Why data is important, and what it can and can’t tell us

Modern football produces more data than ever before, with every pass, tackle, shot and chance tracked across almost every major competition, but collecting those numbers is the easy part.

The value comes from understanding what they tell you, what they don't, and how much weight you should give them when making a betting decision.

Every price on a football coupon is already a data product before you open the bookmaker's app. Behind the match odds sits a model built around scoring rates, squad strength, home advantage, team news and countless other inputs, because bookmakers know better than anyone how important good information is when trying to price a football match. Betting without looking at the same underlying information means starting from a disadvantage and relying far more heavily on results, reputation and instinct.

No individual punter is going to process the same volume of information as a bookmaker or have access to the same modelling resources, but that doesn't mean the numbers should be ignored. Building your own basic models and comparing the underlying performance of two teams gives you a much stronger starting point, particularly when the aim is not necessarily to predict the winner, but to decide whether the price being offered accurately reflects the chance of something happening.

I started many years ago by manually entering data from numerous websites into my model. It took hours, but it create an understanding our how each team was performing from shots, shots on target, shots in the box and a very crude xG output. The model was basic but it gave me more than just league position and current form.

The danger comes when data starts being treated as certainty. It isn't, and even professional clubs working with far greater resources still have to interpret what the numbers are telling them rather than blindly following the output. Data improves the information behind a decision, but it doesn't remove the uncertainty from football.

What data can tell you

It gives you a baseline probability, not a hunch. This is the starting point for most betting models. Historical scoring and conceding data gives you an estimate of how likely different outcomes are, rather than simply deciding which team you think will win based on recent results or what you watched the previous weekend.

Dixon Coles style Poisson models take those attacking and defensive rates, adjust them for factors such as opposition strength and home advantage, then produce a distribution of possible scorelines and outcomes. If your model makes an outcome 55% and the available odds imply a probability of 45%, you have identified a potential edge worth investigating, rather than simply backing something because you fancy it.

Data separates performance from results

This is one of the biggest reasons I use underlying data including xG, and expected points, because the final score does not always give an accurate reflection of what happened during the previous 90 minutes. A team can dominate a match, create several good chances and lose, while another can be outplayed for long periods, score from its only meaningful opportunity and leave with three points.

Last season gave us two good examples. In my article on Premier League data it tells us that Sunderland finished seventh in the Premier League while consistently conceding fewer goals than their xGA suggested at both home and away, while Crystal Palace dropped to 15th despite creating chances at a top half level because their conversion rate collapsed. The league table showed where both teams finished, but the underlying numbers gave you considerably more information about how they got there and whether those results were likely to continue.

Data helps remove bias

Football is a low scoring sport, which means individual moments have a disproportionate influence on how we remember a match. A last minute winner can completely change the conversation around a performance, while a wonder goal can turn an otherwise poor attacking display into three points and make the tactical approach look considerably better than it was.

That creates outcome bias, where performances are judged by the result rather than the process that produced it, and this is one area where data becomes particularly useful. It records what happened across the whole match rather than placing excessive importance on the handful of moments our memory naturally holds onto.

Data shows where a team's threat comes from

Team level numbers only take you so far, particularly when looking at goalscorer and other player markets. Individual shot volume, shots on target, touches in the box, xG and chance creation help identify which players are consistently getting themselves into dangerous positions, rather than simply looking at who happened to score in the previous few matches.

The same applies defensively. A team keeping plenty of clean sheets while conceding high quality chances might be relying heavily on good goalkeeping or poor opposition finishing, while another allowing very few shots and little xG has a much stronger defensive process behind its record. Two teams can therefore have identical goals against numbers while arriving there in completely different ways, which matters when trying to work out what happens next.

Data spots unsustainable results

The league table tells you what has already happened, but it does not tell you whether those results matched the performances underneath them. If a team is sitting fifth while consistently losing the shot count, shots on target, shots inside the box and xG battle, there is a clear reason to investigate whether its league position is sustainable rather than simply assuming the table is an accurate measure of quality.

The same applies at the other end of the pitch. A side repeatedly scoring from a small number of chances and running well ahead of its xG might have excellent finishers, but maintaining an extreme conversion rate becomes increasingly difficult over a larger sample. That doesn't tell you the exact match where results will change, but it highlights teams whose underlying performance is moving in a different direction to the scorelines.

Data works beyond the match result

The same approach applies to markets away from the straightforward match winner, with corners, cards and both teams to score all producing repeatable patterns once the sample becomes large enough. A side consistently conceding high corner numbers regardless of opposition gives you something measurable to investigate, while the same applies to teams repeatedly involved in games where both sides score or fixtures producing unusually high card numbers.

The important part is understanding why the pattern exists rather than finding a percentage and treating it as enough on its own, because a statistic becomes considerably more useful when the underlying style of the team or fixture explains why it keeps happening.

What data can't tell you

It can't overcome a small sample. This is particularly important at the start of a season, when three or four matches can produce numbers that look significant without giving you enough evidence to know whether anything has genuinely changed. One red card, an unusually strong opponent or a freak scoreline has far more influence over an average when you only have a handful of matches to work with.

Recent form still matters, particularly after a managerial change or major squad rebuild, but it needs to be balanced against a much larger body of previous evidence rather than immediately replacing it. Serious models account for that by gradually increasing the importance of current season information as the sample grows.

Data doesn't know what is happening off the pitch

Historical numbers do not know a striker is carrying a knock, a manager plans to rotate half his team or a player is pushing for a transfer, while weather, injuries, suspensions and late team news can completely change the conditions a model was built around. This is why press conferences and confirmed line ups still matter regardless of how detailed your statistical work becomes.

Motivation is harder to measure again. A player who has fallen out with his manager, wants a move or simply has his mind elsewhere might perform differently without anything in his previous numbers indicating it beforehand, because data is excellent at recording what somebody has done but far less useful at telling you what is going on inside their head before the next match starts.

Football still contains a huge amount of variance

A strong model does not mean the favourite always wins, because football contains too much randomness for any set of numbers to remove the possibility of an unlikely result. Research comparing spending and league position across European football found wage spending explains around 70% of the variation in league position across a single season, leaving a substantial amount that cannot be explained by financial strength alone.

Underdogs win, goalkeepers produce outstanding performances, teams score with their only shot and favourites miss several big chances, none of which means the underlying numbers were necessarily wrong. A model is supposed to produce a range of possible outcomes with different probabilities attached to them, and sometimes the 20% outcome is the one that happens.

Not every metric means exactly what you think

Different data providers define events in different ways, which is worth remembering when building models from several sources or comparing numbers from different sites. A survey of top level clubs and national federations found only around a third believed the definitions supplied by commercial data providers were sufficiently clear, despite these being people working with the information professionally.

If people inside professional football do not always agree on exactly what individual metrics are measuring, there is good reason for bettors to avoid treating one statistic as definitive evidence. The stronger approach is to use several related numbers and look for the same pattern appearing across them.

Models create false precision

A model producing a 57.43% probability looks far more precise than saying something has roughly a 57% chance, but the extra decimal places do not mean the estimate itself is more accurate. Change a few reasonable assumptions around weighting, home advantage or team strength and another perfectly defensible model might produce 52% or 62% for exactly the same outcome.

Wolves were a good example last season. The model correctly identified them as a poor side, but their away results became even worse than the underlying process suggested, with their conversion rate from shots inside the box running at around half the league average. The broad assessment was right, but the eventual scale of the collapse was still difficult to predict.

More data does not automatically solve the problem either, because large datasets create thousands of possible patterns and some will appear meaningful purely through coincidence. A model built around every historical relationship it finds risks becoming excellent at explaining what has already happened without being particularly good at predicting what happens next.

Data doesn't understand context

A team with nothing left to play for is in a different situation to one fighting relegation, while a club playing a league fixture between two major European matches might approach it differently to the way its season long numbers suggest. The same applies to managers giving minutes to fringe players, clubs prioritising a cup competition or squads affected by issues that do not appear anywhere in the statistical record.

The data itself also comes from a football environment shaped by its own biases, including recruitment and selection decisions that influence which players receive opportunities in the first place. The numbers are excellent at describing what happened, but they do not always explain why it happened or whether the same circumstances will exist next time.

The takeaway

Data gives you a stronger starting point than relying on league tables, recent results or instinct alone, because it gives you probabilities to compare against bookmaker prices and helps identify teams whose performances are better or worse than their results suggest. It also highlights attacking and defensive patterns hidden by the final score, particularly when several different metrics all point towards the same conclusion.

But the numbers still need context. Sample size, team news, injuries, motivation and the limitations of the model all matter before placing a bet, and the aim should never be to search through enough statistics until you find something that supports the opinion you already had.

The real value comes from using the data to form the opinion in the first place, then asking whether the probability you have arrived at differs enough from the bookmaker's price to justify taking the bet. Data does not make the decision for you, but it gives you far more information to make it with.

🔗 Share on Social Media