WAR, Explained
One number, five moving parts, and a set of assumptions nobody prints on the label.
The Standard take
WAR converts a player's batting, baserunning, defense, and position into runs, adds the runs a replacement-level player would not have produced, and divides by about ten runs per win. It is a good tier-sorter and a bad tiebreaker: the defensive input carries roughly a win of uncertainty, so two players within a win of each other are, for practical purposes, the same player.
Evidence: established methodology, contested inputs · Run the numbers yourself in the WAR Calculator.
The formula
WAR = (wRAA + Def + BsR + PosAdj + Replacement) / RunsPerWin
wRAA = ((wOBA − lgwOBA) / wOBAScale) × PA
PosAdj = positional run value × (PA / 600)
Replacement ≈ +20 runs × (PA / 600)
RunsPerWin ≈ 10
- wRAA — weighted runs above average
- How many runs the bat produced above a league-average hitter, given the same number of plate appearances. Everything hinges on wOBA, which weights each offensive event by how many runs it is actually worth instead of treating a single and a home run as one "hit" apiece.
- Def — defensive runs
- A fielding metric (DRS, UZR, or a tracking-based equivalent) expressed in runs saved relative to average at the same position. This is the noisiest term in the equation.
- BsR — baserunning runs
- Steals, extra bases taken, outs made on the bases. Small in magnitude — usually within five runs either way — but real.
- PosAdj — positional adjustment
- A credit for playing a hard position and a debit for playing an easy one. A catcher and a designated hitter with identical bats are not making identical contributions, and this term is the only place the model says so.
- Replacement
- The gap between "average" and "the guy you could sign this afternoon." Without it, every average player would grade at zero wins, which is obviously wrong — an average player is valuable.
The assumptions
- Ten runs equals one win. An approximation from the run-to-win relationship in Pythagorean expectation. It drifts with the league run environment; in a low-scoring era it is closer to nine.
- Replacement level is a convention. It is set so total league WAR reconciles with total league wins above a roughly 47-win baseline. It is chosen, not measured.
- Contributions are additive and independent. The model adds bat, glove, legs, and position as if they never interact. Real lineups and real defensive alignments interact constantly.
- Context is stripped on purpose. WAR does not care whether the home run came with the bases loaded in September. That is a design choice for predictiveness, and it is exactly why WAR sometimes disagrees with what you watched.
- Playing time is treated as earned. The model credits what happened on the field; it has no opinion about whether a player should have been out there.
A worked example
Take a center fielder with a .340 wOBA over 600 plate appearances, +5 defensive runs, +2 baserunning runs, in a league where average wOBA is .310 and the wOBA scale is 1.25.
| Component | Math | Runs |
|---|---|---|
| Batting (wRAA) | ((.340 − .310) / 1.25) × 600 | +14.4 |
| Defense | season DRS | +5.0 |
| Baserunning | season BsR | +2.0 |
| Position (CF) | +2.5 × (600 / 600) | +2.5 |
| Replacement | +20 × (600 / 600) | +20.0 |
| Total | sum | +43.9 |
| WAR | 43.9 / 10 | 4.4 |
4.4 wins is a clear All-Star season. Now move only the defensive term from +5 to −5 — a swing well inside the year-to-year noise of most fielding metrics — and the same player grades at 3.4. Same bat, same legs, one win of difference, entirely from the component we measure worst.
Where it breaks
- Small samples. Defensive metrics need something like a full season to stabilize. A 150-plate-appearance WAR is a rumor.
- Catchers. Pitch framing, blocking, and game management are either excluded or modeled very differently between systems. Catcher WAR is the least comparable number in the sport.
- Version mismatch. Quoting fWAR against bWAR in the same argument is comparing two different models. Pick one and say which.
- Tiebreaking. Awards debates decided by 0.2 WAR are decided by rounding error in a fielding estimate.
- Team totals. Summed team WAR tracks team wins well in aggregate and misses badly for individual clubs, because sequencing, bullpen leverage, and health are not in the model.
- Era and park. Our calculator does not park-adjust. Published WAR does. Do not compare the two directly.
FAQ
What does WAR actually measure?
WAR estimates how many more wins a team got from a player than it would have gotten from a freely available replacement — a minor-league call-up or waiver claim — given the same playing time.
Why do FanGraphs (fWAR) and Baseball-Reference (bWAR) disagree?
They use different pitching and defensive inputs. fWAR grades pitchers on FIP (strikeouts, walks, home runs), bWAR on runs actually allowed with a fielding adjustment. Disagreements of a win or more on the same pitcher are normal and are a feature of the models, not an error.
Is a 0.3 WAR difference meaningful?
No. Treat WAR as having an error bar of roughly a win, mostly from the defensive component. Use it to sort players into tiers, not to rank two players who sit within a win of each other.
How many wins is a replacement-level team?
About 47 to 48 wins over a 162-game season. That baseline is a modeling convention, chosen so that total league WAR reconciles with total league wins.
Does WAR work for pitchers the same way?
The currency is the same — runs converted to wins — but the inputs are different, and relievers are further adjusted for the leverage of the innings they pitch. Our calculator models position players only.
Can I use WAR to set a contract value?
Roughly. The market has paid something on the order of several million dollars per win in recent free-agent classes, but that rate moves year to year and is not uniform across ages or positions. Use it as a sanity check, never as a quote.