Expected Points: The League Table That Ignores Who Actually Won

A win is worth three points whether a team dominated from start to finish or rode two wonder saves and a deflected winner in stoppage time. That flattening is useful for settling a league, since somebody has to end up champion and somebody has to go down, but it throws away almost all the information about how a result actually happened. Expected points is an attempt to keep that information around.

The calculation starts from a match's expected goals for both sides and runs a probability model, essentially a huge number of simulated versions of that same match based on the two xG totals, to work out how often a team with that shot profile would be expected to win, draw, or lose. A team that generates 2.1 xG against an opponent's 0.6 will win the resulting simulation the large majority of the time, so it earns something close to a full three expected points, even in the specific match where the actual result was a 1-1 draw because of a defensive howler or a moment of finishing brilliance from the other side. Add up expected points across a season and you get an alternative table, one built from the quality of chances created and conceded rather than from what actually went in.

What the gap between the tables means

The value of this second table is mostly diagnostic. A team sitting well below its expected-points total, taking fewer real points than its underlying performances suggest it should, is usually either finishing badly, conceding soft goals its process doesn't explain, or simply running into a stretch of bad luck that tends to even out given enough matches. A team well above its expected-points total is often riding exactly the kind of variance that regresses eventually, whether that's a striker outperforming his xG for a few months or a goalkeeper having the run of his career. Neither situation shows up clearly in the real table, which only ever tells you what happened, not what should have happened given the chances on both sides.

Expected points and the manager debate

This is also why expected points gets used so often in manager-sacking arguments, for better or worse. A club whose real results have collapsed while its expected-points total stays respectable has a decent case that the underlying football is fine and the results are a temporary problem, while a club grinding out results well above what its performances deserve has a much shakier position than the table alone suggests, since that gap tends to close on its own eventually, usually at an inconvenient time for whoever gets credit for the results while they lasted.

Where the model falls short

The model has real limits worth being honest about. It's only as good as the underlying xG numbers feeding it, and it has no way to account for a team that specifically manages games well, protecting a lead by controlling tempo rather than continuing to create chances once ahead, which can make a smart, experienced side look statistically worse than a chaotic one that keeps generating, and conceding, high-value chances until the final whistle regardless of the scoreline. It also treats every match as an independent event with a fixed win-draw-loss distribution, which is a reasonable simplification but not a perfect description of how momentum, fatigue, or a manager's in-game adjustments actually shape a result.

Used well, expected points is a longer memory for a season than the table provides on its own, a way of asking whether last month's results reflected last month's performances or just got lucky. Used carelessly, it becomes another number thrown into an argument that was really about something else, usually whether a manager should keep his job.

The whole model starts from a match's expected goals for both sides, run through a simulation rather than reported as its own separate number. Elo ratings ask a similar question over a team's entire history rather than one match at a time.

How big a gap between xPts and real points actually matters?

There's no official threshold, but analysts generally start paying attention once a team's real points total diverges from its expected-points total by somewhere around 8-10 points over a full season, since that's a large enough gap that it's unlikely to be pure random variance in finishing and defending. A smaller gap of a few points across 38 games is well within the range normal luck produces and isn't worth reading much into on its own.

How are expected points calculated?

Expected points start from a match's expected goals for both sides. A probability model simulates that match many times from the two xG totals to work out how often each team would win, draw or lose. A team generating 2.1 xG against 0.6 wins most simulations, so it earns close to three expected points even if the real result was a 1-1 draw.

What does it mean when a team outperforms its expected points?

A team well above its expected points total is often riding variance that tends to regress, such as a striker outperforming his xG for a few months or a goalkeeper having the run of his career. A team well below it is usually finishing badly, conceding soft goals or running into bad luck that tends to even out over enough matches.

What are the limits of expected points?

Expected points are only as good as the xG numbers feeding them. The model cannot reward a team that manages games well by protecting a lead and controlling tempo, so a smart side can look worse than a chaotic one. It also treats each match as an independent event, ignoring momentum, fatigue and a manager's in-game adjustments.