Playoff Odds, Explained
One probability, one distribution, and an input nobody actually knows for certain.
The Standard take
Playoff odds are the upper tail of a binomial distribution: the chance a team wins at least the number of remaining games it still needs, at a per-game probability adjusted for opponent quality. The mathematics is settled and uncontroversial; the uncertainty lives almost entirely in the one input you have to estimate, which is how good the team truly is.
Evidence: established methodology, estimated inputs · Run it in the Playoff Odds calculator.
The formula
needed = max(0, cutoff − currentWins)
p = (t × (1 − s)) / (t × (1 − s) + s × (1 − t))
P(playoffs) = Σ from k = needed to n of C(n, k) × p^k × (1 − p)^(n − k)
projectedWins = currentWins + n × p
n = games remaining · t = true win% · s = opponent win%
- Wins needed
- The cutoff minus what is already banked. If the team has already cleared the cutoff, the probability is 1; if it needs more wins than it has games, the probability is 0.
- log5 adjustment
- Converts true talent into a win probability against the specific strength of the remaining schedule. Against exactly average opposition (s = .500) it returns t unchanged.
- The binomial tail
- The probability of every remaining season in which the team clears the bar, added up. Computed exactly in log space for 60 games or fewer, and by a normal approximation with a continuity correction above that.
- Projected wins
- The mean of that distribution — a single expected finish, useful for sanity-checking the cutoff you entered.
The assumptions
- Every game is independent. No momentum, no fatigue carryover, no rotation ordering. Independence is what makes the distribution binomial, and it is the assumption most often disputed.
- True talent is constant. One probability applies to every remaining game. Trades, injuries, and call-ups all violate this.
- Opponent quality is a single average. The schedule is compressed into one strength-of-schedule number rather than a list of opponents.
- The cutoff is known and fixed. In reality the cutoff is itself a random variable that depends on how the rest of the league plays.
- No ties or tiebreakers. Reaching the win total is treated as qualifying. Head-to-head records and division structure are not modeled.
A worked example
A team sits at 52-40 with 70 games left, a true talent estimate of .550, remaining opponents averaging .510, and a playoff cutoff of 88 wins.
| Step | Math | Value |
|---|---|---|
| Adjusted p | (.550 × .490) / (.550 × .490 + .510 × .450) | .540 |
| Wins needed | 88 − 52 | 36 |
| Expected wins left | 70 × .540 | 37.8 |
| Projected finish | 52 + 37.8 | 89.8 |
| P(≥ 36 of 70) | binomial upper tail | ≈ 68% |
Note how thin the margin is. The team projects to clear the cutoff by under two wins, and that translates to roughly a two-in-three chance — not a lock. Drop true talent from .550 to .520 and the same record falls under a coin flip.
Where it breaks
- Garbage true-talent input. Enter this season's winning percentage as true talent and you will hand a hot team odds it has not earned. This is the single largest source of error.
- A guessed cutoff. Moving the cutoff by two wins can move the probability by twenty points. The tool does not estimate it for you.
- Late-season roster change. Deadline trades and injuries break the constant-talent assumption exactly when the odds matter most.
- Tiebreakers and structure. Division races, wild-card slots, and head-to-head rules decide real seasons and are absent here.
- Very few games left. With a handful of games remaining, the specific opponents matter more than any average can express.
FAQ
How are playoff odds actually calculated?
Treat every remaining game as an independent coin flip weighted by the team's true win probability. The chance of reaching a win cutoff is then the upper tail of a binomial distribution: the summed probability of every outcome in which the team wins at least as many games as it needs.
What is true win percentage, and how is it different from the current record?
The record is what happened; true win percentage is the best estimate of talent going forward. A 20-5 start does not mean an 80% team. Run differential, underlying rates, and preseason projections are all better guides than the record itself.
What does the log5 adjustment do?
It converts a team's talent into a win probability against a specific quality of opponent. A .550 team facing .510 opposition is not a .550 team in those games — log5 returns .540, and that difference compounds over a long remaining schedule.
Why does the model switch between exact and approximate math?
With 60 or fewer games remaining, the binomial tail is summed exactly. Above that, a normal approximation with a continuity correction is used for speed. The two agree to well within a percentage point in the ranges that matter.
Why do published playoff odds differ from this number?
Full-scale models simulate the entire league, game by game, with opponent-specific ratings, tiebreakers, and division and wild-card structure. This tool models one team against an average opponent and a fixed win cutoff.
Is a 5% playoff chance basically zero?
No. One season in twenty ends with that team playing in October. Small probabilities are the ones people misread most often, in both directions.