HAPM

Held-Out Adjusted Plus-Minus ? HAPM measures how many points per 100 possessions a player's stints produced above or below expectation. The other nine players in each of his stints are rated without any of his possessions. Their ratings give the margin the stint would produce with an average player in his place, and his HAPM is the actual margin minus that. Players with few possessions are pulled toward a typical value for their playing time.

0 is the average rated player. Rotation players average about +1.3. Each cell shows the value and its percentile.
Compare players Data on GitHub
▸ What is HAPM?

HAPM (held-out adjusted plus-minus) is a player's value in points per 100 possessions. Whereas RAPM solves for all 10 players at once and splits each stint's result among them, HAPM holds the player out, estimates each of his stints from the other nine players, and credits him with the actualized difference. 0 is an average player; rotation players average about +1.3.

▸ How HAPM works

HAPM is a player's net impact per 100 possessions, measured against an expectation built without him. The expectation comes from a ridge regression on every stint the measured player did not play, excluding his possessions from the entire regression, including the priors from each player's three previous seasons. His raw value is the possession-weighted mean of actual minus expected margin over his stints, with 0 in his place. That value is shrunk toward a playing-time center by its reliability.

1. Pick the player, rate the other nineTeammates and opponents are rated from every stint where the chosen player did not play, starting from their three previous seasons.
→
2. Estimate each stintThose ratings give the expected margin for each of his stints, with an average player in his place.
→
3. Actual minus expectedThe chosen player's HAPM is what his stints produced above or below that estimate, per 100 possessions.

One stint

What HAPM computes for one stint. Illustrative numbers.

His four teammates+1.8+0.4−0.6+2.1he is 0
The five opponents+1.1−0.2+0.8+0.5−0.9
each rated only from the stints he was not on the floor for
↓
Expected marginteammates' ratings minus opponents', with 0 in his place: +2.4
Actual marginwhat the stint produced: +7.0
↓
His value for the stint+7.0 − (+2.4) = +4.6 per 100
Every stint he plays is scored this way. His season value is the possession-weighted average, pulled toward a typical value for his playing time.

Worked example: Jokić and Caldwell-Pope, Denver 2023-24

The two shared most of their minutes. Per 100 possessions.

BothJokić without Caldwell-PopeCaldwell-Pope without JokićNeither
Possessions4,4031,1054281,962
Net rating+12.4+9.1+2.1−11.8
JokićCaldwell-Pope
Actual margin of his stints+11.1+11.3
Expected with an average player in his place−5.3+6.4
from his four teammates+1.3+13.0
from the opponents and game context−6.7−6.6
Actual minus expected+16.4+4.9
HAPM 2/3 (percentile)+12.6 (100th)+4.3 (82nd)
RAPM 2-year (percentile)+6.9 (100th)+4.3 (94th)

Caldwell-Pope's stints were expected at +6.4 because Jokić was usually beside him (+13.0 from his teammates). Jokić's teammates added only +1.3, so his stints were expected at −5.3. HAPM credits each player with the gap between those expectations and what happened. HAPM 2/3 and 2-year RAPM also use 2022-23; percentiles are among 2023-24 players with 1,500+ possessions.

Versions

Each version uses a different number of seasons of the player's own stints, with newer seasons counting more. The 3 means the other nine are rated with their three previous seasons.

▸ Why HAPM

True value

A player's HAPM is what he adds to his stints, in points per 100 possessions. When a team's top player sits, the team loses about what his HAPM says.

What the team actually lost, divided by what each metric forecast. 1 is the right size; above 1, the metric underrates the player.

Who sitsHAPM 2/3RAPM 4-yearRAPM 2-yearRAPM 1-year
Top player by possessions.901.852.303.67
Second.871.732.033.78
Third to fifth.981.952.313.78
Sixth to eighth1.452.563.207.51

Every team-game from 2013-14 to 2025-26, compared within team and half-season, with home court and opponent strength held fixed. Each forecast uses values from the other half of the season. Closest to 1 in bold.

No teammate bias

A player's HAPM barely moves with the strength of his teammates.

How much a rating moves withHAPM 2/3RAPM 4-yearRAPM 2-yearRAPM 1-year
The strength of his teammates−.02+.02+.08+.23
The strength of the players who replace him−.05−.01−.03−.10

1,359 cases of a player staying on his team from one season to the next, 2013-14 to 2025-26, 2,000+ possessions in both seasons. Each value is the change in next season's rating per standard deviation of his teammates' (or replacements') strength, measured by their previous-season RAPM, given his own rating the season before. 0 means no pull. Closest to 0 in bold.

Next season

How well one season's values predict the next season.

Next seasonHAPM 2/3RAPM 4-yearRAPM 1-year
Game margins (R²).158.149.129
Team ranking (rank correlation).744.742.692
Five-man units (R²).109.116.109
Five-man units never seen together (R²).096.102.095

4-year RAPM uses four seasons of each player; HAPM 2/3 uses two. Best value in each row in bold. Season pairs from 2013-14 to 2025-26.

CSV

NBA HAPM leaders, 2020-21

RkPlayerPosTeamHAPMOffDefPoss
1Rudy GobertCUTA+9.95+4.58+5.374,508
2Kawhi LeonardSFLAC+8.10+5.75+2.353,478
3Royce O'NealeSFUTA+7.09+4.91+2.184,593
4Paul GeorgeSFLAC+7.07+4.10+2.973,577
5Devin BookerSGPHX+6.72+7.47−0.754,445
6LeBron JamesPGLAL+6.46+3.05+3.413,099
7Nikola JokićCDEN+6.02+5.80+0.224,799
8Jayson TatumSFBOS+5.91+4.60+1.314,452
9Duncan RobinsonSFMIA+5.89+3.33+2.564,392
10Jimmy Butler IIISFMIA+5.88+3.11+2.763,353
11Chris PaulPGPHX+5.86+3.86+2.004,298
12Mikal BridgesSFPHX+5.84+3.91+1.934,581
13Trae YoungPGATL+5.83+6.78−0.964,223
14Joel EmbiidCPHI+5.80+2.59+3.213,078
15Giannis AntetokounmpoPFMIL+5.78+2.40+3.384,214
16Stephen CurryPGGSW+5.77+5.05+0.724,668
17Clint CapelaCATL+5.69+1.27+4.423,715
18Pascal SiakamPFTOR+5.65+3.60+2.054,056
19Ja MorantPGMEM+5.41+4.73+0.684,332
20Julius RandlePFNYK+5.39+2.16+3.235,128
21Draymond GreenPFGSW+5.34+3.71+1.634,238
22Kevin DurantPFBKN+5.26+3.63+1.632,442
23Karl-Anthony TownsCMIN+5.19+5.43−0.243,336
24Dorian Finney-SmithPFDAL+5.19+4.52+0.673,687
25Dillon BrooksSFMEM+5.19+3.55+1.644,214
26Damian LillardPGPOR+5.16+6.62−1.454,888
27Joe HarrisSFBKN+5.10+3.68+1.434,441
28Richaun HolmesCSAC+5.10+3.32+1.783,688
29Joe InglesSFUTA+5.02+2.98+2.043,845
30Jamal MurrayPGDEN+4.80+3.81+0.993,228
31Jakob PoeltlCSAS+4.75+2.73+2.023,608
32Mike ConleyPGUTA+4.54+1.65+2.893,080
33Cameron PaynePGPHX+4.43+3.11+1.322,096
34Luka DončićPGDAL+4.43+5.54−1.114,400
35John CollinsPFATL+4.18+2.65+1.533,612
36CJ McCollumSGPOR+4.05+4.20−0.143,296
37Jonas ValanciunasCMEM+4.02+2.08+1.943,682
38Alex CarusoPGLAL+4.01+0.23+3.782,384
39De'Anthony MeltonPGMEM+4.01+2.17+1.842,195
40Gordon HaywardSFCHA+3.98+2.46+1.523,004
41Thaddeus YoungPFCHI+3.90+3.44+0.463,241
42Georges NiangPFUTA+3.72+1.60+2.122,401
43Jrue HolidayPGMIL+3.63+2.98+0.653,992
44James HardenPGBKN/HOU+3.55+3.81−0.263,296
45Deandre AytonCPHX+3.54+2.84+0.704,145
46Robert CovingtonPFPOR+3.46+1.79+1.674,542
47PJ DozierSGDEN+3.43−0.06+3.492,076
48Zion WilliamsonPFNOP+3.43+3.63−0.204,030
49Jarrett AllenCCLE/BKN+3.39+1.97+1.423,664
50Derrick RosePGNYK/DET+3.38+1.72+1.672,353
▸ What the columns show
ColumnMeaningColumnMeaning
HAPMPoints per 100 possessions his stints produced above or below what the other nine were expected to produce with an average player in his place. 0 is the average rated player; rotation players average about +1.3. Off, DefHAPM at each end of the floor.
PercentileThe small number in each cell. It ranks him among all players with the minimum possessions, across positions and teams. Cells run from red (low) through gray to green (high). PossPossessions he played in the seasons selected.
Shot valueSelection (offensive and defensive shot quality impact) and conversion (offensive and defensive shot making impact). Hover the ? next to its title for its tree. Possession valueGeneration (offensive rebounding, forced turnovers) and retention (turnovers avoided, defensive rebounding). Hover the ? next to its title for its tree.
ZonesShot value by zone: rim, mid-range, three and free throws. Defensive conversion has two zones, rim and other, because the rest changes too much from year to year. Turn them on with the Zones button. D ConvChanges the most from year to year, because opponents' shooting luck is a large part of it.
Versions1/3 to 5/3, chosen in the Version menu and explained under "How HAPM works". Versions with more seasons are steadier and react more slowly. Season rangesEach player's values are averaged over the seasons selected, weighted by possessions.
YTYoung teammates. Half or more of his possessions came alongside players with fewer than three earlier seasons, and HAPM tends to run high for these players. Team lineThe last row when a team is picked. It averages the team's players, weighted by possessions; its percentiles rank it among all teams under the same filters. HAPM values do not add up to a team's margin.
Teams viewTeams ranked by the possession-weighted average HAPM of their players, which compares rosters. Click a team to see its players. 1997-98The first season has no earlier seasons to rate the other nine, so its values are less certain.
Games throughThe current season's values include every game up to the number shown. ColumnsThe buttons show or hide each group. The table opens with HAPM, Off and Def.
FiltersMinimum (≥) and maximum (≤) boxes under each column, in points per 100 possessions (Poss in possessions). Rank and percentiles still count every player.
Seasons: 2026-27 2025-26 2024-25 2023-24 2022-23 2021-22 2020-21 2019-20 2018-19 2017-18 2016-17 2015-16 2014-15 2013-14 2012-13 2011-12 2010-11 2009-10 2008-09 2007-08 2006-07 2005-06 2004-05 2003-04 2002-03 2001-02 2000-01 1999-00 1998-99 1997-98 · All seasons
LUMA Basketball Research · Data on GitHub (CC BY 4.0) · © 2026 Court Share LLC