xG or FDR: which predicts FPL points better?
Published 9 Oct 2026 · data to end of 2025/26
Short answer
It depends on the job. To choose between players, an xG fixture rating beats FDR, because it knows how strong your own team is and FDR doesn’t. To pick a player’s best weeks, FDR did as well as every xG rating we built and better than a 10-gameweek one. Market-implied team goals beat both at both jobs.
FPL’s fixture difficulty rating (FDR) gives every fixture a number from 1 (easiest) to 5 (hardest). Many managers now plan with expected goals (xG) instead: how many good chances a team has been creating and conceding. We tested both against what actually happened: 29,334 Premier League starts from 2022/23 to 2025/26, scored with FPL’s own points.
The answer depends on which decision you are making. FDR rates the opponent, not your player’s team. That makes it a weak guide for choosing between players, but a sound one for choosing when to play the same player.
- starts tested, 2022/23 to 2025/26
- 29,334
- starts tested, 2022/23 to 2025/26
- how often FDR was right when it and a 10-GW xG rating disagreed about a player’s weeks
- 55.0%
- how often FDR was right when it and a 10-GW xG rating disagreed about a player’s weeks
- percentage points the xG rating gained over FDR when choosing between players
- +1.9
- percentage points the xG rating gained over FDR when choosing between players
- of these starts had a different FDR by the end of the season
- 24.9%
- of these starts had a different FDR by the end of the season
How we scored each number
Take two starts with different FPL points. A predictor is “right” if it rated the higher scorer’s fixture or player as the better one. Over hundreds of thousands of pairs, 50% is a coin flip and 100% is perfect. One match is noisy, so even the best number here is right about 60% of the time. Where a predictor rates two starts the same, as FDR often does, it gets half a point.
We asked two questions, because FPL asks both:
- Between players: two starters in the same gameweek and position. Which one will score more?
- Same player: two of one player’s starts in the same season. Which week will be better? This is the question behind timing a transfer, a bench decision or a chip.
Our xG fixture rating is the team’s xG per match over its last 10 gameweeks, times the opponent’s xG conceded per match, adjusted for home and away. Defenders and goalkeepers get the mirror image, the goals they are expected to concede. For scale we also scored market-implied team goals (the goals that closing betting prices imply for each team) and three player numbers: xG plus xA per 90 minutes, points per 90 and price. Every input uses only matches played before the gameweek. Every FDR is the one FPL was showing before kick-off, not today’s.
Choosing between players: xG beats FDR
Bars start at 50%, a coin flip. FDR is the weakest of the six.
Show as a table
| Pairs right | |
|---|---|
| Market team goals | 59.9% |
| xG fixture rating | 57.5% |
| FPL price | 57.0% |
| Player points per 90 | 56.1% |
| Player xGI per 90 | 55.9% |
| FDR | 55.6% |
FDR got 55.6% of pairs right and the xG fixture rating 57.5%, a gain of 1.9 percentage points (95% interval +1.1 to +2.7). FDR scored worst of the six numbers, below even price.
The reason is what FDR leaves out. Two defenders both facing an FDR 3 fixture get the same rating, even if one plays for the best defence in the league and the other for the worst. The xG rating includes both teams, so it separates them. The market does the same job with better information and gained 4.3 percentage points on FDR.
By position, the fixture matters most for defenders, where a clean sheet is most of the return. For midfielders, the player matters as much as the fixture: xG plus xA per 90 (58.7%) beat both the xG fixture rating and FDR.
| Pos. | FDR | xG | Market | xGI/90 |
|---|---|---|---|---|
| GK | 54.6% | 55.3% | 57.0% | 51.7% |
| DEF | 56.8% | 58.7% | 61.6% | 52.7% |
| MID | 54.7% | 56.8% | 58.9% | 58.7% |
| FWD | 54.7% | 56.3% | 57.7% | 55.6% |
Timing the same player: FDR holds its own
Once the player is fixed, his own team’s strength drops out. What is left is the opponent and home or away, and FDR rates exactly that. Here the 10-gameweek xG rating did worse than FDR: 54.4% against 55.8%.
That surprised us, so we tried four more xG ratings: longer windows (20 and 38 gameweeks) and versions that look only at the opponent. All of them are shown below. None beat FDR. The best one, the opponent’s xG over a full 38 gameweeks, drew level. The 10-gameweek version was chosen before the analysis ran; the others were added after it lost.
Bars start at 50%. No xG rating beat FDR; market team goals did.
Show as a table
| Pairs right | |
|---|---|
| Market goals (Market team goals) | 57.0% |
| Opp. xG, 38 GWs (xG, opponent only, 38 GWs) | 55.8% |
| FDR (FDR) | 55.8% |
| xG, 20 GWs (xG rating, 20 GWs) | 55.3% |
| Opp. xG, 10 GWs (xG, opponent only, 10 GWs) | 55.3% |
| xG, 38 GWs (xG rating, 38 GWs) | 55.2% |
| xG, 10 GWs (xG rating, 10 GWs) | 54.4% |
“Opp. xG” looks only at the opponent: what it concedes (for attackers) or creates (for defenders). Differences are in percentage points.
| Rating | % right | v FDR | 95% range |
|---|---|---|---|
| Market goals | 57.0 | +1.3 | +1.0 to +1.6 |
| Opp. xG, 38 GWs | 55.8 | +0.1 | −0.4 to +0.5 |
| xG, 20 GWs | 55.3 | −0.5 | −1.0 to 0.0 |
| Opp. xG, 10 GWs | 55.3 | −0.5 | −1.0 to 0.0 |
| xG, 38 GWs | 55.2 | −0.6 | −1.0 to −0.1 |
| xG, 10 GWs | 54.4 | −1.3 | −1.9 to −0.8 |
The disagreements tell the same story. In 47,399 pairs of a player’s weeks, FDR and the 10-gameweek xG rating pointed opposite ways. FDR was right in 55.0% of them.
FDR’s real weakness is coarseness. With only five levels, it rates 32.6% of a player’s pairs of weeks the same. When it does separate two weeks, it is right 58.5% of the time. Most of the market’s edge comes from separating weeks FDR calls equal. When the market and FDR actually disagreed, the market was right only 51.4% of the time, barely better than a coin flip.
Why does a rolling xG rating struggle? Ten gameweeks is a small sample. A team’s xG over that window moves with injuries, red cards, game state and the opponents it happened to face. Longer windows help, which fits that explanation. FPL doesn’t publish how it sets FDR, so we can’t say what it uses instead. But it clearly isn’t just recent xG.
| Pos. | FDR | xG (10 GWs) | Market |
|---|---|---|---|
| GK | 54.6% | 52.9% | 54.8% |
| DEF | 57.0% | 55.5% | 58.8% |
| MID | 54.9% | 53.9% | 56.1% |
| FWD | 55.6% | 54.3% | 56.2% |
| Season | FDR | xG (10 GWs) | Market |
|---|---|---|---|
| 2022/23 | 55.9% | 51.8% | 57.6% |
| 2023/24 | 57.0% | 57.1% | 58.8% |
| 2024/25 | 55.3% | 54.8% | 56.4% |
| 2025/26 | 54.9% | 53.5% | 55.6% |
FDR beat the 10-gameweek xG rating in three seasons out of four; in 2023/24 the two were level. 2022/23 was the xG rating’s worst season, partly because FPL has no xG for 2021/22, so its early-season windows were short.
How much a fixture is worth
Whichever number you use, the fixture matters most for defenders. A defender averaged 3.5 points per start at FDR 1–2 and 1.8 at FDR 5. The market spreads players further apart because it also knows the team. Defenders in the easiest fifth of fixtures by the market averaged 4.2, and those in the hardest fifth 1.8.
| FDR | GK | DEF | MID | FWD |
|---|---|---|---|---|
| 1–2 | 3.72 | 3.49 | 3.88 | 4.61 |
| 3 | 3.31 | 3.01 | 3.55 | 4.32 |
| 4 | 3.10 | 2.49 | 3.20 | 3.35 |
| 5 | 2.67 | 1.85 | 2.67 | 3.44 |
FDR 1 is rare (under 70 starts per position in four seasons), so it is merged with 2. Forwards at FDR 5 scored slightly more than at FDR 4, from only 172 starts.
The two measures mostly agree on which fixtures are hard. For midfielders’ starts, the xG rating’s expected team goals fell steadily from 1.89 at FDR 1 to 0.93 at FDR 5. They differ at the margins, and that is where the tests above are decided.
FDR changes during the season
FPL revises FDR as the season goes on. By the end of each season, between 79 and 255 of its 380 fixtures had a different rating from the one shown before kick-off. The end-of-season ratings score better: 56.6% against 55.8% on timing. But no manager could have used them, so a test that uses today’s ratings for past matches flatters FDR. Here every FDR is the one that was live at the time. The same applies to planning: the FDR you see in August for a match in March is provisional.
What to use
- Choosing between two players: don’t lean on FDR. It can’t see that one team is far stronger than the other. Use something that rates both teams, and for midfielders weigh the player’s own xG and xA at least as heavily as the fixture.
- Timing one player’s weeks: FDR is a fair guide. Don’t swap it for a short-window xG table. If you use xG, use a long window. FDR rates a third of a player’s pairs of weeks the same, so break those ties with a finer number.
- For both: market-implied team goals were the best fixture number we tested. FixtureShark’s match projections use them for each team’s goals where a price exists.
- Before any of it: every number here assumes the player starts. Neither FDR nor xG tells you that; check minutes first.
Fixtures matter even for the very best players. For how much they moved Haaland’s returns home and away, see should you always captain Haaland.
How this was worked out
Sample. Every Premier League start from 2022/23 to 2025/26 (33,495 starts) with FPL’s own points. A start is FPL’s start flag; early 2022/23 rows, before FPL recorded starts, count 45+ minutes. We dropped 4,161 starts that lacked five earlier matches for either team or 270 earlier minutes for the player, which leaves 29,334. Every predictor is scored on the same starts. 2021/22 is excluded because FPL has no xG for it.
FDR before kick-off. We only store FDR for the current season, so past FDRs come from the public FPL archive (github.com/vaastav/Fantasy-Premier-League). Its fixtures file was saved 38 times in 2022/23, 38 times in 2023/24, 37 times in 2024/25, 12 times in 2025/26. For each match we used the last save before kick-off. That was usually 2 to 3 days before; in 2025/26, with fewer saves, a median of 28 days. Fixture ids and teams matched our history for all 1,520 matches. FDR is per team per fixture.
xG fixture rating. Team xG for a match is the sum of FPL’s player xG. Attack rating = own xG per match × opponent’s xG conceded per match ÷ the league average, × a home factor (the square root of home over away goals, 1.51 v 1.22 per match, from 3,040 earlier Premier League matches; divided for away teams). Defence rating is the same for goals against. Windows are the last 10 (or 20 or 38) gameweeks the team played, this season and last only. Midfielders and forwards are rated on attack; goalkeepers and defenders on defence. This was fixed before running.
Market team goals. The Poisson pair of team goals that best fits the closing 1X2 and over/under 2.5 prices (average across bookmakers, margin removed). Used as information, the same way the site’s projections use it.
Scoring. Pairs with equal points are skipped; ties in a predictor count half. Between players: every pair of starters in the same gameweek and position (828,833 pairs). Same player: every pair of one player’s starts in a season (262,875 pairs). The 95% intervals resample gameweeks (between players) or player-seasons (same player) 1,000 times.
Code. scripts/analysis_xg_vs_fdr.py and scripts/analysis/build_fdr_pit.py; run on 9 Oct 2026. This is a one-off analysis; no FixtureShark model was changed.
Between-player disagreements. When FDR and the xG rating ordered two players oppositely, the xG rating was right 51.7% of the time and the market 57.1% (against FDR).