/off shrinks. Same player, very different verdicts.

What Metric 02 proves · On / Off

On/Off compares a team's rating with the player on the floor to its rating with the player off the floor. It is the first metric that tried to factor out the floor itself, instead of just measuring the scoreboard around the player.

The new flaw: the answer depends on who replaces the player. Shai Gilgeous-Alexander's +11.2 measures the gap between Shai and Cason Wallace specifically, the guard who logged the most minutes in his place. If Shai's replacement were a much weaker player — someone barely good enough to hold an NBA roster spot — the gap (and Shai's On/Off number) would balloon to +18 or +20. If his replacement were already a star-level guard, the gap would shrink toward zero. The metric still cannot separate the player from his lineup; it has just moved the lineup problem one seat down the bench.

Metric 03

Net Rating

In one sentence Same point margin can mean different things if one team plays slow and another plays fast. Net Rating fixes that by counting per 100 possessions instead of per game.
✓ Best at

Comparing teams and players across different paces of play on a single, fair scale. The number ESPN, the NBA broadcast, and most podcasts quote when they say "best offense in the league."

✗ Still cannot do

Tell you who specifically did the work. A player's Net Rating is still mostly about which four teammates were on the court with him.

Pace can make raw plus/minus misleading. A team that is winning by 5 points (a plus/minus of +5) across 80 possessions is scoring more per trip down the floor than a team that is winning by 5 points across 115 possessions. Net Rating divides the point margin by the number of possessions and multiplies by 100, so every team and every player can be compared on the same per-possession scale.

Illustrative example · the canonical pace contrastThe two team panels below borrow the most famous pace contrast in modern NBA analytics: Pat Riley's mid-2000s Miami Heat (Shaquille O'Neal half-court system, slower-than-average pace by Riley's final two seasons) versus Mike D'Antoni's mid-2000s Phoenix Suns ("seven seconds or less," league-leading pace). The numbers below are round, chosen to make the pace effect visible at a glance, not exact season totals. The interactive calculator further down uses Oklahoma City's verified 2024-25 numbers (Offensive Rating 120.3, Defensive Rating 107.5, Pace 100.0).

Miami Heat (mid-2000s)
Slow pace · Pat Riley grind-it-out half-court
80 possessions
Scored92
Allowed87
Raw +/-+5
Phoenix Suns (mid-2000s)
Fast pace · Mike D'Antoni seven seconds or less
115 possessions
Scored124
Allowed119
Raw +/-+5
The Net Rating Calculation
Net Rating = (Points Scored − Points Allowed) ÷ Possessions × 100
Miami (92 − 87) ÷ 80 × 100 = +6.25 per 100 poss
Phoenix (124 − 119) ÷ 115 × 100 = +4.35 per 100 poss
Same raw point margin. Different efficiency. Miami squeezed nearly two more points out of every 100 possessions than Phoenix, even though both teams ended the night winning by 5. Per-game numbers said the two teams played equally well; per-100 numbers said Riley's Heat were noticeably more efficient.
What Net Rating does well

It puts every team and player on the same scale. A slow-pace team and a fast-pace team can finally be compared apples-to-apples per 100 possessions. This is the metric ESPN, the NBA's own broadcast, and most podcasts quote when they say "best offense in the league" or "most efficient defense." It is the per-possession common language of modern basketball.

Try It Yourself · Build Your Own Net Rating
Try this first
  1. Leave Points scored and Points allowed alone.
  2. Drag only the Pace slider — first all the way down to 85, then all the way up to 115.
  3. Watch what happens to the two outputs below: the Raw point margin per game changes a lot. The Net Rating per 100 possessions does not budge.
Raw point margin per game
+12.8
Net Rating per 100 possessions
+12.8

What you just saw: the team's scoring efficiency per 100 possessions did not change — the team is playing equally well at every pace. But the raw point margin per game shifted dramatically, because faster pace means more total possessions, which piles up both points scored AND points allowed. Per-game numbers lie about how good a team really is. Net Rating per 100 possessions strips that pace illusion out of the comparison so equally-good teams look equally good, no matter how fast or slow they play.

What Net Rating quietly includes: garbage time

The Net Rating numbers Basketball-Reference and ESPN report include every minute of every game, including blowouts where bench players are running out the clock. Garbage-time possessions are noisy: bench-versus-bench matchups tell a reader almost nothing about a starter's impact. Cleaning the Glass exists almost entirely to filter these out — they strip the fourth quarter once the score gets out of hand (≥25-point gap with 9-12 minutes left, ≥20 points with 6-9 minutes, ≥10 thereafter, plus a check that fewer than two starters are on the floor for either team).

The Cleaning the Glass version of Shai Gilgeous-Alexander's on-court Net Rating is +16.0; Basketball-Reference (which includes everything) reports +16.9. Same season, slightly different filter. Both are correct for the question they are answering; the difference is the question.

What Metric 03 proves · Net Rating

Pace bends raw plus/minus. Winning by 5 against a slow team is more scoring per possession than winning by 5 against a fast team. Net Rating divides by possessions and multiplies by 100 so every team and player can be compared on the same per-possession scale.

What it still misses. Net Rating normalizes pace, but it still does not know who did the work. A player's individual Net Rating depends on his teammates and his opponents. The next metric tries to crack that problem with a regression — a statistical model that takes every possession of the season and tries to solve, mathematically, for each individual player's contribution.

★ Story so far

Raw plus/minus tracked the scoreboard. On/Off asked what changed when the player sat. Net Rating stripped pace out so fast and slow teams compare fairly.

▶ Up next

Every metric so far measured what happened around the player. The next two try something harder: solve mathematically for what the player himself contributed.

Metric 04

Adjusted Plus / Minus

In one sentence Throw every possession of the season into a giant equation, then solve mathematically for each individual player's contribution.
✓ Best at

Trying to actually isolate the player from the lineup using math, not just averages. The math is unbiased — over a long enough sample, it really does converge on the right answer.

✗ Still cannot do

Give you a stable answer in small samples. Five games into a season, the math has wide confidence bands — same player, same skill, very different ratings depending on when the analyst checks.

Adjusted Plus/Minus was the first metric to isolate a player from his lineup using math, not just by comparing him to his bench. For every possession in a season, who was on the floor and what was the point margin? A regression — a statistical model that runs across thousands of possessions — can theoretically solve for each player's individual contribution. The math is rigorous. The data are thin.

Illustrative exampleThe two panels below use round numbers chosen to make the small-sample-versus-large-sample point visible. Pure unregularized Adjusted Plus/Minus with confidence bands is rarely published anymore; every modern source publishes the regularized variant covered in Metric 05. The interactive calculator below shows the math with adjustable sample size.

5 games into the season
Small sample · regression has little data to work with
−100+10
Estimate+9.2
Confidence band±6.5
True value could be+2.7 to +15.7
60 games into the season
Large sample · the regression has enough data to settle on a stable estimate
−100+10
Estimate+2.4
Confidence band±1.8
True value could be+0.6 to +4.2
The Adjusted Plus/Minus Math (in plain words)
For every possession: who was on the floor (10 players at a time) and what was the point margin? Solve the regression for each player's individual coefficient. Repeat across thousands of possessions.
Rosenbaum (2004) showed it could be done. The estimates are unbiased and the math is right. But until tens of thousands of possessions are logged, the answers swing wildly. Same player, same skill, two wildly different ratings depending on when the analyst checks.
What Adjusted Plus/Minus does well

It actually tries to solve for the player, not just describe what happened around the player. The math is unbiased: if Shai Gilgeous-Alexander is genuinely a +6 player, Adjusted Plus/Minus will report that — eventually. The "eventually" is the price of the rigor, and every modern impact metric (Regularized Adjusted Plus/Minus, Estimated Plus-Minus, LEBRON) inherits its core idea.

Try It Yourself · Watch the Confidence Band Tighten
Try this first
  1. Drag the games slider down to 5 games.
  2. Notice the orange confidence band. The estimate could realistically be anywhere from below +1 to above +14 — the regression has no idea who this player is yet.
  3. Now drag to 60+ games. Watch the band collapse to within ±2. That is the cost of rigor in Adjusted Plus/Minus: you have to wait.
Adjusted Plus/Minus estimate
+6.1
Confidence band
±4.0
−10−50+5+10+15

Drag the games slider from 3 to 82 and watch the orange confidence band. At five games, the estimate could realistically be anywhere from below +1 to above +14: the regression simply has no idea who this player is yet. By 60 games, the band has narrowed enough that the estimate is finally trustworthy. Adjusted Plus/Minus needed time. The next metric, Regularized Adjusted Plus/Minus, found a way to skip the wait. Note: the calculator uses a teaching approximation (standard error scales as 1/√games against an assumed All-Star "true value" of +4); real Adjusted Plus/Minus uses a regression with thousands of variables and lineup-specific complexity.

What Metric 04 proves · Adjusted Plus/Minus

Adjusted Plus/Minus is the first metric that tried to solve for the player, not just describe what happened around him. The fix is mathematically rigorous. The new flaw: regression estimates are noisy until the sample is huge.

The October trap. Five games into the season, the same star can post an Adjusted Plus/Minus of +9.2 with a ±6.5 confidence band. By March, with 60 games behind him, it tightens to +2.4 ± 1.8. Same player, same skill. Adjusted Plus/Minus just needed time and possessions.

Sample size is one problem; lineup overlap is another. Two players who share 80% of their minutes — Stephen Curry and Draymond Green, Jayson Tatum and Jaylen Brown, Nikola Jokić and Jamal Murray — are mathematically inseparable in Adjusted Plus/Minus. The regression cannot tell them apart no matter how many possessions are logged, because there is almost no data showing the team's performance with one on the floor without the other. Statisticians call this multicollinearity. The next metric, Regularized Adjusted Plus/Minus, fixes both problems with one math trick.

Metric 05

Regularized Adjusted Plus/Minus

In one sentence Same equation as Adjusted Plus/Minus, but with a brake: pull every wild number back toward zero unless the data really demand otherwise.
✓ Best at

Producing stable answers a coach can plan around. The first plus/minus variant whose number does not lurch wildly between samples. Foundation of every modern impact metric.

✗ Still cannot do

Tell you WHY a player's number is what it is. It reports the answer but does not show its work the way Box Plus/Minus does — it cannot point to a specific skill that explains where the +2.8 came from.

Joe Sill's 2010 MIT Sloan paper added one trick to Adjusted Plus/Minus: a ridge penalty that pulls every estimate toward zero unless the data strongly justify otherwise. Players with thin samples get conservative ratings instead of wild ones. Same data, same possessions, dramatically more stable answers. Regularized Adjusted Plus/Minus is the first plus/minus variant that actually behaves.

Illustrative exampleThe two panels below use round numbers chosen to highlight what ridge regularization does to a thin-sample estimate. The interactive calculator below shows the same math with adjustable sample size and a toggle for ridge.

Adjusted Plus/Minus · five games in
No regularization
−100+10
Estimate+9.2
Confidence±6.5
Regularized Adjusted Plus/Minus · same five games
Ridge penalty shrinks toward zero
−100+10
Estimate+2.8
Confidence±1.8
What ridge regression actually does
Ridge penalty: when the data are thin, pull every player's regression coefficient (their estimated rating) toward zero — the league average. Only let it stray from zero when the data demand it.
Most NBA players are within a few points per 100 possessions of average. A regularized estimator builds that fact into the math itself — a player who has not yet shown a clear effect gets a rating near zero. Real outliers (stars, scrubs) earn their distance from zero over time. Sill (2010) reported that Regularized Adjusted Plus/Minus nearly doubled the predictive accuracy of unregularized Adjusted Plus/Minus.
What Regularized Adjusted Plus/Minus does well

It finally produces stable answers. The same player, sampled across different windows of the season, no longer gives wildly different ratings. This is the first plus/minus variant whose number a coach or front office can actually plan around — and the foundation of every modern impact metric, from Estimated Plus-Minus to LEBRON to DARKO.

Try It Yourself · Watch Ridge Regularization Shrink Estimates Toward Zero
Try this first
  1. Set games to 10 (a small sample).
  2. Toggle ridge on and off. Without ridge, the estimate sits far from zero with a wide band. With ridge on, both the estimate and the band shrink dramatically toward zero — the math says "with this little evidence, I should not be confident this player is unusual."
  3. Now drag games to 70 and toggle again. Notice ridge barely changes anything. The penalty only matters when the data are thin. That is the entire trick.
Estimate
+1.5
Confidence band
±1.8
−10−50+5+10+15

Try a small sample (10 games or fewer) and toggle ridge regularization on and off. The raw Adjusted Plus/Minus estimate is wild — far from zero AND uncertain. Ridge pulls it back toward zero by a lot. Now drag games to 70 and toggle again: at large samples the ridge barely changes anything. The penalty only matters when the data are thin. That is the trick that made impact metrics actually usable. Note: the calculator uses a teaching approximation for shrinkage (a 1/(1+games/10) factor); real ridge regression in basketball analytics picks the penalty strength via cross-validation across the full league.

What Metric 05 proves · Regularized Adjusted Plus/Minus

Same regression, smarter prior. Regularized Adjusted Plus/Minus shrinks unreliable estimates toward zero and lets reliable ones stand. The first plus/minus variant whose answers do not lurch around between samples.

What it still misses. Regularized Adjusted Plus/Minus is stable, but it tells you the answer without showing its work the way Box Plus/Minus does: it reports that a player is +2.8, but not why. It cannot say "because he creates good shots" or "because he protects the rim." The next metric tries to predict Regularized Adjusted Plus/Minus from the box score alone, so the rating becomes computable from public stats — and explainable in terms of what a player actually does.

★ Story so far

The math isolates the player AND the ridge brake stabilizes the answer. Regularized Adjusted Plus/Minus finally gives a number a coach can plan around — but it does not say why.

▶ Up next

The next metric closes that loop. It predicts the stable number from the player's box score alone — points, rebounds, assists, steals, blocks, turnovers, shooting. Numbers anyone can see.

Metric 06

Box Plus / Minus 2.0

In one sentence If you only have a player's box score (points, rebounds, assists, steals, blocks, turnovers, shooting), can you guess their impact? Yes — surprisingly close.
✓ Best at

Computing a usable impact estimate from numbers that already appear in every box score. No subscription, no proprietary data, no play-by-play feed. The foundation of Value Over Replacement Player.

✗ Still cannot do

Catch a defender who never shows up in the box score, or a screen-setter whose value lives in plays the stats do not record. If a skill does not produce a box-score number, Box Plus/Minus cannot see it.

Daniel Myers (2020) ran a regression: given a player's box-score profile per 100 possessions, what does Regularized Adjusted Plus/Minus predict that player's impact to be? The output is Box Plus/Minus 2.0. Anyone with a stat line can compute it. The cruncher below uses the official Basketball-Reference coefficients, applied to Shai's per 36 minutes line. Click the Shai Gilgeous-Alexander 2024-25 preset and the output lands at his actual +11.5 — not an approximation.

Try this first
  1. Click the Shai Gilgeous-Alexander 2024-25 preset button below (the orange one).
  2. Watch all ten inputs snap to his verified line and the output land at his actual +11.5 Box Plus/Minus.
  3. Then click Role Player and watch the same official formula drop toward replacement-level. Same math, different inputs, very different verdict.
Box-score inputs · per 36 minutes
Context · position, role, team
Live Box Plus/Minus 2.0
+0.0
points above league average per 100 possessions
−50+5+10+15
Value Over Replacement Player (season) 1.0
Archetype Average Starter
The Box Plus/Minus 2.0 Formula (official coefficients)
For a point-guard creator, per 100 possessions: BPM = 0.860 × Points + 0.389 × 3PM + 0.580 × Assists − 0.964 × Turnovers + 0.613 × OffReb + 0.116 × DefReb + 1.369 × Steals + 1.327 × Blocks − 0.367 × Fouls − 0.560 × FGA − 0.246 × FTA − 0.818 (position constant) − 2.774 (role constant) − 8 (intercept). For a center, the assist, rebound, steal, and block weights shift toward the position-5 values. For a catch-and-shoot finisher, the FGA and FTA weights and the role constant shift toward the role-5 values.
Every coefficient above is published verbatim on Basketball-Reference's official Box Plus/Minus 2.0 methodology page. The cruncher converts your per-36 sliders to per-100 possessions internally, derives field-goal and free-throw attempts from your points and true-shooting inputs, and applies the position-and-role weights linearly between the two endpoints. The team scoring margin slider applies the team-context points adjustment: a player on a strong team gets their points discounted slightly, because some scoring credit goes to teammates. Load the Shai Gilgeous-Alexander preset and the output lands at his actual +11.5.
What Box Plus/Minus 2.0 does well

It is computable from numbers that already appear in every box score. No play-by-play feed, no proprietary lineup data, no subscription. Anyone with a stat line and a calculator can produce a usable impact estimate. It is also the foundation of Value Over Replacement Player — the closest the NBA has to baseball's Wins Above Replacement, and the metric Basketball-Reference uses to rank every player in league history on a single scale.

What Metric 06 proves · Box Plus/Minus 2.0

Regularized Adjusted Plus/Minus can be predicted from the box score alone. No play-by-play, no tracking data (camera-fed player movement records), no proprietary model. Anyone with a stat line and a calculator can compute it. Box Plus/Minus 2.0 is the public estimate of impact Basketball-Reference uses, and the foundation of Value Over Replacement Player — the closest thing the NBA has to baseball's Wins Above Replacement.

One chapter left. Today's frontier metrics (Estimated Plus-Minus, DARKO, LEBRON) layer tracking data and on/off impact on top of what Box Plus/Minus 2.0 already does. Their formulas are proprietary, but their leaderboards are public. The next section lines up all four side by side and shows in which moments they disagree about who deserved the 2024-25 Most Valuable Player award.

Metric 07

Where the Models Disagree

In one sentence Modern impact metrics split into two families that answer the question differently. One family reads the box score; the other family reads on-court team results. They rank the same season's top players differently.

All modern impact metrics fall into one of two families. The box-score family (Win Shares, Box Plus/Minus 2.0) answers the question by reading a player's box-score profile and weighting the stats to predict impact. The lineup family (Regularized Adjusted Plus/Minus, Estimated Plus-Minus, LEBRON, the now-discontinued RAPTOR) answers the question by reading what happens to the team when the player is on or off the court. Both families look at the same 2024-25 season. They agree on who is in the top tier. They disagree about the order. Below: one representative from each family, side by side, with Shai Gilgeous-Alexander highlighted in both.

Two splits, not one

The box-score-versus-lineup split is one layer. There is another underneath: descriptive metrics (Box Plus/Minus 2.0, Win Shares) tell a reader what already happened this season. Predictive metrics (Estimated Plus-Minus, DARKO, the discontinued RAPTOR) are explicitly forecast-oriented — they bake in age curves, recency weighting, and stability priors to estimate what the player will do next, not just what already happened.

When someone asks "which metric is best?" the honest answer depends on whether they want to evaluate the past or predict the future. The Most Valuable Player debate is a descriptive question. A trade-deadline acquisition decision is a predictive question. Different question, different metric.

Two leaderboards, one from each family. The Box Plus/Minus 2.0 column represents the box-score family. The Estimated Plus-Minus column represents the lineup family. Click any source name to open the methodology page.

Box Plus/Minus 2.0 box-score family
basketball-reference.com
  1. 1Nikola Jokić · Denver13.3
  2. 2Shai Gilgeous-Alexander · Oklahoma City11.5
  3. 3Giannis Antetokounmpo · Milwaukee9.5
  4. 4Luka Dončić · Dallas / L.A. Lakers6.7
  5. 5Victor Wembanyama · San Antonio6.5
  6. 6Stephen Curry · Golden State6.3
  7. 7Tyrese Haliburton · Indiana5.8
  8. 8LeBron James · L.A. Lakers5.6
  9. 9Anthony Davis · L.A. Lakers / Dallas5.4
  10. 10Jayson Tatum · Boston5.2
Estimated Plus-Minus lineup family
dunksandthrees.com
Shai Gilgeous-Alexander
#1 in 2024-25

The Estimated Plus-Minus leaderboard loads inside an interactive web table that requires a live browser session to view, so its full top ten cannot be reprinted here. The Ringer's Most Valuable Player debate (Michael Pina, April 8 2025, published five days before the regular season ended) links explicitly to Shai Gilgeous-Alexander at #1 on the Dunks & Threes Estimated Plus-Minus leaderboard, with Nikola Jokić very close behind. Final-week movement could not be independently verified.

View live leaderboard ↗
Two other modern models, briefly Two more public impact metrics — DARKO Daily Plus-Minus (darko.app) and LEBRON (bball-index.com) — also rank players in 2024-25 using ridge-regression impact models layered on top of tracking data. Their full leaderboards sit behind a paid subscription, so they would only be teased here, not shown. Mentioned for completeness; the agreement-versus-disagreement story between Box Plus/Minus 2.0 and Estimated Plus-Minus carries the same lesson.
The Story in One Line
Shai · Box Plus/Minus rank#2 (11.5)
Shai · Estimated Plus-Minus rank#1
Jokić · Box Plus/Minus rank#1 (13.3)
Jokić · Estimated Plus-Minus rankclose 2nd

Two families, two different Most Valuable Player ballots. The box-score family (Box Plus/Minus 2.0 here, but also Win Shares) rewards Jokić's volume rebounding, playmaking, and shot quality as a center. The lineup family (Estimated Plus-Minus here, but also Regularized Adjusted Plus/Minus and LEBRON) weighs on-court team impact more heavily and ranks Shai Gilgeous-Alexander first behind Oklahoma City's NBA-record-tier 68-14 season and +12.9 scoring margin. Voters split the difference: Shai won Most Valuable Player with 71 of 100 first-place votes and a 0.913 award share, leaving Jokić as runner-up at 0.787.

What Metric 07 proves · The Modern Stack

The two families of impact metrics, fed the same season, can produce two different Most Valuable Player ballots. The disagreement is not noise. It comes from a real philosophical split. The box-score family asks: "What did this player produce?" The lineup family asks: "What happened to the team while this player was on the court?" Different questions, different answers, both defensible.

What this all proves. Plus/minus, in 2026, is a family of arguments dressed up as a number. The single number attached to a player's name is one model's published verdict on a question that is still genuinely open. Every metric in this tutorial is part of how the question gets asked. None of them ends it.

The whole journey, in one diagram

The Family Tree of Plus/Minus

Five metrics build on each other in a single chain. The fifth metric splits the family into two branches that still argue about who is best.

1967-68 1997 2000s 2003 2004 2010 0 Raw Plus/Minus · borrowed from NHL hockey "What did the scoreboard do while you were on the floor?" 1 Raw Plus/Minus (basketball) Same idea, basketball version. The simplest impact number. 2 On / Off Differential First fix: account for what changes when the player sits. 3 Net Rating (per 100 possessions) Second fix: strip pace out so fast and slow teams compare fairly. 4 Adjusted Plus/Minus (Rosenbaum) Third fix: solve for the player with math. Rigorous, but noisy. 5 Regularized APM (Sill) Fourth fix: ridge brake makes the answers stable. The foundation of every modern impact metric. BOX-SCORE FAMILY LINEUP FAMILY

Box-score family — "What did this player produce?"

  • Win Shares · Basketball-Reference, adapted from Bill James's baseball Win Shares. Translates box-score events into marginal offensive and defensive contributions, then converts to wins. Cumulative across the season, so volume matters; league-leading total in 2024-25 was 16.7 (Shai).
  • Metric 06 · Box Plus/Minus 2.0 · Myers, 2020. Predicts Regularized Adjusted Plus/Minus from a player's box-score profile per 100 possessions, with position and team-strength adjustments.

Lineup family — "What happened to the team while this player was on the court?"

  • Estimated Plus-Minus · Dunks & Threes, ongoing. The most widely-cited public metric in this family.
  • Also published, methodology proprietary: DARKO Daily Plus-Minus (Medvedovsky) and LEBRON (BBall-Index). Both layer ridge-regression impact models on top of tracking data; their exact calculations are not public.
  • RAPTOR · FiveThirtyEight, 2019–2023. Updates ceased in 2023 when Nate Silver left ABC News; the site itself fully closed in March 2025.
Now you decide

After everything you just read, who is your pick for 2024-25 Most Valuable Player?

—

—

—

The Payoff

One Season. Six Verdicts.

Shai Gilgeous-Alexander played one Most Valuable Player season. Six different metrics looked at it and saw six different stories. Same player, same 76 games, six numbers that can mean different things to different rooms.

Metric 01
Raw Plus / Minus
+918
season total · 76 games
Oklahoma City outscored opponents by 918 cumulative points across Shai's 2,598 minutes — among the highest single-season plus/minus totals since the NBA started tracking it in 1997. The number the original 1967-68 NHL statistic would have given him. Source: Basketball-Reference game log; historical context from The Ringer (article published April 8 2025, citing Stathead).
Metric 02
On / Off Differential
+11.2
net rating swing · per 100 possessions
Oklahoma City's net rating with Shai on the floor (+16.9) minus its net rating without him (+5.7). The +11.2 swing measures the gap between Shai and Cason Wallace, the guard who logged the most minutes in his place. Source: Basketball-Reference on/off splits.
Metric 03
Net Rating (on-court)
+16.9
per 100 possessions
Oklahoma City's combined offense + defense per 100 possessions while Shai was on the floor (Basketball-Reference primary; Cleaning the Glass, which strips garbage time, reports a slightly lower +16.0 — same season, slightly different filter). For context, the Thunder's team-wide scoring margin was +12.9 per game, the largest in the NBA since the 1971-72 Los Angeles Lakers. Sources: Basketball-Reference; Cleaning the Glass; team record from Wikipedia.
Metric 04 · 05
Regularized Adjusted Plus/Minus
top tier
dunks & threes
Shai's stabilized Regularized Adjusted Plus/Minus lives in interactive applications that require a browser session to read; the published rankings place him in the league's top tier alongside Nikola Jokić. Source: dunksandthrees.com (manual access required).
Metric 06
Box Plus / Minus 2.0
+11.5
2nd in league · Basketball-Reference
Shai finished second only to Nikola Jokić (+13.3) in 2024-25 Box Plus/Minus. His Value Over Replacement Player of 8.9 was also second to Jokić's 9.8. (Replacement Player means a typical end-of-bench player or G-League call-up; Basketball-Reference fixes the baseline at −2.0 points per 100 possessions, so everything above that becomes a player's Value Over Replacement.) Box-score metrics tend to reward Jokić's volume rebounding and playmaking from the center position. Source: Basketball-Reference Advanced.
Metric 07
Estimated Plus-Minus
#1
dunks & threes
The Ringer's Most Valuable Player debate (April 2025) explicitly links to Dunks & Threes showing Shai at #1 in Estimated Plus-Minus. Impact-based metrics flip the Box Plus/Minus verdict: Jokić leads on box-score, Shai leads on on-court impact. Source: The Ringer, hyperlinked to dunksandthrees.com.
Per 36 minutes · 34.4 Points · 5.3 Rebounds · 6.7 Assists · 1.8 Steals · 1.1 Blocks · 2.5 Turnovers · 63.7% True Shooting · 34.8% Usage Rate · 16.7 Win Shares (led league)
Per game · 32.7 Points · 5.0 Rebounds · 6.4 Assists · 51.9% Field Goals · 37.5% Three-Pointers · 89.8% Free Throws · 76 Games · 34.2 Minutes per Game · Most Valuable Player (913 of 1,000 award share, 71 first-place votes)
Sources: Basketball-Reference player page · Basketball-Reference on/off splits · Basketball-Reference 2024-25 advanced leaderboard · Cleaning The Glass · Dunks & Threes Estimated Plus-Minus · The Ringer Most Valuable Player debate
Built by Kevin Luczynski · May 2026
Future companion explainers

This was the plus/minus family. These earn their own walk-through next.

Plus/minus is one branch of the impact-metric tree. Each metric below answers the same "how good was this player?" question through a different lens — and each one deserves its own page with the same TLDR, pros and cons, and interactive cruncher this one used.

Tell Kevin which one to build next.