Premier League Title Race: Data Breakdown

Our model's projected points table and the three fixtures that decide it all.

Title races are decided by fewer matches than the table suggests. By the time a season reaches its final third, most of the variance has already been squeezed out and what remains is a small number of fixtures where the projected points swing is large enough to change the outcome. Here is how we model that, and what the model says matters.

Points projections, not form tables

A form table tells you what has happened. It is a poor predictor because it weights a 1-0 win earned from two shots on target identically to a 1-0 win built on twenty. Our projections run on underlying performance instead — shot volume, shot quality, and the share of expected goals each side takes from a match.

For every remaining fixture we generate a distribution of scorelines rather than a single prediction, convert that into win, draw and loss probabilities, and then simulate the rest of the season thousands of times. The output is not "Team A finishes on 84 points" but "Team A finishes on 84 points as the median, with an 80% band running from 78 to 89."

That band matters more than the median. Two clubs projected to finish level on points can have very different title probabilities if one has a wider spread of outcomes ahead of it.

Where the race is actually decided

Three categories of fixture do most of the work, and only one of them is obvious.

• Head-to-heads between the contenders. A six-point swing in a single afternoon, and the market usually prices these efficiently because everyone is watching.
• Away trips to mid-table sides with strong home records. These are where title challenges quietly die, and where the market consistently overrates the favourite because the badge is bigger than the fixture.
• The fixture immediately following a European away tie. Rotation risk and travel fatigue are real and measurable, and the closing line often moves late once the team news lands.

Schedule strength is not a single number

Most published "remaining fixture difficulty" ratings average the league position of upcoming opponents. That is close to useless. A run of three matches against defensively organised sides who concede few chances is a different problem to three matches against open, high-scoring teams, even if the opponents sit in identical league positions.

We split difficulty into attacking difficulty and defensive difficulty. A side that wins by outscoring opponents suffers against low-block teams and thrives against open ones. A side built on clean sheets has the opposite profile. Two clubs can share an identical average opponent rating and face genuinely different runs.

What the model is bad at

Worth saying plainly, because a projection presented without its weaknesses is marketing rather than analysis.

• Injuries to a single irreplaceable player. Squad-level metrics absorb this too slowly, and the market reacts faster than any model.
• Managerial change. The bounce is real but its size is close to unpredictable, and the sample of comparable changes is always small.
• Motivation once a race is settled. A side with nothing to play for in May does not behave like the same side in February, and no amount of shot data captures that.

Turning a projection into a bet

The outright title market is usually efficient at the top and soft in the middle. The value is rarely in backing the favourite — it is in the second and third favourites when the model's spread of outcomes is wider than the price implies, and in the European qualification places, where public attention is thinner.

Our approach is to compare the model's implied probability against the price after removing the bookmaker's margin, and take a position only where the gap clears a threshold wide enough to survive our own error bars. Most weeks, that means no bet on the outright at all. That is the correct answer more often than it feels like it should be.