Docteur Bet
Methodology

How we finda value bet

A probability, a price, a gap. Here is how each one is obtained, how we check that it is right — and where our models get it wrong.

The method in three steps

We model football as we would any statistical phenomenon: from data, with models whose assumptions are known — not from an intuition dressed up as a percentage. A value bet comes out of three steps, always the same.

A probability. For every fixture, the model estimates the probability of each outcome: the result, the number of goals, both teams to score.

A price. We collect the odds of the operators licensed in your country. Odds are a probability in disguise: 2.50 means 40%, the operator's margin included. A published bet stays recorded at the odds collected when it was published — never at better odds found after the fact.

A gap. When our probability exceeds the one implied by the odds, the price pays more than the risk. The gap is measured by expected value: EV = probability × odds − 1. A bet is only selected if that gap exceeds a threshold set market by market, because a small gap is indistinguishable from the model's own error.

The rest of this page walks through the construction of the first one, from the most naive model up to the one we publish.

Starting with the simplest thing

The most naive model looks at nothing: it takes the average number of goals scored at home and away across the league and applies it to every fixture. In Ligue 1 that gives about 1.9 goals for the home side and 1.7 for the visitors, whatever the fixture.

It is a bad model, but it asks the right question: from two expected goal numbers, you can already compute the probability of every scoreline, and therefore of a win, a draw, or over 2.5 goals. All the work that follows consists of getting better numbers.

Giving each team its own strengths

In 1982 Michael Maher proposed describing each team with two numbers rather than one: an attacking strength and a defensive strength. A single parameter would not distinguish a side that wins 3-2 from one that wins 1-0.

λ = attack(home) × defence(away) × home advantage μ = attack(away) × defence(home)

An attack of 1.60 means the team scores 60% more than the league average against an average defence. A defence reads the other way round: it is a multiplier of goals conceded, so the lower the better. Home advantage is common to the whole league, and comes out at about ×1.23 on average in our data.

These strengths cannot be read off the table: they are estimated over thousands of matches, and adjusted for the quality of the opponents faced. A side that started well against promoted teams does not earn the same strength as one that did as much against the top of the table.

What the Poisson distribution cannot do

Maher assumes the two sides score independently of one another, following two Poisson distributions. That is convenient, and it is wrong on one specific point, which Dixon and Coles measured in 1997.

Tight scorelines are more frequent than independence predicts. There are more 0-0s, 1-0s, 0-1s and 1-1s in reality than in the model. That is no accident: at 0-0 in the 80th minute, the two sides do not play as they would at 3-1. The score influences the game, so goals are not independent.

Forgetting the past, but slowly

Dixon and Coles's second idea is that a match from five years ago does not say as much as one from last week. Each match receives a weight that decays over time. A match one half-life old counts for half.

That half-life is not chosen by intuition. It is calibrated league by league: we replay each past season week by week, predict the next matchday under several settings, and keep the one that was least wrong. The result varies enormously, and that variation is itself information — where the half-life is short, hierarchies move fast.

Dixon-Coles model settings by league: time-decay half-life (days), ρ parameter, number of matches, effective sample and last calculation date.
LeagueHalf-lifeρMatchesEffective sampleLast computed
Admiral Bundesliga365 j—98843522 November 2026
Allsvenskan365 j—1,80768722 November 2026
Bundesliga (Germany)365 j-0.0552,19185522 November 2026
Champions League— j—1447222 November 2026
Championship (England)240 j—3,9431,02122 November 2026
Ekstraklasa1095 j—2,0761,70722 November 2026
Eliteserien180 j—1,75434722 November 2026
Eredivisie (Netherlands)365 j—2,11386622 November 2026
La Liga (Spain)365 j-0.0392,7401,07222 November 2026
La Liga 2 (Spain)240 j—3,27182322 November 2026
Liga Portugal (Portugal)1095 j0.0502,0241,6659 November 2026
Ligue 1 (France)730 j-0.0452,3841,59522 November 2026
Premier League (England)180 j-0.0662,70950822 November 2026
Premiership (Scotland)730 j—1,5441,04822 November 2026
Pro League (Belgium)365 j—2,13483122 November 2026
Serie A (Italy)240 j0.0192,73868422 November 2026
Serie B (Italy)120 j—2,74932922 November 2026
Super League (Switzerland)240 j—1,43140322 November 2026
Süper Lig (Turkey)120 j—2,41627222 November 2026
Superliga— j—1,2761,27621 November 2026

The effective sample is the number of matches it would take, all of equal weight, to carry as much information as the weighted set. In the Premier League, 2,709 matches are worth only 508: that is the price of a short memory, and it is also why probabilities there are less clear-cut.

The case of promoted sides

A newly promoted team has no history in its new league. Estimating a strength from three matches would produce nonsense: three wins would make it a title contender.

Its estimate is therefore shrunk towards the average level of a promoted side, measured over 149 observed promotions since 2020 across ten European leagues. On average, a promoted side attacks at ×0.82 and concedes at ×1.34 relative to the rest of its new league. The more it plays, the less that anchor weighs.

Comparing two leagues

Dixon-Coles was written for a single league, and there is a reason for that: for the model to be identifiable, the attack strengths are recentred around the league average at every fit. Each index therefore reads relative to its own league — Porto's relative to Portugal, Manchester City's relative to England. Setting them side by side is comparing two temperatures where one is in Celsius and the other in Fahrenheit.

For European competitions, we added two terms per league to the model, one for attack and one for defence, estimated on European fixtures between clubs from different leagues:

λ_home = α_i · β_j · exp(δ_L + ε_M + g)
λ_away = α_j · β_i · exp(δ_M + ε_L)

g is the home advantage specific to European competition, estimated on these fixtures rather than carried over from the domestic leagues. Each past fixture is explained by the strengths in force on its date, never by today's: no fixture takes part in estimating its own strengths.

Do these terms earn their place? The question is settled out of sample, on a season never seen during fitting. The verdict is clear: comparing two domestic indices with no correction barely beats historical frequencies. That is the measure of what the shortcut is worth, and of what this offset avoids.

Replaying the season

The remaining fixture list is replayed 50,000 times. Each match is drawn at random according to the model's probabilities, points accumulate, and the final table is settled by the league's real rules — points, then goal difference, then goals scored.

A title probability of 60% means exactly this: across fifty thousand simulated seasons, that team finishes first in thirty thousand of them. And is wrong four times in ten.

Since the strengths are themselves estimated, the model is refitted 60 times on resamples of past matches, and each simulated season draws one of those fits at random. The probabilities thus carry the uncertainty on the strengths, not only the randomness of the matches.

What the app publishes, competition by competition

Not every competition lends itself to the same calculation. A knockout cup cannot be simulated like a league where everyone plays everyone, and a league without enough history does not yield reliable strengths. Rather than publish a wrong figure, we publish what the model can compute: of the 24 competitions in the app, 16 have match-by-match probabilities — 14 of them with an end-of-season projection —, 5 only have team strengths, and 3 have no published computation yet.

When the figures move

The model is refitted after every matchday played, and only if matches have been played since the last computation. Without a new result it would return exactly the same figures under a more recent date — which would suggest a movement that did not happen. In practice: one computation a week, two in weeks with midweek fixtures, none during international breaks.

Every computation carries its date. If the computation chain fails, the app and the site keep serving the last valid computation — and that date is the only thing that tells you so.

What this model does not know

It knows results and nothing else. Not injuries, not transfers, not a change of manager, not a match played with ten men. A side that loses its striker in January will keep its attacking strength until results say otherwise.

And the Dixon-Coles correction, while it settles the question of tight scorelines, does not lift the independence assumption: it corrects it on four cells, not everywhere. That is an acknowledged limit of the model, and it has been known since 1997.

What it was not built for

This model explains a result well from the overall strength of two sides. It was not built for a different task: finding precisely where an operator has mispriced. An odds market already aggregates the information of thousands of participants; beating it does not call for a good result model, but for a model calibrated for that particular contest.

That is why, and not out of any taste for secrecy, the value bets come from another model. We do not detail its construction: it is what carries the product's value. Its page, however, says what it looks for, how we judge it and where it fails.

How we check

A model is not judged by the successes it highlights, but by everything it has produced. Three checks are public.

  • The track record. Every value bet is published before kick-off, then settled — wins and losses, since day one, market by market. Markets are never added together: the same bet appears in several of them. See the track record.
  • Calibration. When we announce 60%, the event should happen about six times in ten. The gap between what we announce and what happens is measured market by market, and shown where we overestimate as well as where we are right.
  • The out-of-sample backtest. The same selection rule, applied to fixtures the model had never seen, with the expected curve and its margin of uncertainty next to the real one.

Calibration and the backtest can be viewed in the app's Track record tab, free and with no account needed.

The limits we own

  • A probability is not a prediction. 61% means that, across a hundred similar situations, you would expect the outcome in about sixty-one of them — not that it is written in advance.
  • The model always starts from a history. An unprecedented event — a major injury, a radical change of manager — takes time to be absorbed, since it is only seen through results.
  • No model guarantees a profit. An edge measured over one period has to be measured again over the next, and a losing run is part of the game, even when the probabilities are right. Only stake what you can afford to lose.

A question about the method?

A step seems unclear, a figure looks wrong, a limit goes unstated? Write to us rather than guess. We answer on what is measurable — and on what is not.

Question about the method

Tell us which step or which figure raises a question, with the page address if you have it.

Today’svalue betsIn your pocket

Model probabilities, operators’ odds, value gaps: it’s all in the app. Seven-day trial, no commitment.

18+ · Gamble responsibly · BeGambleAware.org · National Gambling Helpline 0808 8020 133