Event Data vs. Tracking Data

Data companies employ analysts who watch the match and tag every on-ball action with its type, time and location.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Opta began in 1996, rating Premier League players with an index built from logged match actions.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Every pass, touch, duel and shot is an event, which adds up to thousands in 90 minutes.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Clients want numbers during the match, and a post-match pass fixes errors in locations and event types.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Knowing who did what, when and where is the core of event data; details come as extra labels.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

A run that drags a defender away leaves no event, because nobody touched the ball in that action.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Custom tools let a logger click the pitch location and pick the event type as fast as the game moves.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Speed comes first for live products; accuracy is added in the checking stage before the final release.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Reep began in the 1950s, Opta in 1996, and StatsBomb added 360 location snapshots in 2021.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Events carry details such as the body part used; the other options are not part of an on-ball event.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Qualifiers add detail such as body part, pass height or whether the ball was a cross.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

A pass can be headed, long, a cross and under pressure all at once; qualifiers let it say all of that.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The 'under pressure' flag marks actions taken with an opponent closing in, which changes how they should be judged.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

xG models treat a header from a spot very differently from a strike with the foot from the same place.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

What counts as a tackle, a dribble or a key pass varies, so the labels depend on each provider's rules.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Where a pass went astray matters for turnovers and counters, so the intended or actual end point is kept.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Passes are completed or not, duels won or lost, shots on target or off: the outcome records which.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

The type says what happened, the outcome whether it worked, and qualifiers add the finer details.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The count is built from the basic event plus a qualifier and a location test.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Height, body part, pressure and type are qualifiers: extra descriptions added to the basic event.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Opta's pitch is a percentage grid, so 50 on the length axis is halfway, whatever the real size of the ground.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

StatsBomb records locations on a 120 x 80 grid in yards-style units, not in percentages.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

A point at (60, 40) means the centre spot for one provider and somewhere else entirely for another.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Normalising direction means a location near 100 is always close to the opponent's goal.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A pass across 50 units of width is a different number of metres on a narrow ground and a wide one.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Tracking systems work in real distances, commonly with the centre spot as the origin.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

The length runs from 0 to 120, so 60 is exactly halfway.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Opta measures length as a percentage, so higher x means closer to the goal being attacked.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Direction and scale have to match before any point can be compared.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Halfway along 120 and halfway across 80 gives (60, 40); (50, 50) is the centre on a 100 x 100 grid.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Cameras placed around the roof cover the pitch from several angles, and software turns the images into positions.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

The software has to know exactly where each camera looks and how big the pitch is to turn pixels into metres.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

When bodies overlap, the system can lose or swap identities, which is why several angles are used.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A small object moving at up to 100 km/h and leaving the ground is much harder to follow than a player.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Software identifies players from appearance and position, with operators correcting any swaps.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Hawk-Eye's camera systems are used for goal-line decisions and, increasingly, for player and ball tracking.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Automation does most of the work, but people correct the moments when the software loses track.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Detection comes before geometry, and identity is attached last so that every point belongs to someone.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The more bodies overlap, the harder it is for the cameras to keep each player separate.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Three players close together block each other from any one camera, so views from other sides keep them apart.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Agreement between independent loggers shows whether event definitions are applied the same way.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A goal is a goal; whether a challenge was a tackle, a duel or a foul depends on the logger's reading.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A logger judges where the action happened on screen, which is quick but not exact.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

An impossible jump between frames means the system lost him and found him again somewhere else.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Spot checks against video show whether the definitions and locations match what the club sees.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A tighter definition of a tackle or a key pass lowers the counts without anyone playing differently.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

If an event happens twice a season, one mistake is half the total; common events absorb errors better.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Goals are unmistakable, passes mostly clear, and tackles depend on judgement.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

An audit needs a trusted reference, which the club builds itself before comparing.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

18 against 13 for the same match shows that deciding what counts as a tackle is a judgement call.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Events say what happened on the ball; tracking adds the positions of all 22 players at that moment.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A logger's timestamp can be a fraction of a second off, so the event must be matched to the right frame.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The moment the ball's tracked position starts to move away from the player marks the real time of the pass.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

At sprinting speed a player covers several metres in half a second, so a mismatched frame puts him in the wrong place.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

360 gives a snapshot around each event from broadcast video, not a continuous record of all players.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Event clocks restart at 45:00 while tracking runs on its own time, so each half needs its own alignment.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Knowing where team-mates and opponents stood shows which passes were available, not just the one chosen.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Clock alignment comes before event matching, and a video check confirms the join worked.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

An event is one action, a freeze frame adds visible players, and a tracking frame has everyone.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Aligning the event with the frame where the ball actually leaves the foot puts every player in the right place.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Perform, which had bought Opta, merged with STATS in 2019 to create Stats Perform.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Hudl, known for video analysis software, added Wyscout's scouting platform and database.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Fast, reliable live data feeds broadcasts, apps and in-play betting, so leagues license it for a fee.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

One provider may be strongest on event detail, another on tracking, another on smaller leagues abroad.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Detailed logging and tracking cost money, so coverage is deepest where the paying customers are.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Broadcasters, apps and betting firms need the events almost instantly, which is what live feeds provide.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Detailed positional data on their own players and opponents is a competitive advantage.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Opta began in 1996, Hudl bought Wyscout in 2019, and StatsBomb launched 360 data in 2021.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · recall

Speed is the product: the whole chain runs in seconds.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Providers collect and check the data at the stadium and sell it on to clubs, broadcasters and others.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

StatsBomb's open data also covers women's football and several major tournaments, for learning and research.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

Metrica's anonymised sample matches give full positional data, which is otherwise very hard to obtain.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

The paper in Scientific Data released Wyscout events for European leagues and major tournaments.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

Camera systems are expensive and the data reveals a lot about teams, so little of it is released.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Researchers and club analysts ran it on YouTube during the pandemic, using open data sets for lessons.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

Working with public data is how many analysts built portfolios that led to jobs in the game.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Releases are usually selected tournaments or seasons, so they cannot answer every question about every league.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

The Wyscout paper came out in 2019, Friends of Tracking in 2020, and open 360 data from 2021 on.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

The same steps, done on public data, are how many analysts learned the trade.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Free event data sets far outnumber free tracking data, so most public projects are built on events.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

From players' positions and speeds, it works out which side would reach a ball played to any point first.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

Spearman's paper modelled how likely off-ball positions were to lead to goals, using tracking data.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

Who controls a space depends on where all 22 players are and how fast they are moving.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

EPV rates every moment of a possession by how likely it is to end in a goal, so each action's effect can be measured.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

The Boston conference is where researchers and clubs share analytics papers across sports.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · recall

A decoy run that drags a defender away increases his team's control of another area, which the model can show.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Models assume set reaction times and speeds, which real players, tired or alert, do not always match.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Positions and speeds come from tracking; the map shows which side owns each area at that moment.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

Basic xG needs only events, freeze frames add visible players, and pitch control needs full tracking.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

The home side controls the areas around 10 and 9, so passes into them are likely to be won.

Source: MIT Sloan Sports Analytics Conference papers and published data sets · applied

A run without a touch leaves no event, so only continuous positions can show it.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Passes are on-ball events with start and end locations, which event data records directly.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The line's height is a position held all match long, which only tracking records.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Shots against, with their locations and quality, are logged events.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Events record the tackles and turnovers; tracking shows the distances closed and the lanes cut off.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Events find the set pieces and the shots; video shows the marking and movement behind them.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Which way a player faced, whether he checked his shoulder and how he communicated are clearest on video.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

Shot counts are events, defenders at the shot need freeze frames, and pressing runs need full tracking.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

The question decides the data, not the other way round.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied

A freeze frame records the visible players' positions at the moment of each event, including the shot.

Source: Data provider documentation (Opta, StatsBomb, Wyscout) and the analytics literature · applied