How do xG models estimate a shot?
Most shot-based models learn from large samples of historical attempts. Typical inputs include shot distance, angle to goal, body part, the type of assist and the situation preceding the attempt. Some providers also use defender and goalkeeper locations, pressure, or tracking data. A model estimates the likelihood of scoring for shots with similar characteristics.
Different data providers can assign different xG values to the same chance because their data, features, modeling assumptions and definitions differ. Comparing xG across sources therefore requires care.
How is team xG calculated?
Suppose a team takes four shots with estimated scoring probabilities of 0.05, 0.12, 0.28 and 0.45. Adding these values gives 0.90 expected goals. This is the sum of individual shot probabilities, not a forecast that the team will score exactly 0.90 goals.
Over many matches, the relationship between xG and actual goals can help evaluate chance creation, finishing and goalkeeping. Small samples can be misleading: a few exceptional finishes or missed penalties may move results dramatically.
What xG does not measure
Traditional shot xG does not directly measure a player's ability to create a shot before it happens. Ordinary shot xG also does not fully represent possession value, pressing, off-ball movement or the quality of attacks that never produce a shot. Some providers offer related metrics such as expected threat (xT), expected assists (xA), or post-shot expected goals; these answer different questions and should not be treated as interchangeable.
Is xG the same as a pre-match expected-goals forecast?
No. A match's observed shot xG is calculated from chances actually taken. A pre-match forecast estimates future scoring rates using information available before kickoff, potentially including team strength, recent performance, player availability and tactical context. A model may use historical xG as an input to that forecast, but observed shot xG and projected goal rates are distinct quantities.
This distinction matters for Cateo: hypothetical goal-rate inputs in a research simulator must not be represented as measured xG from real matches.
How can xG support BTTS and scoreline models?
Pre-match goal-rate estimates can be fed into a probability model such as an independent Poisson model. That model assigns probabilities to possible home and away goal counts; from those distributions, analysts can derive outcomes such as both teams to score (BTTS), over 2.5 goals, or a particular scoreline. Real match outcomes can be correlated, however, and a basic Poisson model may require adjustments or validation.
Higher projected scoring rates generally increase the probability of certain high-goal outcomes, but that does not establish an advantage over market prices. To evaluate a forecast, compare its time-stamped probabilities with verified outcomes using calibration, Brier score or log loss over appropriate out-of-sample periods.
How to read xG responsibly
Start with the time frame, sample size and source. Ask whether the number is observed shot xG or a future goal-rate estimate. Check whether penalties are included and whether the figures come from the same provider. Use uncertainty ranges and contextual information rather than treating a single decimal as definitive.
Cateo's current public research experience includes explicitly fictional model examples. This educational article does not claim that live fixture predictions or a validated historical performance record are available.