How the odds work

As of 29 September 2026. Women's Super League, NWSL, Northern Super League and USL Super League.

What the numbers are

Estimates of how the season will finish. Each day, after the latest results, we play every remaining match 100,000 times and count how often each club finishes in each place. A figure like 40% means the club finished there in about 40,000 of those seasons.

Proven results come first

Separately, we check every possible combination of remaining results. When that proves a club has clinched a place, or can no longer reach it, the Odds tab says Clinched or Out instead of a percentage, exactly as the Scenarios tab does. A simulation can put a number on a possibility but can't prove one, so an estimate is never shown as 0% or 100%: the smallest and largest are "<1%" and ">99%".

The WSL model

• Each club gets an attacking and a defensive strength, estimated from the last two years of league results. Recent matches count for more than older ones.
• Those strengths, plus home advantage, give the chance of every scoreline in a match (a Dixon-Coles model). Scorelines matter because tables are split on goal difference and goals scored.
• A club new to the league (just promoted, or back after time away) starts from the strength of the league's weaker clubs, not its average, and its own results take over as it plays.
• Each simulated season is ranked by the league's own rules, including its tiebreakers and any points deductions.

How we checked the WSL model

We rewound every finished WSL season since 2017-18 to its start and to a quarter, half and three quarters of the way through, forecast the rest using only what was known then, and scored the forecasts against what happened. Settings were chosen on the other seasons each time, so no season was used to tune its own forecast.

• Match forecasts (ranked probability score, lower is better): 0.171 for this model, against 0.182 for a standard club Elo rating, 0.244 for the league's average home/draw/away rates, and 0.250 for treating every result as equally likely.
• Calibration: things we gave about a 20% chance happened about 20% of the time, and so on across the range, within what chance alone would explain.
• Before the 2026-27 season we compared ours with the Opta supercomputer's published forecast. We named the same three title contenders; we're more confident than Opta at both ends of the table.

The NWSL model is different

The NWSL is so evenly matched that the WSL model doesn't beat a plain club Elo rating there. So the NWSL model borrows from both:

• An Elo rating, updated after every result, decides how likely each side is to win.
• The WSL-style model above decides how likely a draw is, and which scorelines are likely, because tables are split on goal difference.

We checked it the same way, on nine NWSL seasons (2016 to 2019 and 2021 to 2025). Its match forecasts were the best of every model we tried: 0.225, against 0.226 for Elo on its own and 0.235 for the league's average rates. But it only matches Elo; it doesn't clearly beat it. We publish it anyway, because it is the best we have and, unlike Elo alone, it can play out a whole season.

Treat the NWSL odds with more caution than the WSL's. Its season forecasts were mostly well calibrated, but one range was off: things it gave a 60–70% chance happened about half the time (41 cases). That may be chance, since some range usually misses when there are ten of them, but it may also mean the model is too sure of itself in the middle of the range.

Leagues with little history

The Northern Super League and USL Super League have too few finished seasons to test a model on, so their odds rest on limited data, and their Odds tabs say so. They use the WSL-style model described above, with settings chosen on the NWSL. On each league's one finished season we could check, it forecast matches better than the NWSL model and better than Elo, but one season proves little.

What they don't know

Only results go in. Neither model knows about injuries, transfers, suspensions, a new manager, fixture congestion or chance quality (expected goals). Both will be slow to react when a club changes quickly.

Data credit

The NWSL model was checked, and its settings chosen, partly on NWSL seasons from 2016 to 2019 provided by American Soccer Analysis.