NBA Playoff Probability: Series Models and Assumptions

A constant-probability illustration

Assume independent games and an unchanged team win probability p in every game. The series-win probability is the sum of C(7,k) × p^k × (1-p)^(7-k), for k = 4, 5, 6, 7. Here C(7,k) counts ways to place k wins in seven hypothetical outcomes. Unplayed outcomes after a clinch are only a mathematical device: they cannot change who reached four wins first.

These figures are derived, not observed NBA results. For p above 50%, this model gives a higher series probability than the single-game probability. It does not establish that every higher seed has that advantage.

Sensitivity matters as much as the central estimate

Moving the assumed game probability from 55% to 65% moves the modeled series probability from about 60.8288% to 80.0154%: a range of 19.1866 percentage points. Even a precise calculation can be highly uncertain because its inputs are uncertain.

Real series may have different probabilities by venue, lineup, or game state. The NBA uses a 2-2-1-1-1 home-court format; see the NBA’s format explanation. The team assigned home-court advantage hosts Games 1, 2, 5, and 7 when needed. This scheduling fact supplies no universal point adjustment.

Stop the forecast when the series ends

Let F(w,l) be the chance of eventually winning the series from w wins and l losses. Set F(4,l) = 1 and F(w,4) = 0 for reachable clinched states. Before a clinch, F(w,l) = p_next × F(w+1,l) + (1-p_next) × F(w,l+1). If venue and history change the next-game chance, the model must track that information too.

At a hypothetical constant 60% chance for each remaining independent game, a team leading 3-1 needs at least one win in at most three games: 1 - 0.40³ = 93.6%. Trailing 1-3 requires three consecutive wins: 0.60³ = 21.6%. If a team clinches in Game 5, there is no Game 6 to forecast.

Make changing assumptions reproducible

For a venue-dependent forecast, list a separate assumed win probability for each scheduled game, together with its information cutoff. Work backward from clinched states using the recurrence. If an injury changes the probabilities, preserve the original forecast and document the replacement inputs. Averaging the home and away probabilities and inserting that average into the constant-game table generally gives a different model.

Keep price, evidence, and assumptions separate

A series moneyline converts to a break-even threshold, not a known true series probability. Compare like-for-like series prices and settlement terms; use the conversion guide to distinguish raw thresholds from normalized market estimates.

Rest, availability, and matchup adjustments belong in documented assumptions. This guide assigns no fixed rest bonus, momentum percentage, or automatic Game 2 covering effect. A Game 1 result can provide information, but one result alone cannot identify the cause of a price change or validate a new model.

Review predictions without treating a short run as proof

Record probabilities, prices, timestamps, and the information available before games. Evaluate calibration and results over additional observations, with uncertainty and dependence accounted for. A short playoff run or strong simulated standing cannot establish a durable forecasting advantage.

Continue with Why NBA Playoff Series Usually Favor the Better Team, NBA Spreads and Totals, or compare the single-elimination tournament path.

Frequently Asked Questions