Understanding Basketball's Four Factors: A Deep Dive on the Metrics That Actually Win Games
Every basketball fan has an intuition about what wins games. Points, obviously. Maybe rebounding. Perhaps a superstar taking over in the fourth quarter. These intuitions aren't wrong, but they're incomplete. In 2004, statistician Dean Oliver published Basketball on Paper and gave the analytics world something far more useful: a framework that strips the game down to four fundamental pillars that explain, with remarkable precision, why teams win and lose.
We call them the Four Factors. After training our XGBoost_Top17 model on five seasons of NBA data, we can say with confidence that these metrics remain relevant today — but their relative importance is not what Oliver originally proposed, and not what most analysts assume. Here is what they are, why they matter and what we actually learned about how they behave in the modern game.
The Four Factors, Explained
Oliver's insight was elegant: basketball games are decided by four things and four things only. How efficiently you shoot. How often you turn the ball over. How well you rebound your misses. And how many free throws you generate. Everything else — pace, isolation scoring, switching defenses, load management — ultimately flows through these four channels.
Effective Field Goal Percentage (eFG%)
eFG% accounts for the fact that a three-pointer is worth 50% more than a two.
eFG% = (FGM + 0.5 × 3PM) / FGA
Oliver assigned this the highest weight of the four factors — roughly 40% of offensive performance. The intuition is sound: shooting efficiency is the most direct path to points. In the modern pace-and-space era, the gap between a team shooting 52% eFG% and one shooting 48% compounds quickly over 100 possessions.
However, here is where our model diverges from conventional wisdom: eFG% differential did not survive feature selection into our Top17. After testing 266 candidate features, eFG% was filtered out during regularization. This does not mean shooting efficiency is unimportant — it means its signal is already captured by more holistic metrics. Specifically, PIE (Player Impact Estimate), which holds the top two spots in our feature importance rankings at a combined 37.9%, encapsulates shooting efficiency alongside turnovers, rebounds and free throws in a single integrated metric.
When you see a High Confidence pick on DataProven, you are not seeing a raw eFG% comparison. You are seeing a PIE differential that already has shooting baked in.
Turnover Rate (TOV%)
TOV% measures what percentage of possessions end in a turnover.
TOV% = TOV / (FGA + 0.44 × FTA + TOV)
Oliver weighted this at roughly 25% of offensive performance. Our model tells a different story: away team turnover rate over the last 10 games (away_tov_pct_l10) is the highest-ranked standalone Four Factor feature in our model at 4.1% importance — making it more predictive than any other individual Four Factor metric.
The road environment amplifies turnovers in ways that home games do not. A visiting team that has been sloppy for two weeks is not going to suddenly tighten up in a hostile arena. Our L10 window captures this behavioural pattern: if a team has been turning the ball over on 18%+ of possessions over their last ten games, that is a habit, not a fluke, and it is the single clearest warning sign we have for road losses.
Offensive Rebound Percentage (ORB%)
ORB% captures how often a team recovers its own misses.
ORB% = ORB / (ORB + Opponent DRB)
This is the most tactically contested factor in the modern era. Many teams deliberately sacrifice offensive rebounding to protect against transition defence — a calculated trade-off that makes ORB% an asymmetric signal depending on team philosophy.
In our model, away team offensive rebound rate over the last 5 games (away_oreb_pct_l5) ranks 9th at 3.5% importance, while the defensive side — dreb_pct_l10_diff — ranks 13th at 2.6%. The asymmetry is meaningful: a road team crashing the glass aggressively is a high-signal indicator of energy and motor. Defensive rebounding, by contrast, is a baseline fundamental — important, but expected.
Free Throw Rate (FTR)
FTR is calculated as FTA / FGA, reflecting how often a team earns trips to the line.
FTR = FTA / FGA
Oliver assigned this the lowest weight of the four at roughly 15%. Our model largely agrees, though with a nuance: free throw features appear twice in our Top17, but in a specific form. home_fta_rate_l10 (16th, 1.8%) captures the home team's recent ability to draw fouls, and away_ftm_l10 (17th, 1.8%) captures free throws actually made by the road team — not just attempted.
Free throw rate behaves more asymmetrically than Oliver's framework implies. A team with an elite FTR — say, above 0.35 — gains a meaningful edge. But the difference between average and slightly below-average FTR is nearly negligible in our predictions. The factor matters most at the extremes, and this is why it ranks last among the Four Factors in our model.
What the Weights Actually Look Like
Oliver's original proposed weights were roughly 40% for shooting, 25% for turnovers, 20% for rebounding, and 15% for free throws — on both offense and defense.
The most important divergence from Oliver's framework is that shooting efficiency, while genuinely important, is not a separable signal at the game-prediction level once you control for overall team quality via PIE. Two teams with identical PIE differentials but different eFG% splits are essentially equally matched in our model's eyes — the efficiency has already been priced in.
What the model adds on top of PIE is context: how a team is performing right now (pie_l10_diff at 17.7%), whether the road team is protecting the ball, whether either team is getting to the line, and whether rest or schedule creates an asymmetric edge.
The Hierarchy Our Model Actually Uses
Understanding the Four Factors is useful. But understanding how they interact with our full feature set is more useful still. Here is the actual priority order our model assigns when generating predictions:
1. Overall Efficiency Gap (37.9% combined) PIE differential over the full season and last 10 games. This is the foundation. Before any Four Factor analysis, the holistic efficiency gap between the two teams is the single largest predictor of outcome.
2. Team Quality Baseline (15.6% combined) Win/loss differentials and season record. Strong teams do not suddenly become weak — season-long quality creates a floor.
3. Home Advantage Calibration (8.0%) Estimated net rating adjusted for strength of schedule. Home court is real but it is a setting, not a guarantee.
4. Recent Offensive Form (7.1%) Offensive rating differential over the last 10 games. Captures current trajectory independent of season averages.
5. Four Factors and Discipline (15.9% combined) Turnovers, rebounds, free throws and personal fouls — in that order of importance. Ball security is the highest-ranked individual Four Factor. Shooting efficiency is embedded in PIE above.
The most common analyst mistake is leading with eFG% when evaluating a matchup. Our model suggests this is backwards. Start with the PIE differential — it already contains the shooting signal. Then ask whether the road team has been protecting the ball over their last ten games. Those two checks will cover more predictive ground than any individual Four Factor in isolation.
The Defensive Side of the Ledger
Oliver's critical insight was that each factor has an offensive and a defensive component. Winning the eFG% battle requires both shooting well and preventing efficient shots. Winning the rebounding battle requires limiting opponent offensive boards, not just generating your own.
In our model, the defensive versions of these factors carry slightly different weights than their offensive counterparts. Defensive rebounding rate (dreb_pct_l10_diff) ranks 13th at 2.6% — important, but below the away team's offensive rebounding rate. The implication is that controlling your own defensive glass is a baseline expectation; what surprises the model is a road team that crashes offensively against the grain of modern NBA philosophy.
Personal foul differential (pf_l10_diff) ranks 15th at 2.0%. This is Oliver's framework extended: teams in foul trouble give opponents free points and force stars to the bench. Over ten games, a team that has been consistently fouling more than its opponent is signalling defensive tardiness — rotations arriving a step late, contests turning into fouls. It is a subtle indicator of a team playing reactive rather than proactive defence.
What the Four Factors Miss
No framework is complete, and Oliver acknowledged the limitations. The Four Factors describe what happened in a possession framework — they do not capture shot quality directly. Two teams can have identical eFG% but radically different offensive processes: one generating open corner threes off ball movement, the other forcing mid-range pull-ups under pressure. The former is more sustainable.
This is why our model supplements the Four Factors with context that Oliver's framework cannot see: rest and fatigue data (b2b_diff at 3.8%, away_optimal_rest at 3.2%), and head-to-head history (h2h_dominance at 2.9%). Some matchups are tactically misaligned in ways that season-long Four Factor averages cannot detect.
The Four Factors say nothing about clutch performance, officiating tendencies or the specific matchup dynamics of a given game. These are real factors that create variance in individual outcomes — which is precisely why even an excellent model cannot exceed roughly 70–75% accuracy. Some unpredictability is intrinsic to basketball. That is what makes it worth watching.
Using the Four Factors as a DataProven Subscriber
When you view our predictions, you are seeing the downstream output of a model that treats PIE differential as its foundation and Four Factor rolling metrics as context layers on top. A High Confidence pick typically reflects a team that holds a meaningful PIE advantage, is protecting the ball over recent games, and benefits from favourable rest and home court conditions.
A Low Confidence pick often involves teams that are competitive in PIE but where Four Factor signals are mixed — one team turning the ball over more, the other getting to the line more — creating genuine uncertainty that the model cannot resolve with high confidence.
Understanding the framework does not just help you consume our predictions — it helps you evaluate them critically, which is exactly the kind of informed audience we are trying to build.
Dean Oliver gave basketball a language for its fundamentals. Twenty years later, that language is still the clearest way to explain why teams win. But the data adds a footnote: the hierarchy is not what he thought and shooting efficiency — while real — is already embedded in the metrics that matter most.
Next up: Kelly Criterion and bankroll management — how to size your bets when you have a genuine edge.