The new manager bounce in football: signal or regression to the mean?
By FootInsights · Published · 6 min read
A club loses six of eight, the manager goes, and the replacement wins their first two. The new manager bounce in football is the most reliably repeated causal story in the sport, and unlike most punditry it comes with numbers attached: teams really do pick up more points after a dismissal than before one. That part is not in dispute.
What is in dispute is whether the manager caused it. The before-and-after gap is real and the causal claim built on top of it is mostly an artefact of how the comparison is constructed — and understanding why is a good way to learn what separates a football statistic that means something from one that only looks like it does.
What the bounce looks like in the raw numbers
Take an illustration with round figures. A team averages 0.7 points per game across the eight matches that get its manager sacked. Under the replacement, it takes 1.3 points per game over the next five. That is an 86% improvement, arriving immediately, exactly when the new appointment did.
Presented that way it is nearly impossible to read as anything but cause and effect. Every element of the story lines up: a clear intervention, a clear before, a clear after, and a plausible mechanism in fresh ideas and a dressing-room reset.
The problem is the “before”. It was not a random sample of that team’s matches. It was the worst run the team had produced — which is precisely why the sacking happened.
Regression to the mean does most of the work
Suppose the team’s true strength is worth 1.35 points per game. Football is low-scoring and one goal decides most matches, so a team’s actual return over any short window scatters widely around its true level. An eight-match stretch at 0.7 is not evidence that the team is a 0.7-points-per-game side. It is a normal, expected excursion for a 1.35 side having a bad run.
Now add the selection step. Managers are not dismissed at random points in the season; they are dismissed at the bottom of an excursion. That means the “before” window has been deliberately chosen for being extreme, and anything chosen for being extreme tends to be followed by something less extreme. Not because the situation was corrected, but because the extreme was partly noise, and noise does not persist.
So the honest counterfactual for that team is not 0.7 points per game. It is something close to 1.35 — the level it would have drifted back toward with nobody sacked at all. Measured against that baseline, the observed 1.3 is not an 86% improvement. It is slightly below where the team was heading anyway.
This is the central error, and it is worth stating plainly: the comparison everyone makes (before versus after) is not the comparison that answers the question (with versus without).
The fixture run is not random either
There is a second confound stacked on the first, and it points the same way.
Sackings cluster after difficult stretches — a run against the top of the table, an away-heavy schedule, a period of cup and league fixtures compressed together. A club that has just played its four hardest fixtures of the season and lost them all is a club whose next four are, mechanically, easier than average. The new appointment inherits that easier run and gets credited with the points it produces. Work examining fixture difficulty around managerial changes has generally found that incoming managers face a lighter-than-average opening schedule more often than not.
Calendar effects compound it. Dismissals bunch into specific windows, and those windows sit next to breaks that give a squad recovery time and training days a mid-season week never provides. None of that is managerial skill, and all of it lands inside the “after” column.
The control group that settles the question
The fix is not a cleverer statistic. It is a control group.
The right comparison takes every club that sacked its manager after a bad run and matches it to clubs that had an equally bad run and kept theirs — same league, similar squad quality, similar points deficit, similar remaining schedule. Then it compares what the two groups did next. That is the design that isolates the manager from everything else the sacking is entangled with.
Studies built this way do not all agree, and it would be dishonest to claim they do. What they consistently show is that the raw before-and-after gap shrinks dramatically once the comparison group exists — the teams that kept their manager also improved, for the same statistical reasons. Whatever is left over after that is small relative to the run-to-run variance of a football team, which is another way of saying it is hard to measure and easy to overstate.
“Smaller than it looks and hard to pin down” is a much less satisfying headline than “the bounce is real”. It is also the conclusion the evidence supports.
What this means for a prediction model
A model that reads only results has no field for who is standing on the touchline, and no adjustment fires when a club changes manager.
The cost of that is real and worth naming. If a new appointment genuinely does change a team’s level, such a model learns it only as new results arrive, weighted by how quickly recent matches are allowed to count. For a stretch of matches after a real change, the model is pricing a team that no longer exists in quite that form.
The alternative is worse, though. Adding a “new manager” adjustment requires a number: how much, for how many matches, and for which kinds of club. The evidence above does not supply a stable one. A hand-tuned boost would encode a guess about an effect that mostly dissolves under a control group, and it would fire on every dismissal — including the majority where nothing changes except the noise reverting. A feature that carries no clean signal does not sit harmlessly at zero; it fits noise, confidently. The known limits of what our model can see are written up on how our predictions work.
There is a narrower case where a change genuinely does move the numbers, and it has nothing to do with motivation: a manager who changes the team’s tactical approach in a way that alters how many goals the fixture produces. That shows up in results within a handful of matches and a decay-weighted model absorbs it — slower than a human observer would, but without needing to guess in advance.
How to check any claimed bounce yourself
The same four questions work on any before-and-after football statistic, not just this one.
- What is the counterfactual? If the comparison is the team’s own bad run, there is no counterfactual, and regression to the mean explains the result before the manager gets a chance to.
- Was the “before” window selected for being extreme? If the window ends at the sacking, then yes, by construction.
- How hard were the fixtures either side? Compare opponent strength, not just points. An easier run produces more points with the same manager.
- Was the “after” window chosen after seeing the results? Five matches, ten matches and end-of-season all give different answers, and picking the flattering one afterwards is the oldest trick in the genre.
A claim that survives all four is worth taking seriously. Most of what gets published about managerial change survives none of them.
The broader habit matters more than this one topic. Football produces a constant supply of plausible causal stories with real numbers behind them, and the numbers are usually accurate while the story attached to them is usually wrong. The only defence is to ask what the comparison group is — which is the same standard we hold our own predictions to on the public track record, where every call is settled after full time and the misses stay up alongside the hits.
Treat probabilities as probabilities, and stake only what you can afford to lose.