Multiple regression, delta method
What scores runs: OBP or SLG?
Thirty-odd team seasons across four eras, one regression, and the old Moneyball question with error bars.
Computed on request from the record book (Oct 13, 2026) and 720 finals in the game log.
The finding
Across 36 team seasons (2003, 2010, 2025, 2026), runs per game are described well by on-base and slugging together: the model accounts for 83% of the spread and typically misses by 0.25 runs a game. A point of OBP is worth 1.88 times a point of SLG, but the 95% interval runs from 0.29 to 3.47 and includes 1 — the data cannot rule out that the two are worth the same.
In season terms, ten points of OBP (.010) are worth about 27 runs over 144 games, ten points of SLG about 15. Comparing clubs only with others in their own season — a separate baseline for each year absorbs the era — the ratio is 1.21 (0.60–1.81), again not clearly different from 1. Measured against the spread each actually shows between clubs, they matter about equally — standardised weights of 0.47 for OBP and 0.50 for SLG — because slugging varies more from club to club. On its own, OPS explains the most (82%).
OBP ÷ SLG weight
1.88×
95% interval
0.3–3.5
Model r²
0.83
n = 36 team seasons
Typical miss
0.25
runs per game
Small sample. 36 team seasons, with OBP and SLG correlated at r = 0.75: the two coefficients trade off against each other, which is why the ratio's interval is wide.
How much more is a point of OBP worth?
The ratio of the two regression weights, with 95% intervals. Right of the dashed line, on-base is worth more than slugging.
Predicted against actual runs per game
Each dot is one club's season. On the dashed line, the model is exactly right.
Share of the spread each measure explains
r² of runs per game on each measure alone, and on OBP and SLG together.
Method
R/G = β₀ + β₁·OBP + β₂·SLG + ε
Ordinary least squares on every club's season batting line in the four snapshots on disk. OBP and SLG are recomputed from the counting stats rather than read from the rounded columns. Coefficients: OBP 19.1 (95% 10.2–27.9, p < 0.001), SLG 10.2 (5.7–14.6, p < 0.001); adjusted r² 0.82.
The ratio's interval uses the delta method, which carries the covariance between the two coefficients: Var(β₁/β₂) ≈ (β₁/β₂)²·[Var β₁/β₁² + Var β₂/β₂² − 2·Cov/(β₁β₂)]. The within-season model adds an indicator for each season after the first. Variance inflation from the OBP–SLG correlation is 2.3. OPS alone reaches r² 0.82.
Team seasons, not players: this says how a lineup's on-base and slugging turn into runs, not what any single hitter is worth. The current season is incomplete, the 2003 and 2010 leagues had eight clubs, and no park adjustment is possible. With OBP and SLG this correlated, the individual weights are fragile even when the fit is good — the classic estimate that a point of OBP is worth about 1.8 points of SLG is well inside these intervals, but so is parity.