How Basketlog calculates its NBA predictions

In short

For every game, Basketlog estimates the expected point margin between the two teams, then turns it into a win probability. The margin starts from each team's real strength, then adds home-court advantage, fatigue and injuries. Backtested on the 2025-26 season (1322 games), the model picked the right winner in 68% of cases, and in 81% of cases when it gave the favorite at least 70%. This season (2026-27), it stands at 66% over 68 completed games.

1. Each team's strength

The foundation is the average point differential per game: points scored minus points allowed. It is one of the most reliable indicators of an NBA team's real strength, more telling than its win-loss record. Early in the season, a handful of games isn't enough, so the model blends the current season with the previous one, which counts as 12 extra games. The further the season goes, the more the current season weighs. Finally, this differential is pulled 15% toward the average, because very good and very bad starts tend to even out.

2. Home-court advantage

Playing at home is worth about 2.3 points in today's NBA (and 3.2 points in the EuroLeague, where arenas matter more). This value is added to the home team.

3. Back-to-back fatigue

A team that played the night before loses around 1.6 points on average. Every night, the model checks which teams are playing two games in two days.

4. Injuries

The injury report is checked before every game. A player ruled out takes about 0.9 points away from his team and a questionable player about 0.35 points, capped at 4.5 points to avoid overreacting. Injuries are only factored in for upcoming games.

5. From margin to probability

The expected margin is never spot on: in the NBA, the final score misses it by about a dozen points on average. So the model uses a normal distribution centered on the expected margin, with a standard deviation of 12.5 points. From it, the model derives each team's win probability, the chances of a close game (a margin of 5 points or fewer) and of a blowout (10 points or more). The projected score combines each team's offense with its opponent's defense.

6. Verified results, no peeking at the future

Every prediction is compared with the final score on the Model track record page. For past games, the model is recalculated using only the results known before the game: it never sees the future. Probabilities remain estimates: a 70% favorite loses about three times out of ten. Basketlog is an analysis tool, not a betting service.

See today's NBA predictions · EuroLeague predictions · Track record