Regression to the Mean vs Gambler’s Fallacy in Football

“Does regression to the mean mean a hot football team is about to start losing?” Regression to the mean means unusually strong results are likely to become less extreme if they were partly driven by temporary or random factors. It does not mean the team is “due” a defeat.

That second idea is the gambler’s fallacy, regression’s frequently confused neighbour. Both can point a bettor towards opposing a winning streak, but they arrive there by different routes. One is a statistical expectation based on repeatable performance; the other incorrectly treats previous outcomes as creating a debt that future outcomes must repay.

A hot streak can contain improvement, luck or both

Suppose a team wins eight league matches in a row. Calling it “hot” describes what happened, not why it happened. The sequence may reflect a genuine rise in playing strength: a better coach, fit attackers, an effective tactical change or several returning defenders. It may also include penalties, deflected goals, weak opponents and finishing that cannot reasonably continue at the same rate.

Regression to the mean concerns the temporary part.

The “mean” is the level around which results or performances would be expected to settle over a larger sample, assuming the relevant conditions remain broadly similar. It is not necessarily the team’s long-term historical average. A promoted side with an excellent new manager should not be dragged mentally towards figures recorded under an old manager two divisions lower.

Nor does regression say the next match must be poor. It concerns the likely direction of future averages. A striker who scores ten goals from ten ordinary chances may score again on Saturday, but his conversion rate over the next 30 similar chances is unlikely to remain at 100%.

The football complication is that team quality is constantly moving. Players get injured, tactics develop, confidence affects decisions and schedules vary in difficulty. There is no fixed ability figure sitting in a drawer at the club. A bettor has to estimate the team’s current underlying level, then decide how much of the observed streak exceeds it.

Useful evidence includes:

• Chance quality created and conceded, rather than goals alone
• The strength and location of recent opponents
• Penalties, own goals, red cards and late goals
• Injuries, suspensions and changes to the starting side
• Whether improved results began with a credible tactical or personnel change
• How performance looks across a longer sample

Expected goals, usually shortened to xG, can help because it estimates the quality of chances. It is not an oracle. Different models disagree, and xG can miss tactical context, game state and certain repeatable player skills. Still, seven wins accompanied by modest chance numbers deserve different treatment from seven wins built on sustained territorial and chance superiority.

Regression to the mean is a reason to reduce the weight given to extreme recent results, not an automatic reason to bet against the team producing them.

The gambler’s fallacy invents a result that is “due”

The gambler’s fallacy says a run must soon reverse merely because it has continued for a while. A team has won six consecutive matches, so a loss is due. A striker has failed to score in five, so he must score next. Neither claim follows from the sequence alone.

Imagine a fair coin lands heads six times. The next toss remains 50% heads and 50% tails because the coin has no memory. Tails is not more likely to restore balance. Over thousands of tosses, the overall proportion may move closer to 50%, but that happens because many ordinary future observations dilute the unusual run, not because the next toss is forced to compensate.

Football matches are not independent coin tosses. Results can change confidence, selection, fatigue and the prices offered by bookmakers. Even so, the same error appears whenever someone treats a streak itself as evidence that the opposite outcome has become more likely.

Regression and the gambler’s fallacy do share one sensible warning: do not project an extreme sequence indefinitely. An eight-match winning rate of 100% is not a sound forecast for the next eight matches. But only regression asks what sustainable level sits beneath those wins.

Consider two sides that have each won six in a row.

The first has dominated strong opponents, created far more high-quality chances than it has allowed and improved after recruiting a suitable defensive midfielder. Its sustainable level may genuinely be much higher than its earlier season average. Six more consecutive wins would still be unlikely, but continued strong performance would not be surprising.

The second has faced four bottom-half teams, scored from two deflections, benefited from three opposition red cards and conceded enough chances to lose several matches. Its perfect record is much further above its plausible playing level.

A gambler’s-fallacy bettor opposes both because each has won six. A regression-based bettor distinguishes between them, then asks whether the market price already reflects that distinction.

That final step matters. Bookmakers do not usually ignore a visible run. Customers like backing winners, especially recognisable teams whose matches have recently been televised. Trading teams know this and may have room to shade the fashionable side’s price—shorten the odds slightly—without losing demand.

If a team’s fair win probability is estimated at 50%, fair decimal odds are 2.00 because 1 divided by 0.50 equals 2.00. A bookmaker offering 1.80 implies 55.56% before allowing for the margin across all outcomes. The relevant question is not whether the team can win. It plainly can. The question is whether its chance is greater than the probability represented by the available price.

The dividing line is evidence about sustainable performance

The cleanest way to separate the two concepts is to ask what would change your mind.

If you are opposing a hot team only because “runs always end,” you are relying on the gambler’s fallacy. A run can end next week or continue for another two months. Its length does not set a deadline.

If you are opposing it because goals have greatly exceeded chance quality, the schedule is about to become harder and the price assumes the recent conversion rate will persist, you have a regression argument. That argument can still be wrong, but it is testable.

Regression is strongest where three features meet:

• The recent measurement is extreme
• The sample is small enough for randomness to have a large effect
• There is a more reliable estimate of the team’s underlying level

Goal difference over four matches is highly volatile. Shot quality across 20 matches is generally more informative, though never complete. Penalty conversion, opponent finishing and red-card timing can heavily distort a short run.

Be careful with the choice of mean. Pulling every team back towards the league average ignores real differences in quality. Manchester City at their strongest and a relegation candidate do not share the same baseline merely because they play in the same competition. The target should combine longer-term performance with current information.

There is also a market mean, in effect: the level already assumed by the odds. Public discussion often identifies likely regression correctly but reaches the betting market late. If everyone has noticed that a side keeps winning despite poor underlying numbers, the next price may already be bigger.

From the bookmaker side, that is where customer behaviour matters. Traders price the event, add margin and adjust for expected money. A popular hot team may be shorter than a pure model suggests because recreational money tends to follow visible form. Yet a fashionable “regression” story can also become crowded, particularly among statistically minded bettors. There is no permanent value in either camp.

The same discipline applies when reading football betting predictions: separate the forecast about team strength from the judgement about whether the quoted odds are generous enough. Correctly predicting that performances will cool does not guarantee a profitable bet. Variance remains, and betting always carries the risk of losing the stake.

Three questions that expose a false regression argument

Does regression mean I should back the draw or away team?

No. It means the hot team’s future results may be less impressive than its recent record, not that a particular opposing selection is automatically good value.

A team can regress from winning 80% of matches to a sustainable 60% and still deserve to be a strong favourite. If its true win chance is 60%, fair decimal odds are about 1.67. Backing its opponent merely because the streak looks unsustainable could still be a poor decision.

You need to estimate all relevant outcome probabilities and compare them with prices. In a three-way match market, also remember that converting each bookmaker price into an implied probability will usually produce a total above 100%. That excess is the bookmaker’s margin, so the displayed implied figures are not all fair probabilities.

Can better players consistently outperform expected-goals figures?

Yes, to a degree. Elite finishers can score more than an average finisher would from the same recorded chance quality, and strong goalkeepers can save more than average models expect. Team style can also create details that broad models do not capture perfectly.

The mistake is assuming every short burst proves a new, repeatable skill. A forward scoring eight times from chances worth roughly four expected goals may be an excellent finisher, unusually fortunate, or both. Career history, shot type, age and role provide better evidence than the eight goals alone.

Regression does not require everyone to become average. It predicts movement towards an appropriate individual or team baseline. For an elite striker, that baseline may remain well above the competition average.

What if the winning run has clearly improved the team’s confidence?

Confidence can be real without making the run self-sustaining. Players may make quicker decisions, attempt difficult passes or cope better with pressure after several wins. Opponents may also become more cautious.

Treat confidence as one possible change in the underlying level, not as permission to extrapolate a perfect record. Ask whether there is observable support: more aggressive pressing, better movement, improved availability or stronger chance creation. If the only evidence for confidence is that the team keeps winning, the reasoning becomes circular.

Confidence is also fragile and hard to price. A model may understate it, while supporters and media may greatly overstate it. Either error can matter.

Choose regression for pricing and streak logic for nothing

A bettor should want regression analysis when recent form is extreme and there is enough underlying information to estimate a more sustainable level. It is particularly useful for teams with large gaps between goals and chance quality, unexpectedly high conversion rates or records inflated by unusual match events.

You should want evidence of genuine improvement when considering backing the hot team. A stronger squad, repeatable tactical edge, healthy first-choice players and good numbers against credible opposition can justify moving the baseline upwards. Regression may still reduce a projected 100% win rate, but it need not erase the upgrade.

The gambler’s fallacy is not the right tool in either situation. It offers no probability estimate, no price and no explanation beyond the streak’s existence. “They cannot keep winning” may eventually be true, yet a bet can lose several times before that observation is vindicated—and the odds may never have offered value.

Before placing a bet, check whether the hot run is supported by sustainable performance, whether the opposition and team conditions have changed, and whether the available odds already account for likely cooling. If those checks do not produce a probability that beats the price after allowing for margin and uncertainty, there is no obligation to bet.