Over/Under 2.5 Goals: Model Accuracy Across 50 Featured Leagues

By Tactiq AI · 2026-05-05 · 7 min read · Methodology

The over/under 2.5 goals threshold is the second-most asked question about a football fixture, after the win-draw-loss outcome. It is the cleanest signal of whether a fixture is going to be tight and tactical or open and end-to-end. And it is one of the most modelable: the expected goals literature gives a clear path from squad quality and matchup data to a probability distribution over total goals.

Tactiq computes over/under 2.5 probability for every fixture in the 50 featured leagues, and the model's calibration varies meaningfully across leagues. This article walks through how the model works and where its strengths and limits sit.

The core machinery

The over/under model is downstream of the same expected-goals engine that powers the win-probability output. Each side has an expected goals scored figure (their offensive output, weighted by recent form) and an expected goals conceded figure (their defensive output, similarly weighted). The fixture-specific expected total is the sum of expected scored by team A against team B's defense and expected scored by team B against team A's defense.

That gives a single number: expected total goals for the fixture. To convert it into an over/under 2.5 probability, the model assumes total goals follow a Poisson-like distribution centered on the expected total, calibrated to the league's historical distribution shape (most leagues are slightly over-dispersed compared to a pure Poisson, so the model uses a tuned variant).

The probability of over 2.5 goals is the area under that distribution above 2.5. A fixture with an expected total of 2.4 will have an under-50-percent over probability, because the median outcome is below the threshold. A fixture with expected total of 3.0 will sit clearly above 50 percent over.

Per-league baseline correction

Different leagues score at different rates. The Bundesliga averages around 3.1 goals per fixture across recent seasons. Ligue 1 hovers around 2.5. The Greek Super League is closer to 2.3. The Eredivisie often runs above 3.2.

A naive model that assumed a global average of 2.7 goals per fixture would systematically over-estimate over 2.5 probability in low-scoring leagues and under-estimate it in high-scoring leagues. Tactiq corrects for this with per-league baselines, computed from rolling 3-season historical averages.

The correction is straightforward: when the model produces an expected total for a fixture in the Greek Super League, it is implicitly comparing against the Greek baseline distribution, not the global one. The probability output is consequently calibrated against what is normal for that league.

Calibration figures

Tactiq tracks the Brier score of the over/under 2.5 model per league. The aggregate figures for 2024 and early 2025 sit in the following ranges:

Tier 1 calibration (Brier under 0.20). Premier League, Bundesliga, La Liga, Eredivisie, Serie A. These leagues have deep historical data, stable fixture counts, and well-calibrated baselines. The model performs strongly and the over/under output is reliable.

Tier 2 calibration (Brier 0.20 to 0.23). Ligue 1, Primeira Liga, Süper Lig, Championship, Belgian Pro League, Greek Super League, Brazilian Serie A. Solid calibration with occasional larger misses driven by league-specific anomalies (Ligue 1's known low-scoring profile, Süper Lig's higher variance).

Tier 3 calibration (Brier 0.23 to 0.25). Most other featured leagues, including most regional Scandinavian, Eastern European, and Asian leagues. Calibration is meaningfully better than the 0.25 random-guessing baseline but with more variance than the Tier 1 leagues.

A Brier score of 0.20 corresponds roughly to: when the model says 60 percent over 2.5, the over hits 60 percent of the time, with typical errors of plus or minus 5 percent in either direction. This is genuinely useful calibration for fans assessing whether a fixture is likely to be tight or open.

Where the model is most accurate

Three categories of fixture where the over/under output is most reliable:

Top-flight fixtures with deep data. A Premier League or Bundesliga fixture between two stable mid-table sides is the model's best case. Both sides have hundreds of fixtures of historical data, recent form is well-tracked, and league-specific baselines are robust.

Fixtures with clearly differentiated profiles. A fixture between a high-attacking side and a low-blocking side has a clear expected pattern (high attacking output suppressed by tight defense, leading to a fixture often in the 1.5 to 2.5 goals range). The model captures these mismatches well.

End-of-season fixtures. Late-season fixtures benefit from a full season of in-season form data plus the previous season's baseline. The model's expected total is most precise in this window.

Where the model is less reliable

Three categories where the output is noisier:

Early-season fixtures. The first five matchdays of a new season have the model leaning heavily on the previous season's baseline because in-season form data is sparse. Squads change in the summer; managers change; tactical setups evolve. The first ten matchdays have higher variance in over/under output.

Knockout cup ties. Cup fixtures are not always part of the featured leagues' regular over/under output. When they are, the data is sparser (cup fixtures are fewer) and the calibration weaker.

Fixtures with very high or very low expected totals. A fixture with an expected total of 4.5 or above is in the tail of the distribution. The model still produces a probability, but the calibration in those tails is less tested than around the 2.5 to 3.0 expected-total band where most fixtures live.

How fans should read the output

A practical reading: treat the over 2.5 probability as the model's best read of how open the fixture will be. A 65 percent over probability says the fixture is meaningfully more likely to be open than tight, and 65 percent of the time, that read will be correct. A 45 percent over probability says the fixture is leaning tight.

The model does not predict specific scores. It predicts the distribution of total goals. A fixture that hits over 2.5 with a 4-3 scoreline and a fixture that hits over 2.5 with a 2-1 scoreline are both "over 2.5 hits" by the model's measure, even though the texture of the two fixtures was wildly different.

The over/under output is most useful when you read it alongside the win probability. A fixture with high over probability and balanced win probabilities suggests a chaotic open contest. A fixture with low over probability and tilted win probabilities suggests the favourite will grind out a 1-0 or 2-0. Reading the two outputs together gives a fuller picture than either alone.

The over/under 2.5 model is included in Basic and Premium tiers across iOS, Mac, and Android. Free tier users see win probability without the over/under output.

Frequently Asked Questions

How does Tactiq compute over/under 2.5 goals probability?

The model produces an expected total goals figure per fixture by combining each side's expected goals scored and conceded, weighted by recent form and home advantage. The expected total is converted into a probability distribution over total goals, and the over 2.5 probability is the area of that distribution above 2.5.

Why 2.5 goals specifically?

Two and a half is the most-tracked total-goals threshold in football because it splits the typical fixture distribution near the median. Lower-scoring fixtures fall below 2.5, higher-scoring ones above. Modeling other thresholds (1.5, 3.5) is a straightforward extension and Tactiq computes them internally for analysts who request them.

Does per-league calibration vary?

Yes, meaningfully. Some leagues are systematically higher-scoring (Bundesliga, Eredivisie) and some are systematically lower-scoring (Ligue 1, Greek Super League). The model applies per-league baseline corrections so that a 65 percent over 2.5 probability in the Bundesliga represents the same calibration as a 65 percent in Ligue 1.

How does the model handle extreme fixtures?

Fixtures with unusually high expected total goals (over 3.5 expected) are common in some leagues and rare in others. The model treats them with the same probability machinery but flags them in the analysis output so the user knows the fixture is on the tail of the league's distribution.

Is over/under available for all 1,200+ leagues?

It is computed for the 50 featured leagues where Tactiq has full historical data and squad context. For non-featured leagues, the over/under output is shown but with a note indicating the lower-data calibration band. Premium users see the full output regardless.

How is the model's performance measured?

Same way as the win-probability model: a Brier score against the actual outcome (over or under). Lower Brier means better calibration. The model's published per-league Brier figures range from approximately 0.18 (Bundesliga, Premier League) to 0.24 (lower-data leagues), well below the 0.25 random-guessing baseline.