1 · The Four Factors
In Basketball on Paper (2004), Dean Oliver proposed explaining a game's result with four box-score factors: shooting (eFG%), turnovers (TOV%), offensive rebounding (OREB%) and free throws (FTM/FGA). He attached rule-of-thumb weights of 40 · 25 · 20 · 15.
Those weights are not the output of a statistical fit. Poropudas (2023) describes them as an ad hoc weighting, and published regression estimates on NBA data give more weight to shooting and less to free throws. This study does not reuse those figures: it measures the weights in the ACB, with its own data and an explicit method, and compares them with Oliver's in order and proportion.
2 · Data and method
- Source: play-by-play and box scores of the Liga ACB, ten seasons, 2016/17 to 2025/26.
- Weights: one row per game, from the home team's perspective. Each factor as a home-minus-away differential, standardised (z-score) within season. Linear regression of the game margin on the four differentials; a factor's weight is its absolute coefficient normalised to 100. It is an explanatory model of games played, not a forecast.
- Levers: team-by-season grain. Pearson correlation between each factor and each style lever. Association, not causation.
- Levers are shares, not accuracy: shot share by zone (rim, inside = rim + paint, mid-range, three) as attempts in that zone over the team's total field-goal attempts; assisted baskets as a share of makes; pace as possessions per 40 minutes.
- Two windows, stated in each table. Zone and passing levers are measured the same way across all ten seasons (180 team-seasons). Shot-clock levers are comparable only from 2020/21 (109 team-seasons): the earlier play-by-play records twice as many shots in the first three seconds, a recording artefact rather than basketball.
- Uncertainty: 95 % intervals by bootstrap, 2,000 replicates. An interval that excludes zero is a real association at this level; one that includes it is not distinguishable from zero.
3 · Factor weights in the ACB, against Oliver
| Factor | ACB weight | 95 % interval | Oliver (2004) |
|---|---|---|---|
| Shooting (eFG) | 46.8 % | 46.3–47.3 | 40 |
| Turnovers (TOV) | 23.7 % | 23.3–24.2 | 25 |
| Off. rebounding (OREB) | 20.1 % | 19.6–20.5 | 20 |
| Free throws (FTM/FGA) | 9.4 % | 8.9–9.9 | 15 |
Oliver's ordering holds; his weights do not. Shooting carries seven points more and free throws six points less, with intervals one point wide. Turnovers and rebounding land where he put them. The four differentials account for 93 % of the variance in the game margin (R² 0.932), which is why the weights are so tightly determined.
The result is stable: across the ten seasons shooting never falls below 45 and free throws never rise above 12, with no trend. And the gap with Oliver is not a Spanish anomaly. His are rule-of-thumb figures, and published regression estimates on NBA data also give more to shooting and less to free throws. That is cited as context, not as a number-for-number comparison: different method, different data.
4 · What moves each factor
A factor is a result, not a decision: nobody coaches "make more shots". The useful question sits one level below. Each factor has between three and four levers whose interval excludes zero. The strongest for each:
- Shooting ← assisted baskets: r +0.38 [+0.25, +0.49].
- Turnovers ← inside share: +0.29 [+0.14, +0.42]. Three-point share runs the other way, −0.28 [−0.39, −0.16]: attacking the paint risks the ball more than shooting from outside.
- Offensive rebounding ← mid-range share: −0.29 [−0.43, −0.13]. A missed shot from close falls under the rim; a long two comes off long.
- Free throws ← pace: +0.29 [+0.15, +0.42]; and mid-range share, −0.26 [−0.40, −0.10].
Two decisions carry the pattern: moving the ball, which goes with shooting, the factor that weighs most; and staying out of the mid-range, which touches three of the four factors.
Two coaching beliefs do not show up at team level: shooting closer to the rim is not associated with more free throws (+0.12 [−0.02, +0.25]), and crashing the offensive glass is not associated with conceding transition (+0.07 [−0.14, +0.29]). Both intervals include zero. The effect could still exist possession by possession and wash out when averaged over a season.
5 · The shot clock: where the early shot comes from
Within the possession the clock matters: eFG of 60.6 % on shots taken in the first eight seconds, 55.0 % between nine and sixteen, and 49.5 % at seventeen or more. As a team trait the sign flips: a larger share of early shots goes with lower shooting (−0.25 [−0.42, −0.06]) and more turnovers (+0.28 [+0.07, +0.46]).
Splitting by where the possession started explains it. Of every 100 shots taken in the first eight seconds, 37 follow the team's own offensive rebound, 23 an opponent turnover, 27 a defensive rebound and 12 an inbound after an opponent make. They are not worth the same: eFG 66.7 % after an opponent turnover and 63.2 % after a defensive rebound, which is real transition; 58.3 % after an offensive rebound; 50.1 % after an inbound.
At team level it is the putback share that drives the relationship, −0.45 [−0.60, −0.28] with shooting. Clean transition, without an offensive rebound, is not associated with the team's shooting at all (−0.13 [−0.32, +0.07]).
The early shot is efficient when it happens, but no team shoots better by taking more of them. What separates teams with a high early-shot share is that they rebound and put back more, and that goes with worse shooting overall. The shot clock is not a style lever; where the possession starts is what explains the pattern.
Caveats
- Association, not causation, and at season grain. An r of 0.38 accounts for 14 % of the variance; the message is the sign and the order of magnitude, not the decimal.
- Two windows. Zone and passing, ten seasons (180 team-seasons); shot clock, six (109), because of the play-by-play recording change in 2020/21.
- Coordinate coverage. 4 % of shots arrive without coordinates in the ACB feed, almost all made twos (dunks and putbacks), 41 % of them in the first eight seconds. They are excluded from the zone shares and included everywhere else.
- Compositional data. Shot shares sum to 100 %: more threes is necessarily less of something else, so the contrasts are read, not the raw levels.
- Weights are not levers. That shooting carries 47 % does not say how shooting is improved.
- No player names: this is a study of leagues and teams.
References and full version
Oliver, D. (2004). Basketball on Paper. Potomac Books.
Poropudas, J. (2023). Dean Oliver's Four Factors Revisited. arXiv:2305.13032.
Full study in Spanish, with every table, interval and the code that produced them.
Interactive simulator: move the four factors of two teams and watch win probability and
margin change, with the weights of each Spanish category.