Regression to the Mean in Football
Why a team's hottest or coldest stretch tends to drift back toward normal, how to spot it in football numbers, and why a good explanation of it is not a betting edge.
Picture a running back who scores four touchdowns on a Sunday. By Monday morning he is the most added player in fantasy leagues and the subject of every highlight show. By the following Sunday he has one short touchdown and a mediocre yardage total, and the same people who rushed to pick him up are asking what went wrong. Usually nothing went wrong. The week before was the unusual one.
That drift back toward ordinary results is called regression to the mean, and it is one of the most useful ideas in sports analysis and one of the most misused. It explains why great seasons are so often followed by merely good ones, why a team that wins five close games in a row is probably not as good as its record, and why a bad team that finally wins is not always "turning the corner."
This article explains what regression to the mean is, why it happens even though nothing mystical is going on, how it shows up in football numbers, and why understanding it does not turn into a guaranteed way to beat a betting line. It is educational only. Nothing here is a pick or a prediction, and sports betting is legal only in some places and only for adults of legal age.
What regression to the mean actually means
The "mean" is simply the average, the long-run typical result for a player or team. Any single result is a mixture of two things: the underlying level of ability, and luck. Luck here is not superstition. It means everything that varies from game to game for reasons the player or team does not control, such as a tipped pass landing in the wrong hands, a fumble that bounces out of bounds instead of to the defense, or a field goal that clips the upright and goes through.
When you pick out an extreme result, such as the best week, the worst month or the highest-scoring stretch, you are very likely picking a moment when luck pointed strongly in one direction. The next sample is a fresh draw. Luck on the next draw is not tied to luck on the last one, so it will most likely be closer to neutral. The ability did not change, but the luck went back to normal, and the results moved toward the average.
Three points are worth fixing in your mind before going further.
- Regression is not a force. Nothing pulls a hot team back down. The old luck simply is not repeated.
- Regression is not guaranteed in any single case. It describes what is likely across many cases. A particular player can have two big weeks in a row.
- Regression does not mean "due." A player who had a bad month is not owed a good one. The next sample is just a fresh draw, centered on his real level.
Why extreme results are mostly unrepeatable
A simple thought experiment makes this concrete. Imagine two hundred kickers who are all exactly equal in skill, and each kicks the same number of field goals from the same distances in the first month of a season. Some will finish above the group's overall average, some below, just by chance. Pick the ones who finished highest and call them the best kickers. In month two, with their skill unchanged, that group will on average land back near the overall average. The original group did not get worse. They were never better.
Real life is messier because real kickers do differ in skill. But the same logic holds in a weaker form. The top performers in any short sample are a blend of the genuinely skilled and the genuinely lucky, and you cannot tell the blend apart from one sample. The next sample keeps the skill and sheds some of the luck.
A useful way to say it: the more of a result came from skill, the less it regresses; the more came from chance, the more it regresses. The practical question for any statistic is therefore how much of it is stable skill and how much is noise.
Measuring how stable a number is
A simple test: for any number that looks impressive, ask whether the leaders in the first half of the games would also be the leaders in the second half. For some football statistics the answer is "mostly," and for others it is "barely." The ones that scramble from sample to sample are mostly noise, and results built on them will regress hard.
Where regression shows up in football numbers
Football is a sport with few games, high variance on individual plays and many events that depend on bounces. That makes it especially full of regression candidates. Here are the places where it appears most often.
Touchdowns
Touchdowns are rare, high-value events, and a single one swings a player's weekly line. A receiver who catches the same number of passes and gains the same yards as the week before can score one touchdown or none. Over a full season, touchdown totals tend to track opportunities, such as how often a player gets the ball near the goal line, better than they track last week's touchdowns. A player who scores far more than his opportunities would suggest is a regression candidate, and one who gets plenty of chances but scores rarely is often a candidate to improve. Our article on red zone opportunities and touchdown regression goes through that idea from the fantasy side.
Turnovers
Fumbles are the clearest case. A defense or offense can run into a stretch where loose balls repeatedly bounce one way. Research on football fumble recoveries has long described the recovery of a loose ball as close to a coin flip for either side, which is why a team that recovers nearly all of its fumbles over a few weeks is likely to see that fade. Interceptions are less random than fumbles because they depend on quarterback decisions and coverage, but part of an interception total, such as a pass that bounces off a receiver's hands into a defender's arms, is still luck. A team's turnover margin in one stretch is a poor guide to its turnover margin in the next stretch.
Close games and one-score results
A team's record in games decided by a small margin is a classic regression case. Imagine a team that wins six games, each by a field goal or less. Close games turn on a handful of plays: a late missed kick, a successful onside attempt, a fumble in the final minutes. If outcomes in such games were close to a coin flip, a team winning all or nearly all of them would be expected to win a smaller share of them later. This is why analysts compare a team's actual record with a record estimated from its points scored and allowed. Wins that exceed what the scoring margins suggest are the ones most likely to be followed by a slip. Our standings page shows win-loss records, and the scores page lets you see the margins behind them.
Special teams
Field goal percentage in a small sample, return touchdowns and blocked kicks are all high-variance events. A kicker who makes every attempt in a month may be good, or may have faced short kicks in good weather. Rare events regress hard.
Quarterback and skill-player efficiency
Yards per attempt, completion rates and yards per carry can all run hot or cold in a small sample. Sack rates and rushing averages are partly produced by the offensive line and the opponent, and partly by the occasional long run that inflates an average. A single 80-yard run can change a back's yards per carry for weeks. These averages settle only as the number of attempts grows.
The same idea applied to coaches, rosters and "new head coach bounce"
Regression also explains stories that sound like cause and effect. When a team with a bad record hires a new head coach and then improves, the improvement may be partly real and partly regression. A team that had an unlucky year, with injuries at key positions and a pile of narrow losses, will tend to improve the next year even if the coach changes nothing. The same goes in reverse for a team that wins a championship after a lucky year. It is rarely as strong the following season as its record said.
Coaching and roster building matter a great deal. The point is only that part of any big jump or drop is likely to be temporary, so be careful before crediting or blaming a single change for all of it.
A worked example, with made-up numbers
Everything in this example is hypothetical. Imagine a receiver who, over a long stretch of seasons, catches a touchdown on about one of every twenty-five targets. In a certain five-game stretch he is thrown thirty targets and scores five touchdowns, which is one for every six targets. A reasonable analyst asks how much of that rate to expect going forward.
If the player's real touchdown rate is one in twenty-five, thirty targets would typically produce about one or two touchdowns. Five is possible but unusual. The next thirty targets will most likely produce something closer to the usual one or two, because the underlying rate is the best guide and the five-touchdown stretch carried a lot of luck.
Now imagine the opposite case, a receiver with the same long-run rate who goes thirty targets with zero touchdowns. The same reasoning says not to assume he has "lost it." Zero touchdowns on thirty targets happens to ordinary receivers. In both cases the best guess for the future mixes the long record with the recent stretch, giving the long record much more weight, and the recent stretch less.
That mix is the practical form of regression. You do not throw out recent data. You shrink it toward the longer-run average, and you shrink it more when the sample is small and the event is rare.
Common misreadings of regression
Because the idea is popular, it is also often misused. Here are the usual mistakes.
- "He's due for a bad game." Regression does not work by balancing accounts. A player on a hot streak is not scheduled for a cold one. The expected next result is just closer to his real average.
- "Fade the hot team automatically." A team that has won many games may have won them because it is good. Regression says part of the record was luck. It does not say all of it was, and the question is always how much.
- "Regression means the best are not really the best." Genuine skill persists. The very best players and teams stay near the top across seasons. They just rarely repeat their single best result.
- "If I know about regression, I can predict the next game." Knowing the idea tells you which direction results tend to drift. It does not tell you the outcome of any one game, and it does not tell you how far results will move.
Why regression is not a betting edge by itself
This is where careful readers should slow down. Sportsbooks know about regression. The people who set and adjust point spreads and totals are professionals who study the same patterns, and the lines they post already account for the fact that a hot team is probably not as good as its record. A team's price reflects what the market thinks of its true strength, not simply its last few results.
That has several consequences.
- The lines already shrink. If a team has won six close games, the spread is typically set with a lower opinion of that team than its record suggests. Betting against it because "it must regress" can mean agreeing with a number that has already regressed it.
- Public attention can push lines in either direction. A very popular team can attract so much interest that a book moves its price to balance the action, which is a business choice rather than a statement about quality. You can read more about how this works in our guide to why betting systems fail, and see lines for upcoming games on our odds page.
- A small edge is easily swallowed. Even if you are right about direction, the sportsbook's built-in margin and the randomness of a single game mean that being correct about regression does not translate into winning bets. Our article on sample size and small edges explains why you need an enormous number of bets to tell skill from luck.
- Football has few games. With an NFL regular season of well under twenty games, and similar or shorter schedules at the college and high school levels, the amount of data is small, so the estimate of a team's true level is itself uncertain.
In short, understanding regression makes you a better reader of football. It does not make the sportsbook's price wrong.
How to use the idea well
You can apply the concept in ways that have nothing to do with wagering.
As a fan
When a team's record looks surprising, compare it with its scoring margin and its results in close games. A team that is 8-2 with a large margin of victory is on firmer ground than one that is 8-2 with seven wins by a field goal. You can check scoring margins on our scores page and standings on the standings page.
As a fantasy manager
Look at opportunity, not just output. Targets, carries and snaps are more stable than touchdowns. A player who produced on few opportunities is a regression risk, and one who has many opportunities but few points may be undervalued. If you want to go further, the league leaders pages show raw production that you can compare with usage.
As someone who bets or is thinking about it
Treat regression as a reason for humility rather than a system. The most reliable lesson is that confident stories built on short streaks are fragile. If you choose to wager, set a limit you can afford to lose, treat the outcome as entertainment spending, and do not chase losses. Sports betting is legal only in some places and only for adults of legal age. If gambling is causing you or someone you know worry, the National Problem Gambling Helpline (1-800-GAMBLER) offers confidential help.
A short checklist for any hot or cold streak
When you see a streak, run through these questions.
- How many games, attempts or events are in the sample?
- Is the statistic driven by rare events such as touchdowns, fumbles or long plays?
- Is there a longer record to compare against, such as the previous season or career average?
- Did the opportunities change, or only the results?
- Was the schedule unusually easy or hard during the streak?
- Did something real change, such as an injury, a new starter or a scheme change, that could make the new level the true level?
That last question matters. Regression applies to luck, not to real changes. If a team loses its starting quarterback, its lower scoring may be a genuine decline and not noise. The skill is knowing when the numbers have changed because the team changed and when they have changed because the dice did.
The short version
- Regression to the mean is the tendency for unusually high or low results to be followed by results closer to the long-run average, because the luck behind them does not repeat.
- It is not a force and not a guarantee. It describes what is likely across many cases, not what will happen in a given game.
- Football is a high-regression sport: few games, rare scoring events and bouncing balls mean that touchdowns, fumble recoveries, close-game records and special teams plays swing widely.
- Measures built on opportunity, such as targets, carries and snaps, are steadier than measures built on outcomes, such as touchdowns.
- Real changes, such as injuries or a new starting quarterback, are not regression, so check whether something genuine changed.
- Sportsbooks already account for regression, so knowing the concept is not a betting edge. Treat wagering as entertainment with a fixed limit, and call 1-800-GAMBLER if it stops being fun.