Impact of the Euro 2020 championship on the spread of COVID-19

Passion for competitive team sports is widespread, and the tradition of watching and celebrating matches together may pose risks to COVID-19 mitigation, especially in large or indoor gatherings. Prior events under substantial restrictions showed minor transmission impact, but large, highly publicized tournaments with stadium attendance, increased travel, and viewing parties can substantially affect spread—particularly when few restrictions are in place, as during UEFA Euro 2020 (June 11–July 11, 2021). While stadium attendance itself may be minor, it likely increases engagement and social gatherings, suggesting considerable impact. The Euro 2020 is well suited to quantify effects because match timing and team progression are largely independent of national epidemic conditions, resembling a quasi-randomized setting. Moreover, attendance at match-related events is expected to be gender-imbalanced, enabling identification of match effects using gender-resolved case data. The study aims to quantify how much Euro 2020-related gatherings contributed to COVID-19 spread and under which epidemiological conditions their impact is largest, thereby informing mitigation strategies for mass events.

The paper references evidence that spectator sports under strict contact restrictions had minor effects on COVID-19 transmission, whereas large-scale events with extensive media coverage and viewing parties can amplify spread. Stadium attendance per se may have a small effect, but it likely increases social mixing and viewing parties. Early observational analyses and news reports indicated gender-imbalanced participation in football-related activities, supporting the use of gender-resolved data to disentangle effects. Previous evaluations of sports events’ impacts were inconclusive when focusing on in-stadium transmission; this work aligns by finding hosting effects small for Euro 2020, though effects could be larger for single-country championships accompanied by extensive ancillary events and international travel.

The authors develop a hierarchical Bayesian model to quantify the effect of Euro 2020-related gatherings on COVID-19 spread across 12 countries with daily gender-resolved case data (England, Czech Republic, Italy, Scotland, Spain, Germany, France, Slovakia, Austria, Belgium, Portugal, Netherlands). The epidemic dynamics in each country are modeled via a discrete renewal process with daily time steps and separate compartments for male and female populations. The effective reproduction number R_eff(t) is decomposed additively into: (1) a slowly varying base component R_base(t) reflecting non-football-related dynamics (NPIs, behavior, immunity), (2) a match-induced component ΔR_football(t) concentrated on match days and allowing gender imbalance, and (3) a slowly varying gender-asymmetric noise term ΔR_noise(t) capturing other imbalances unrelated to football. Gender interactions are implemented via symmetric contact matrices for base, match, and noise components. Prior for female participation in match-related gatherings centers at 33% (Beta(10,20)), implying a higher per-capita match-day effect among men. The match effect is modeled per match m with a hierarchical structure: ΔR_match,m = α_prior,m (ΔR_mean + Δα_m), where α_prior,m indicates participation; Δα_m ~ Normal(0, σ_α), σ_α ~ HalfNormal(5), and ΔR_mean ~ Normal(0,5). An extended variant also considers hosting-related effects ΔR_stadium,m with analogous hierarchy for matches held within a country. R_base(t) is parameterized by a series of logistic change points approximately every 10 days with priors on timing, scale, and effect (Δy_n) to capture gradual shifts in transmission. The ΔR_noise(t) term similarly uses change points to allow smooth gender ratio variability. The model accounts for delays between infection and reported case using a Gamma-distributed kernel with country-dependent priors on median delay (reflecting whether dates correspond to symptom onset, sample collection, or report date) and an additional weekday-dependent recurrent delay to capture weekly reporting/testing patterns (separate priors for Tue–Thu vs Fri–Mon). The likelihood for observed gender-specific daily cases is a Student’s t-distribution with variance proportional to expected counts to accommodate overdispersion and outliers. Inference uses PyMC3 with NUTS sampler, multiple chains, and convergence assessed via R-hat (good for reproduction-number-related variables). The model estimates primary cases (infections occurring due to match-day gatherings) and subsequent cases (additional infections propagated from primary cases) through counterfactual simulations setting ΔR_football to zero. Cross-country averages are computed with hierarchical models that incorporate uncertainty from country-specific posteriors. Correlational analyses use Bayesian linear regression to relate Euro 2020-attributed cases to explanatory variables: initial incidence N0 (pre-event cumulative cases), pre-event reproduction number R_pre, event duration T (matches played), mobility changes, and a proxy for popularity (matches played/hosted). Robustness is explored by varying priors (generation interval, delays, change-point spacing, female participation prior), shifting match dates, and adding hosting effects.

• Average match-day effect: A single Euro 2020 match increased the reproduction number by ΔR ≈ 0.46 (95% CI [0.18, 0.75]) for one day across analyzed countries (excluding the Netherlands from averages due to confounding).
• Attributed cases: Across 12 countries, approximately 0.84 million COVID-19 cases (95% CI [0.39M, 1.26M]) were attributed to Euro 2020-related gatherings, corresponding to about 2200 cases per million inhabitants on average during June 11–July 31, 2021. Using contemporaneous case fatality risk, this implies about 1700 deaths (95% CI [762, 2470]) assuming age-independent effects.
• Primary vs subsequent: Primary infections (directly linked to match-day gatherings) are only a small fraction; subsequent infections dominate roughly 4:1. On average, only 3.2% (95% CI [1.3%, 5.2%]) of new cases during the analysis period were direct primary cases.
• Country heterogeneity: Strongest effects in England and Scotland. Scotland experienced a particularly strong single-match effect (Scotland vs England in London) with R_match = 3.5 (95% CI [2.9, 4.2]); subsequent cases from this match accounted for ~30% of Scottish cases in the following weeks. England, as runner-up (max matches), showed sustained and increasing effects toward later rounds; approximately 48% of cases until July 31 were related to the championship. Some countries (e.g., Germany) showed small primary contributions and minimal gender imbalance; low incidence and reporting imprecision reduced detectable effects in Italy and Slovakia. The Netherlands showed an apparent inversion due to a coinciding lifting of restrictions (“freedom day”) with opposite gender imbalance, confounding the football effect.
• Determinants of impact: The potential for spread N0·R_pre·T strongly correlates with attributed cases (R² = 0.77; 95% CI [0.39, 0.90]; p < 0.001; slope 1.62 [1.0, 2.26]). Analyses excluding England and Scotland yield a non-significant but consistent slope. Trends exist with R_pre and N0 individually (significant for R_pre). Mobility changes during the Euro 2020 did not correlate with attributed cases (R² = 0.06; p = 0.54). Hosting effects were small to non-existent for this multi-country tournament.
• Nonlinear dependence and counterfactuals: Impact depends nonlinearly on initial incidence, pre-event R, and number/duration of matches. Counterfactuals indicate that fewer matches (e.g., England exiting at group stage) or favorable epidemic conditions (e.g., Czech Republic’s lower N0 and R_base) would have markedly reduced attributed cases, even with similar gathering intensity.

The study addresses whether and how a mass sports event measurably increased COVID-19 transmission and under what conditions the impact becomes substantial. By leveraging gender-resolved data and quasi-random match timing, the Bayesian framework isolates match-day gathering effects from broader epidemic trends. Findings show that Euro 2020 gatherings significantly increased transmission in multiple countries, with substantial downstream effects due to subsequent infection chains. The magnitude of impact scales with the baseline epidemic situation (initial incidence N0 and R_pre) and event duration (number of matches), underscoring the importance of timing mass events during favorable epidemiological periods. The lack of correlation with mobility suggests that coarse mobility indices may not capture social mixing relevant to viewing parties. Hosting effects were minor in this distributed-host format, though they may be larger for single-country championships with extensive ancillary activities and travel. Policy implications include promoting vaccination, masking, limiting gathering sizes, testing and COVID-pass requirements for viewing parties, and encouraging post-event self-quarantine/testing. The schedule cadence (matches every 4–5 days) can resonate with incubation/generation intervals, potentially amplifying spread between rounds. The modeling indicates that some overall increases in transmission were captured by R_base rather than being attributed to matches, rendering ΔR_match estimates conservative. Overall, careful preconditions (low N0 and R_pre) and targeted mitigations can substantially reduce the public health impact of future mass events.

This work quantifies the impact of Euro 2020-related gatherings on COVID-19 spread using a gender-resolved Bayesian renewal model, attributing roughly 0.84 million cases across 12 countries. The effect of matches on transmission was significant and highly heterogeneous, driven by pre-event epidemiology and event duration. Primary infections at gatherings seeded sizable subsequent chains, highlighting the lasting influence of transient mass gatherings. The results provide a framework and quantitative guidance for policymakers to balance societal benefits of mass events against health risks and to select mitigation measures. Future research could extend to single-country tournaments with extensive travel, incorporate age-structured dynamics to refine severity impacts, assess variant-specific generation intervals and vaccine-modified transmission, and evaluate real-time decision tools that combine event scheduling with epidemic indicators to minimize risk.

• Confounding events: In the Netherlands, a coincident lifting of restrictions with opposite gender imbalance confounded attribution, leading to exclusion from averages.
• Detection sensitivity: Low incidence and imprecise timing between infection and case confirmation reduce sensitivity to detect match effects (e.g., Italy, Slovakia); wide posteriors reflect this.
• Age distribution: Death estimates assume age-independent effects; primary infections likely underrepresent older high-risk groups, potentially overestimating deaths, though subsequent chains may mitigate age bias via mixing.
• Potential correlation between team progression and incidence is unclear; limited evidence suggests negligible net bias.
• Modeling assumptions: Priors on gender participation, delays, generation intervals, and change-point spacing, though tested in robustness checks, remain assumptions. Weekday-dependent reporting adds complexity and some convergence challenges (non-central variables).
• Data constraints: Reliance on gender-resolved case data and reporting conventions may introduce country-specific biases in delays and case ascertainment. Mobility and policy indices may be too coarse to reflect relevant social mixing patterns.