Why naive before-and-after fails
Managers are often replaced after extreme poor performance. Regression-to-the-mean research explains why repeated measurements selected at an extreme can improve even without a causal intervention. Upcoming opponents, injuries and home-away mix can also change.
Define the causal question
Specify whether the target is points, goal difference, xG difference or forecast residuals, and over what fixed post-change period. Decide whether caretaker appointments and off-season changes are included before viewing results.
Comparison design
The manager-change study provides a counterfactual design reference for the workflow below.
- Estimate pre-change team strength using only earlier matches.
- Match changed clubs with similar non-changing club-periods or use another declared counterfactual method.
- Adjust for opponent, venue, dismissals and schedule.
- Report the full effect distribution and uncertainty.
- Repeat on later seasons or leagues.
A study of in-season Premier League changes used a comparison design and found no average in-season performance effect in its stated 2000/01 to 2014/15 sample. That finding should not be turned into a universal claim about every manager, league or mechanism.
Descriptive record
If a club moves from 0.8 to 1.4 points per match across five-match windows, the raw difference is 1.4 - 0.8 = 0.6. Publish fixture strength and uncertainty before attributing the difference to management.
Forecast use
Treat manager change as a timestamped feature whose relationship can vary with appointment type, squad and time. Validate on future changes; do not tune the post-change window after seeing when results peaked.
Worked comparison design
The manager-change study provides a counterfactual design reference for the workflow below.
Define an event date and a fixed post-change window, such as the next five eligible league matches. For each changed team, estimate a pre-change strength baseline using only information available before the appointment. Match it with comparable team-periods that did not change manager, or use a model with team strength, venue and opponent controls.
| Design risk | Control or sensitivity check |
|---|---|
| Change follows a losing run | Model pre-change trend and regression to the mean |
| Easier fixtures follow | Adjust for opponent and venue |
| Transfer window overlaps | Record squad changes separately |
| Caretaker period varies | Predefine inclusion rules |
| Multiple changes | Decide whether later events censor the window |
Interpret the estimate carefully
Report the average difference, uncertainty and the distribution across clubs. An average short-term improvement does not mean every appointment causes improvement or that the effect persists. Results, goal difference and xG difference can answer different questions, so declare the outcome before analysis.
For forecasting, test whether a manager-change indicator improves later-match probabilities beyond the strength and schedule baseline. The indicator should be timestamped to the announcement information actually available. Narrative explanations can follow the estimate, but they should not be used to recode the sample after outcomes are known.
Keep permanent appointments, caretaker spells and off-season changes visible as separate groups. Combining them may increase the sample while erasing the practical distinctions a reader needs.
Related resources
Use form tables for window design or limits of statistics for causal boundaries.
Continue learning
- Next guide: Poisson Football Score Model
- Related guide: Weather and Pitch Conditions in Football Models
Assumptions and limitations
Observational manager changes are not random and data on tactics, training or injuries can be incomplete. Average effects can hide heterogeneous appointments, while individual success stories do not establish a general rule.
GambleAware
GamCare
Gambling Therapy
GAMSTOP