Messi Goals vs Expected Goals: A Finishing Case Study

Two public sources carry long runs of Messi's shot data, and they cover different spans.

Understat publishes its own xG model for league matches from 2014/15 onward. For Messi it covers La Liga with Barcelona (2014/15 to 2020/21) and Ligue 1 with Paris Saint-Germain (2021/22 and 2022/23). It does not cover MLS, cup competitions, or internationals. Figures here were read from Understat's player data on 22 October 2026.

FBref displayed Opta's xG model from 2017/18 onward, which extends to Messi's MLS seasons with Inter Miami. Sports Reference announced on 20 January 2026 that Opta had terminated its data feed and that the advanced data had been removed from FBref, as reported by Awful Announcing and The Athletic. The Opta figures below therefore come from an Internet Archive snapshot of Messi's FBref page dated 17 December 2025.

Neither source covers his first decade at Barcelona, so whole-career claims go beyond these two sources. Other public datasets reach further back and are not directly comparable with either model.

Non-penalty goals minus non-penalty xG, Barcelona, La Liga: 2014/15 +6.6, 2015/16 +1.1, 2016/17 +9.3, 2017/18 +6.0, 2018/19 +9.7, 2019/20 +2.9, 2020/21 +4.0. Every season was positive. At PSG the sign flipped: 2021/22 was 3.9 below and 2022/23 was 1.7 below.

FBref's Opta numbers point the same way for the overlapping Barcelona seasons. Its MLS figures for 2023 to 2025 show 48 non-penalty league goals from 34.5 npxG, though MLS shots are not directly comparable with La Liga shots.

An xG model estimates how often an average shooter scores a given chance. If each shot is treated as an independent trial with scoring probability equal to its xG, the expected goal total is the sum of xG values and the variance is the sum of xG times one minus xG.

Applying that to Understat's shot-level values for Messi's 1,205 non-penalty La Liga shots gives a standard deviation of about 10.4 goals (our calculation). The observed excess of 39.6 goals sits roughly 3.8 standard deviations above expectation. For a calibrated model and an average finisher, a gap that size is very unlikely.

Two caveats keep this honest. First, the test asks whether Messi beats the model, not why; a model that misses something his shots share would produce the same result. Second, analysts look at thousands of players, so a few large deviations appear by chance. The stronger evidence is consistency: seven positive seasons in a row on Understat, with the same direction on Opta's model.

Splitting Understat's Barcelona shots by situation shows where the gap was built. Open play produced 166 goals from 137.0 xG on 852 shots, the bulk of the excess. Direct free kicks produced 28 goals from 19.8 xG on 284 attempts, a gap of about 1.9 standard deviations on its own. Penalties went 28 goals from 29.0 xG, essentially at expectation, which is why npxG is the fairer comparison for finishing.

For Barcelona from 2017/18 to 2020/21, Understat's npxG totals 88.3 while Opta's is 77.1, against the same 111 non-penalty goals. Opta's model includes shooter pressure and goalkeeper position, per FBref's explainer, so the same shot can carry a different value. The gap above xG is about 22.7 goals on Understat and 33.9 on Opta. The direction agrees; the size depends on the model. Never mix sources in one comparison.

Finishing skill is real but noisy, so estimate it with shrinkage: start each player at the population average and move toward their observed goals minus xG only as shots accumulate. Messi is the rare case where the data overwhelms the prior. For most players, a season or two above xG says little.

This page describes historical data and method, and is not betting advice.