Goal Totals & Over/Under 2.5: Soccer's Key Number
Over 2.5 needs at least three settled goals; under 2.5 needs at most two. The half-goal line avoids a push. Neither a league label nor a projected average above 2.5 establishes the probability of going over.
A mean is not a tail probability
Different goal distributions can share the same mean. To evaluate over 2.5, estimate P(G ≥ 3) for the contract’s settlement period. Confirm whether that period is regulation plus stoppage time, a single half, or something else; do not mix extra-time goals into a regulation-only sample.
A hypothetical Poisson model with mean λ = 2.85 gives P(G ≥ 3) = 1 − exp(−λ)(1 + λ + λ²/2) = 54.2379%. At -115, break-even is 115/215 = 53.4884%. Expected net return is P(G ≥ 3) × (1 + 100/115) − 1 = +1.4013% of stake. These are modeled figures, not observed league rates, and depend on the Poisson assumption.
Reducing the assumed mean to 2.70 gives P(G ≥ 3) = 50.6376%, below that same price threshold. The sensitivity can reverse the conclusion. A mean-goal disagreement with a line should therefore lead to a distribution and uncertainty check, not an immediate value claim.
What expected goals can contribute
Expected goals assigns a model-estimated scoring probability to a shot using its recorded context. Adding shot probabilities gives a chance-quality summary. Hudl Statsbomb’s explanation describes model inputs and why estimates can differ by provider. No single penalty or header value is universal.
Source: Hudl Statsbomb expected-goals explanation.
A post-match xG total describes the chances observed in that match. A pre-match forecast must separately estimate which chances will occur. A gap between past goals and xG is not proof of future regression or of an error in the current price. Record provider, model version where available, competition, dates, and sampling uncertainty.
League and half-specific research
To publish a scoring table, count eligible matches and goals in a declared season window, and compute over rates from actual scores. Do not infer a near-50/50 over rate from a mean near the line. The former 2020–2025 league table and first-half percentages have no retained dataset and are removed.
A first-half forecast needs its own distribution and settlement period. Do not automatically halve full-match xG or assume a universal share of goals before halftime. Validate any timing model on comparable matches and prices.
Continue reading: Three-Way Moneyline · Asian Handicap Mechanics.
Frequently Asked Questions
Does a 2.85-goal projection prove value on over 2.5?
No. A mean alone is insufficient. The worked example adds an explicit Poisson distribution and a -115 price, and its result changes when the assumed mean changes.
Is xG a known probability?
No. It is a model estimate for shot context. Providers can use different data and methods; a pre-match forecast also needs to model future chance creation.
Can I use a universal first-half scoring percentage?
This guide establishes none. Define and validate a half-specific sample and model, including the treatment of stoppage time.