Devin Harris

DAWG — Recorded Activity combines ten recorded actions. Four come from the box score: steals, blocks, offensive rebounds and assists. Six come from the NBA's hustle tracking: deflections, loose balls recovered, charges drawn, contested shots, screen assists and box-outs. It does not capture rotations, help defence, transition effort or missed assignments. It is a recorded-activity profile, not effort, defensive quality or total value, and it is separate from the published six-family NBA DAWG.

Each action is counted per 36 minutes. Hustle tracking covers only the games the league tracked, so hustle rates use the minutes from those games. In this edition, those games account for 99% of all minutes played. The adjusted rate blends a player's record with the league population, using a Gamma–Poisson model fitted on nearby seasons (2017, 2018, 2020, 2021, 2022), never on the season being scored. Fewer minutes means more pull toward the league average.

Each player is judged against his own position. Guards, forwards and centres come from each player's most frequent starting slot: a coach's listing, not something inferred from the stats being ranked. A guard's blocks are compared with other guards' blocks, and a centre's assists with other centres'. Unusual activity for a role (a guard blocking shots or drawing charges, a centre passing) therefore counts for what it is, instead of being drowned out by the specialists. The rulers are frozen and come from the same era: guards 144, forwards 143, centres 62 player-seasons from 2019 with at least 492 minutes and all ten actions recorded.

The DAWG index is the plain average of the ten position-relative percentiles. The league-wide index beside it averages percentiles against every position (350 player-seasons, reference eee9d64b98f2), and it leans toward centres, who rack up box-outs, contests and screen assists. League rates have moved a lot; box-outs per 36, for example, fell by about four-fifths between 2017-18 and 2022-23. So percentiles are relative to the era, not absolute across eras. The equal weights are a fixed policy, not learned values. If any action is unknown, the index is not shown, and the rest are never re-weighted.

Defensive rebounds, turnovers and personal fouls are shown separately and are not subtracted.

• Hustle counts cover only the games the provider tracked; each player's hustle rates use the minutes from those games, shown as coverage.
• Ten recorded actions only: box-score steals, blocks, offensive rebounds and assists, plus provider hustle counts. Rotations, help defence, transition recovery and missed responsibilities are not recorded.
• This is not the full DAWG model: there are no eligible opportunities, harmful outcomes or validated role model, so it is a recorded-activity profile, not effort, defensive quality or total value.
• Percentiles compare each player with his own position group (guards, forwards or centres, from his most frequent starting slot) in the same era. They are not absolute comparisons across eras or positions. The league-wide index is shown alongside.
• Position groups come from starting-lineup slots, which are coarse: a power forward who starts beside a centre is a forward. Players who never started are compared with the whole league.
• Adjusted rates are pulled toward the league-wide average, not the position average, so low-minute guards' blocks and box-outs (and centres' assists) read slightly high within their group.
• An average of ten equally weighted component percentiles, not a percentile of the composite, a probability or points added.
• Turnovers and fouls are shown separately and are not subtracted.
• Hustle counts exist only as season totals, so no game-resampling stability range is shown; component intervals are conditional empirical-Bayes estimates that ignore hyperparameter uncertainty.
• The 80% ranges are poorly calibrated. On the sealed 2023-24 and 2024-25 holdout, next-season counts fell inside them 56–88% of the time, depending on the action. Treat them as a rough sense of uncertainty, not a guarantee.

Edition

nba_dawg_recorded_activity_v1 v1.1.0

Sources

NBA Stats hustle totals and player box scores, via SportsDataverse

Scale

Adjusted (passed the holdout gate)

Snapshot

389996c63a51