Why do score apps show different stats for the same match?

Providers define and count some stats, such as possession and xG, in different ways, so two apps can show different numbers.

Elliot Hayward

Editor

Updated , 5 min read

Illustration of two phones showing the same pitch with different bar charts.
On this page
  1. Do all apps use the same data?
  2. How is possession calculated?
  3. How are shots and big chances defined?
  4. Why do xG numbers differ?
  5. How reliable is event data?
  6. What should you do when two apps disagree?

In a 2014 Slate report on the World Cup, one provider put Spain's possession against the Netherlands at 64% and Deltatre at 57%. Some stats also arrive late.

Do all apps use the same data?

Not necessarily. Providers collect events and define stats in their own ways, as the sections below show, so two apps can show different numbers for one match.

Flashscore's FAQ says that if you see different stats elsewhere, such as on UEFA or FIFA sites, the cause may be a different provider. Sofascore says its statistics may sometimes differ from official sources while staying consistent with its own definitions.

We haven't compared any two apps' numbers for one match, so we can't tell you how often they differ.

How is possession calculated?

There are at least two methods: counting passes and timing the ball.

Pass-based. In 2014 StatsBomb's Alex Olshansky described the figure that, in his words, the majority of media outlets use as a pass ratio: a team's passes divided by both teams' passes. He wrote that an employee of that provider confirmed this method. Slate, the same year, described the method as using completed passes, with nothing to do with time.

Time-based. Slate described Deltatre's method as a "chess clock": a person at the match presses a button for team A, team B or ball out of play. Wyscout's glossary says its figure is based on "pure time spent in possession of the ball".

These are 2014 third-party accounts and a vendor's glossary, so none tells you what that provider does now. The gap can be large: Slate listed Spain v Netherlands at 64% (the pass-based provider) against 57% (Deltatre), Spain v Chile at 63% against 56%, and USA v Germany at 68% against 63%. Slate also relayed that provider's claim that its tests show pass-based figures correlate closely with timing methods. We haven't seen the tests.

We don't know which method Flashscore, FotMob or Sofascore uses. Sofascore's definitions page covers shots, assists, tackles and clean sheets, and not possession.

How are shots and big chances defined?

Sofascore's definitions page says a shot that hits the frame of the goal isn't counted as on target unless the ball goes in and a goal is awarded. We haven't compared it with other apps' definitions, and we haven't tested whether the apps' numbers match.

A big chance is more of a judgement call, and Sofascore's definitions page doesn't define it. The KU Leuven authors found, in their comparison of two providers, that providers use different shot definitions too: some count shots blocked at the feet of the player, others don't.

Why do xG numbers differ?

xG, or expected goals, is a modelled number, so a different shot location or a different model gives a different value. Davis and Robberechts of KU Leuven tested what provider differences do to it. Their 2020 blog post used shots from the 2018/2019 top-five European leagues from two unnamed providers. Shot locations differed by up to a couple of metres, which they attribute to human annotators logging events in near real time. Summed team xG favoured opposite teams in 238 games (13%), and player xG rankings were more stable.

The authors conclude that data quality matters mainly when you analyse small amounts of data. That fits a single match, where one shot moves the total. The post is a blog, not a peer-reviewed paper.

How reliable is event data?

It depends on the provider and the event, and the studies we found are old and narrow.

  • Between providers. In 2019 FIFA described an internal study at the 2018 Club World Cup. Goals were fairly consistent, but free-kicks, corners and yellow cards weren't always accurate, and subjective indicators such as successful tackles and completed crosses showed discrepancies between providers upwards of 50%. The providers aren't named, there is no sample size, and it isn't peer reviewed.
  • Feeds. Wang and Kuzmanovic compared three anonymised feed providers on English Premier League games. The share of real events each captured was 0.996, 0.665 and 0.851, with no false events. It is a 2017-era study of events such as goals, cards, fouls and shots, and doesn't cover possession or xG.

Timing matters too. Sofascore says that in some leagues detailed player stats and ratings can appear several hours or even days after a match. Flashscore says some competitions or historical data may be missing because reliable information is hard to source.

What should you do when two apps disagree?

Compare like with like, then check the definition.

  • Read the app's definition of the stat. Possession, big chance and xG are the stats where the sources above found different methods or judgement calls.
  • For a final result or a table, use the competition's own site.
  • Give small samples less weight, because the KU Leuven authors found provider gaps matter most in small samples.

Our football stats apps comparison lists which stats each app names.

Frequently asked questions

Why can two apps show different possession for the same match?

Providers can calculate it in different ways. Slate reported in 2014 that two providers calculated possession in different ways. Sofascore's definitions page doesn't define possession. We haven't compared any two apps for any match, so we can't say how often they differ.

Is possession measured in time or in passes?

Both methods exist. In 2014, StatsBomb and Slate described one provider's figure as a share of completed passes, and Slate described Deltatre's as a clock operated by a person. Wyscout says its figure is based on time in possession. We haven't confirmed which method any app uses today.

What counts as a shot on target? Does hitting the post count?

Sofascore's definitions page says shots hitting the frame aren't counted as on target unless the ball goes in and a goal is awarded. We haven't compared it with the definitions of other apps.

Why is xG different on different sites?

xG, or expected goals, comes from a model, and models take different inputs from different event data. A 2020 KU Leuven blog post compared two unnamed providers' data for the 2018/2019 top-five leagues. Shot locations differed by up to a couple of metres, and summed team xG favoured opposite teams in 238 games (13%).

Are app stats the same as the Premier League's official stats?

We don't have a comparison that says so. Sofascore says its figures may sometimes differ from official sources but stay consistent with its own definitions.