Method
How basketball ELO ratings work.
A team starts at 1,500. After every game, its rating moves by an amount shaped by the opponent, the result, home court, and the score margin.
- Starting rating
- 1,500
- Default model
- Adjusted ELO
- Updated
- After every result
The chess idea, in three moves.
ELO does not reward every win equally. It first estimates what should happen, then moves each rating according to how surprising the actual result was.
- 1
Expect
The rating gap becomes an expected score. Equal players start at 50–50; a 400-point advantage is worth roughly 91%.
- 2
Score
Chess records 1 for a win, 0.5 for a draw, and 0 for a loss. Basketball uses 1 or 0 because rated games require a winner.
- 3
Update
The new rating is the old rating plus K × (actual − expected). An upset therefore moves more points than a routine win.
new rating = old rating + K × (actual − expected)Basic and Adjusted ELO answer slightly different questions.
Basic ELO
Looks only at who won, who lost, and how strong each team was beforehand. It is the clearest reference model.
- Win or loss
- Opponent strength
- No extra adjustments
Adjusted ELO
Adds the information in the final score. A convincing win can matter more than a narrow one, within a controlled limit.
- Everything in Basic
- Score margin
- Competition weight
“Default” simply means this is the model shown first in the public rankings. You can switch models there; their histories remain separate.
A 1,600 team hosts a 1,500 team.
Using the current adjusted model, the calculation starts like this:
1,600 + 70 − 1,500 = 170The home team receives a 70-point home-court adjustment.
1 ÷ (1 + 10^(-170 ÷ 400)) = 72.7%The classic ELO curve uses 400 as its probability scale. Substituting the 170-point gap produces 0.72683, rounded here to 72.7%.
20 × (1 − 0.72683) = +5.46If the favourite wins, it gains a modest 5.46 points before the score-margin adjustment.
A useful signal, not a verdict.
ELO describes what past results support right now. It does not know about injuries, line-ups, travel, motivation, or a tactical change that has not appeared in the results yet.
Every rating is tied to a versioned model, so an improved formula can be tested and rebuilt without silently rewriting an older history.