The Non-League Ranking applies the ELO rating system to rank every club competing at Steps 1 to 4 of the English non-league pyramid — from the National League down to the fourth tier of non-league football.
The ELO rating system gives every club a single numerical value — their ELO score — that represents their current strength relative to all other clubs in the ranking. After every match, points are transferred from the losing side to the winning side. The amount transferred depends entirely on the difference in ratings between the two clubs: beating a much stronger opponent earns far more than beating a weaker one, and losing to a weaker side costs far more than losing to a stronger one.
Draws are treated as half a win and half a loss (value 0.5). The higher-rated club loses a small number of points and the lower-rated club gains the same amount, reflecting the outcome's deviation from expectation.
Matches that are cancelled, abandoned, or forfeited are excluded entirely — no points are exchanged unless the game is completed.
Before each match, a win probability is calculated for the home team based on the difference in ELO ratings between the two clubs. This probability drives how many points are exchanged after the result.
A club with a 100-point ELO advantage carries a win probability of approximately 0.64, while their opponent sits at 0.36. The larger the gap, the more extreme the probabilities become in either direction.
Once the result is known, the difference between the actual outcome and the expected win probability is multiplied by the K-factor (set to 18 in this ranking). This product is the number of points won by one club and lost by the other — always equal in magnitude.
Where Actual Result = 1 (win), 0.5 (draw), or 0 (loss).
To illustrate how the system works in practice, consider two clubs with ELO ratings of 1500 and 1400 meeting on a neutral ground.
Home teams consistently win more often than their raw ELO rating would suggest. To account for this, the home team is awarded a temporary ELO bonus for the purpose of calculating the win probability only — their actual rating is never altered by this adjustment.
Because clubs with larger crowds tend to benefit more from home support, the bonus scales with the team's current ELO rating:
For a club rated 1500, this gives a home bonus of 40.5 points, effectively making the matchup feel like a 1540.5 vs 1400 contest for the purposes of win probability.
When Team C (rated 1500) hosts Team D (rated 1400), the adjusted ELO difference is 140.5 rather than 100, shifting the probabilities from 0.64 / 0.36 to approximately 0.69 / 0.31. The same K-factor calculation is then applied to these adjusted probabilities.
If the roles were reversed — Team D hosting Team C — the adjusted difference narrows to just 62 points, giving win probabilities of roughly 0.59 / 0.41, a considerably more balanced contest.
Every fixture on the site displays a predicted probability for a home win, draw, and away win. These figures are derived entirely from the two clubs' ELO ratings — including the home advantage adjustment where applicable — and are calibrated against the full historical match database.
The starting point is the draw. Rather than treating it as a leftover once wins are accounted for, the draw probability is calculated first as a function of how evenly matched the two sides are. The closer the contest, the more likely a draw; the more one-sided the matchup, the rarer a draw becomes.
This relationship follows a bell-curve shape, peaking at around 22–23% when the two clubs are equally matched and falling toward zero as the ELO gap widens.
Where WinProp is the home team's win probability after applying the home advantage adjustment. At a 50/50 contest (WinProp = 0.5), this produces a draw probability of 22.75%. At WinProp = 0.7 (a clear home favourite), draw probability falls to roughly 19%.
Once the draw share is set aside, the remaining percentage is split between a home win and an away win. Rather than dividing it in simple proportion to WinProp, a small tilt is applied to reflect an empirical pattern found in the data: as the ELO gap widens, the stronger side's win probability grows slightly faster than a pure proportional split would imply.
The tilt term — 0.40 × (WinProp − 0.5)² — is zero when teams are perfectly equal and grows symmetrically as one club becomes a heavier favourite. The away win probability takes whatever is left: 100 − P(Home) − P(Draw).
The two coefficients — 0.91 (draw curve) and 0.40 (win tilt) — were fitted by minimising prediction error across the full match database. Tested against over 80,000 historical results, the model's errors for home win, draw, and away win each stay within approximately ±2 percentage points across all ELO bands.
With home advantage applied, Team A's adjusted ELO advantage is ~140 points, giving WinProp ≈ 0.69. Draw = 0.91 × 0.69 × 0.31 × 100 ≈ 19%. The remaining 81% splits with a slight tilt toward the home favourite: ~58% home, ~23% away.
For the head-to-head comparison page, the same model is applied on a neutral ground basis — home advantage is removed so the prediction reflects the clubs' underlying ELO ratings alone.
Each league view includes a predicted final table, built by summing a club's actual points to date with the expected points from every remaining fixture — including any postponed matches that have yet to be replayed.
For each unplayed fixture the model calculates:
E[pts] = P(win) × 3 + P(draw) × 1
These are summed across all remaining games per club and added to their current points tally. The result is sorted to produce the predicted final standings. Sorting uses the full decimal expected value — not a rounded integer — so that fine margins between closely ranked clubs are preserved.
Because a team can only score 0, 1, or 3 points from any single game, the expected value (e.g. 1.8) will never exactly match any achievable outcome. To give context to the prediction, each club's row in the table also shows a points range — the minimum (lose all remaining games) and maximum (win all remaining games) totals physically possible. A club the model expects to finish on 51 points but whose range spans 44–65 is in a much less settled position than one whose range is 67–70.
When five or fewer games remain, predicted points are shown to one decimal place to reflect the precision that matters at that stage of the season.
When the ranking was established at the start of the 2014/15 season, clubs received a starting ELO score based on the step of the pyramid they competed in. This created an average gap of 200 points between adjacent levels — equivalent to a win probability of approximately 0.76 for the higher-ranked side.
| Level | Examples | Starting ELO |
|---|---|---|
| Step 1 | National League | 1800 |
| Step 2 | Nat. League North / South | 1600 |
| Step 3 | NPL / Southern / Isthmian Premier | 1400 |
| Step 4 | NPL / Southern / Isthmian Div 1 | 1200 |
Each season, clubs join the ranking through promotion from Step 5 or relegation from the English Football League. Rather than simply assigning a flat starting score, new entrants receive an initial rating based on two factors: their own recent performance (finishing position and results in their promotion or relegation season), and how clubs from the same Step 5 league have historically performed after entering the ranking. This approach gives a more accurate picture of a club's likely strength from day one.
Non-league clubs occasionally face opponents from the Premier League, EFL, or Step 5 and 6 in the FA Cup. Since those clubs are not part of the Non-League Ranking, they are assigned representative ELO ratings that reflect their relative strength. These ratings are reviewed and adjusted each season to reflect recent inter-level results.
| Level | Representative ELO |
|---|---|
| Premier League | 2200 |
| Championship | 2100 |
| League 1 | 2000 |
| League 2 | 1900 |
| Step 5 | 1100 |
| Step 6 | 1000 |
Standard ELO treats every win as 1 and every loss as 0. However, in cup football a team that battles through 90 minutes level before losing on penalties has performed quite differently to a side that concedes in normal time. To reflect this more fairly, cup results decided in extra time or on penalties use adjusted outcome values:
This means a team that loses on penalties in the FA Cup still gains (or loses fewer) ELO points than if they had been beaten in 90 minutes — rewarding the performance without over-rewarding the winner.
The ELO system was originally developed to rate chess players and has since been adapted for many competitive sports, including FIFA's international football rankings. The links below offer more background on the system and other football applications.