Match the model to the line
Over 2.5 goals wins when the settled goal total is at least three; under 2.5 wins at zero, one or two. Whole and quarter lines introduce pushes or split settlement. DraftKings' current soccer rules are one operator example; always use the actual operator's current terms.
Worked example: from a goal distribution to probability
A simple Poisson model can start from a supplied total-goal rate lambda. OpenStax's Poisson guide defines the probability mass function. Football models often require extensions for dependence and changing team strengths; Dixon and Coles is a foundational score-model example.
For an illustrative lambda = 2.70:
- P(0 goals) = 0.06721.
- P(1 goal) = 0.18147.
- P(2 goals) = 0.24499.
- P(under 2.5) = 0.06721 + 0.18147 + 0.24499 = 0.49367.
- P(over 2.5) = 1 - 0.49367 = 0.50633.
These values are illustrative outputs from the stated rate, not a current match forecast.
Compare with the accepted price
At decimal odds 2.05 for over 2.5:
EV = (0.50633 * 2.05) - 1 = 0.03798 units.
If a reasonable lower rate produces p = 0.47, EV = (0.47 * 2.05) - 1 = -0.0365 units. The decision is sensitive to the goal-rate estimate, so report the range.
Audit the two-way market
Capture both over and under prices simultaneously and calculate the booksum. Keep alternate lines because the shape across 2.0, 2.25, 2.5, 2.75 and 3.0 can expose input or settlement mistakes. Do not compare a model's 2.5 probability with an operator's 2.25 price.
Validate on later fixtures
Check log loss or Brier score, calibration, coverage and results by league, line and odds band. Calibration guidance supports comparing predicted probabilities with observed frequencies. Freeze rates and features at the forecast cutoff; corrected xG or confirmed lineups cannot appear early in historical rows.
Diagnose the probability before using the price
| Check | Passing evidence | Failure response |
|---|---|---|
| Distribution total | Score probabilities sum close to one | Fix truncation or normalization |
| Goal-rate calibration | Later total frequencies align by band | Refit or recalibrate on earlier data |
| Line consistency | Over and under complements reconcile | Check pushes and quarter lines |
| Baseline comparison | Model improves a declared simple baseline | Retain the baseline |
| Market mapping | Target and settlement match exactly | Exclude the row |
Repeat these checks by league and forecast lead time. A model can produce a plausible overall over-2.5 rate while failing at high totals or in competitions with different scoring patterns. Publish the failed slices beside the aggregate result. Preserve the original probability grid as well as the final over/under aggregate so a reviewer can locate whether an error entered through the goal rates, score cells, line mapping, or price comparison.
Continue learning
- Next guide: Football Betting-Market Efficiency
- Related guide: How Bookmakers Set Football Odds
Assumptions and limitations
The Poisson probabilities and prices are illustrative and rounded. Independent team-goal assumptions can fail, and operator settlement differs for abandoned matches or alternate lines. The Poisson over/under calculator provides the full arithmetic.
GambleAware
GamCare
Gambling Therapy
GAMSTOP