Rest and Recovery Impact on Shooting Performance

Today I want to share research I conducted after being inspired by a recent post exploring the relationship between player rest and shooting performance in football. This wasn't straightforward to analyze, but I believe there's value in the findings.

To summarize my results upfront:

First of all, what I did was to scrape all the schedule of teams and competitions in season I could given my event datasets. So many leagues from 2021 to this season, I also decided to exclude the 2021 season being the one of the covid, and I had to sacrifice MLS, Liga Profesional, Brazilian Serie A and Liga Portugal for different reasons, with the last two being because I could not scrape their full schedules. For schedules we are of course talking about all the games from teams in that competition, League Cup or Cups, Super Cups, League and those that participate, European Competition. So I decided to also leave MLS and Liga Profesional out of it.

There was another problem, I had to handle those games in which I have European teams I only have events data for their continental competition. In that case I kept only event data for the teams I also have leagues event data. For Example, I have Slovan Bratislava Champions League event data, but not the league so we exclude this type of teams. This also pose a problem when we watch difference in rest.

Methodology

I scraped team schedules across multiple competitions from 2022 to the current season. I excluded the 2021 season due to COVID-19 disruptions, and had to omit MLS, Liga Profesional, Brazilian Serie A, and Liga Portugal for various data collection challenges.

For each team, I tracked all matches: league games, cup competitions, super cups, and continental tournaments. One limitation arose with European competitions where I only had event data for continental matches but not domestic leagues for certain teams. In these cases, I restricted analysis to teams where I had complete data sets.

This creates some data reliability issues. For example, when Milan played against Slovan Bratislava, we have accurate rest data for Milan but not for Slovan. While this skews the data slightly, the overall trends remain consistent even when excluding continental competition data entirely. Even though I’d prefer to leave this specific metric alone as I cannot find a reasonable cause to analyze for why teams with more rest than their opponents do worse than those with the same amount.

Weighted xG Approach

I'm using "goals - weighted xG" as my performance metric. Weighted xG is calculated by multiplying standard xG by the shooting team's offensive rating (derived from my Dixon and Coles model). For instance, if Liverpool takes a 0.3 xG shot and their offensive rating is 2, that shot represents 0.6 weighted xG.

Without this weighting, the results appear counterintuitive:

It initially seems strange that teams with less rest perform better, but this reflects how elite teams (who generally have superior shooting ability) play more frequently. By using weighted values, we get more sensible results showing that teams with less rest typically underperform:

The analysis reveals that teams playing with 3-6 days of rest achieve shooting performance at approximately 93% of those with 6+ days rest. Teams with 0-3 days rest perform at around 89% efficiency.

Seasonal Factors

Following previous Ali’s approach, I also considered cumulative fatigue effects by dividing the season into distinct periods:

The data shows teams shoot most effectively at the beginning of the season (August through October). Performance declines from November through December, improves slightly from January through March, and edges up again from April onward.

This pattern aligns with rest patterns: during the holiday season, teams have a median of 125 hours between matches, compared to 149 hours at both the beginning and end of the season. However, the total accumulated match time also differs significantly across these two periods.

Teams have slightly less rest in January-March (median 144 hours) than at season's end, but many clubs are eliminated from cup competitions from that period, potentially creating more recovery time. I expected stronger performance at season's end compared to mid-season, but cumulative fatigue appears to be a significant factor.

Interestingly, when using unweighted xG values, late-season performance does indeed exceed all other periods:

This analysis reveals several key insights about rest and shooting performance in football. First, adequate recovery time between matches does matter — teams with 6+ days of rest demonstrate superior shooting efficiency compared to those with minimal recovery time. Second, the seasonal cycle shows clear patterns, with early-season freshness yielding the best performance, followed by a mid-season dip and a slight recovery toward season's end.

The weighted approach helps control for team quality, revealing the true impact of rest that might otherwise be masked by stronger teams simply playing more matches. While this methodology has limitations — particularly regarding data availability for some competitions and teams — it provides compelling evidence that rest intervals significantly impact shooting performance.

I would've liked to find a way to include these findings in my Dixon and Coles model to create better ratings, but I haven't developed a convincing method to incorporate these values without sacrificing data from competitions I consider valuable from a scouting perspective, such as the Brazilian Serie A, Liga Portugal, and MLS. The challenge remains in finding an effective way to incorporate both the seasonal period effect and the specific rest intervals between games while maintaining model integrity and data coverage. I think this will be one for the future.