Availability Is the Best Ability

In January 2021, Liverpool’s centre-back options were Virgil van Dijk (out with a cruciate ligament tear), Joe Gomez (recovering from knee tendon damage), and Joël Matip (absent with a serious ankle injury). The reigning champions started Nathaniel Phillips and Rhys Williams, two players with a combined zero Premier League starts before that season, in a Champions League knockout tie. Phillips affectionately came to be referred to as 'The Bolton Baresi' in that time. They finished third. A year earlier, with everyone fit, they had won the league by eighteen points.

Sports science departments have known for years that injuries affect results. Hägglund, Waldén, and Ekstrand published the first large-scale evidence in 2013. Every Premier League club now employs GPS analysts, load management specialists, and soft-tissue physios. None of what follows will surprise them.

But most public analysis of injuries still trips over a basic statistical problem that makes the relationship invisible. We built a dataset of two million matchday entries from Transfermarkt, covering 858 club-seasons across six European leagues over ten years, to show what the raw numbers hide and to make the data available for anyone to replicate.

The confound

The flat scatter is misleading. Rich clubs have deeper squads, better medical departments, and more depth on the bench. When Manchester City lose a midfielder, they replace him with another player worth tens of millions. When a newly promoted side loses one, the alternative is a youth player or a January loan. Squad value predicts both health and results. Without controlling for it, the real relationship is hidden.

A panel regression with clustered standard errors strips out squad market value, European competition participation, manager changes, squad age, promotion status, and COVID disruption. In the Premier League, Bundesliga, and Serie A, each 10 percentage-point increase in injury burden is associated with roughly 4.5 fewer league points. That is the difference between a European place and mid-table. The effect is weaker or absent in La Liga and Liga Portugal, likely due to noisier injury reporting in those markets.

The result holds within clubs. A separate model with club fixed effects (the same club compared to its own ten-year average, rather than to other clubs) produces a nearly identical coefficient. When a club has a healthier season than its own baseline, it gains points. When it has a worse one, it drops.

Two caveats. First, causation: a temporal test (do early-season injuries predict late-season points?) gave a coefficient in the expected direction but fell short of significance (p = 0.20). A placebo test in the other direction was also non-significant but closer to the threshold (p = 0.09). We cannot rule out reverse causation. Second caveat, scope: our data captures domestic and European club fixtures but not international duty. Players returning from national team camps with injuries or accumulated fatigue are a real and unmeasured source of load that this analysis does not capture. The methodology page has the full regression tables and robustness checks.

The cascade

An injury does not stay contained. Player A goes down with a hamstring injury. Player B, previously rotating, now starts three games in seven days. Two weeks later, in training Player B pulls a calf. The squad thins from the inside out.

This is not anecdotal. Across 42,000 club-matchday observations, teams carrying a high injury load see new injuries among available players at roughly 2.5 times the rate of low-load squads in the following two to three weeks. These are genuinely new injuries, not the same player still being out. The association is unadjusted and we cannot rule out confounders like schedule density, but the pattern holds across leagues and seasons.

The cascade is strongest for muscle injuries (2.6x rate increase per available player) and weaker for trauma like fractures and concussions (2.1x). That pattern is what you would expect if the mechanism is overwork rather than bad luck. A broken leg is random. A hamstring after three games in a week is far likelier to be load-related.

This is why the first hamstring of the season is more "expensive" than it looks. It is not one player out for four weeks. It is the chain reaction that follows: the cover player overworked, the rotation options exhausted, the training schedule compressed. Clubs that manage the first injury cleverly, with versatile cover or smart rotation while absorbing the short-term quality drop (which otherwise may spiral into desperation), may break the chain before it starts.

Clubs below 17% show little systematic deviation from expected performance. Above 17%, the top decile finishes 3.4 points below expectations on average. This is a single decile (86 club-seasons), so the threshold should be treated as indicative rather than precise.

What the data says

Across 858 club-seasons, six leagues, and a decade of data, the clubs that finished highest relative to their spending were not the ones that avoided injuries entirely. They were the ones that stopped the first injury from becoming the third.

None of this, of course, is new to the people who work in sports science. The practical levers are well understood: manage training load between fixtures so the first XI can sustain three games in eight days without an acute spike. Protect forwards, who are hardest to replace. Monitor the cascade in real time, because the second and third injuries in a cluster are more costly than the first.

The harder question is perhaps why clubs do not always act on it. Manager tenure averages eighteen months, and this is getting shorter every season. The payoff from load management is measured in seasons. A manager who rests his star striker for a midweek cup tie and loses may not survive to see the benefit. The data says availability is a measurable competitive advantage. The incentive structure says short-termism is rational. That tension is not going away.

The underlying dataset (two million matchday entries, six leagues, eleven seasons including 2025/26 in progress) is available on GitHub for anyone who wants to replicate, extend, or challenge this analysis.

Methodology, full regression tables, and data sources →