COVID and Home Advantage in Football: An Analysis of Results and xG Data in European Leagues

Introduction

This study investigates whether the COVID-19 pandemic had an impact on the home advantage in football. The home advantage is a well-known phenomenon, and understanding its potential alteration during a period without crowds offers unique insights. By analyzing expected goals (xG) and points data across major European leagues, this study sheds light on the broader effects of the pandemic on competitive sports.

This study will try to reproduce and replicate this article by Mathieu Acher, https://blog.mathieuacher.com/FootballAnalysis-xG-COVIDHome/, written on May 23, 2021.

Reproducibility

How to Reproduce the Results

1. Requirements

   • System Requirements: Any operating system that supports Docker.
   • Dependencies:
   • Python: 3.8
   • Libraries:
     • pandas: 1.5.2
     • scipy: 1.6.0
     • matplotlib: 3.7.5
     • Jinja2: 3.1.4
     • jupyterlab: 3.5.0
     • ipykernel: 6.21.3

2. Setting Up the Environment
   Use the provided Dockerfile to create a reproducible environment:

   docker build --build-arg MY_PYTHON_VERSION=3.8 --build-arg WORKFLOW=default -t covid-home-advantage .
   docker run -it -p 8888:8888 covid-home-advantage

   Open the jupyter-notebook : covid_home_advantage_analysis.ipynb

3. Reproducing Results
   Access JupyterLab If you just runned the Dockerfile :

   1. Open your browser and go to http://localhost:8888 and use the token displayed in the terminal to log in.
   2. Open analysis/covid-home-advantage_analysis.ipynb in your notebook environment with the kernel named Python 3.8.x (Docker REP project)
   3. Follow the steps to reproduce the analysis. Else :
   4. Run this command first and redo the steps above :

     jupyter lab --ip=0.0.0.0 --port=8888 --no-browser --allow-root

Encountered Issues and Improvements

• Challenges:
  • Data scraping involved manual extraction of JSON patterns from the website as no API was available.
  • No clear documentation on the data preprocessing steps was provided.
  • No clear instructions on what versions of libraries were used in the original study.
• Improvements:
  • Introduced a structured data pipeline for efficient preprocessing.
  • Documented the data scraping process for future reference.

Is the Original Study Reproducible?

The original study's methodology was reproducible with the given data and tools. Our results closely matched those presented in the study, as evidenced by comparison tables and visualizations.

Results and Metrics

1. Difference Between Home and Away Points

   • Displayed in tables and bar charts across six leagues from 2014–2020.
   • Example metric: DIFF_POINTS_HOMEAWAY, DIFF_XPOINTS_HOMEAWAY.

2. Evolution of Mean Points (Per Match)

   • Plotted the seasonal trends for home and away matches.

3. Statistical Tests

   • Wilcoxon Signed-Rank Test: Assessed differences between home and away metrics.
   • Mann-Whitney U Test: Compared seasonal variations in home metrics.

Analyze and Document Reproduced Results

• Results align with the original study's conclusions. COVID-19 had a measurable impact on home advantage across analyzed leagues.
• Supporting visualizations and statistical summaries validate findings.

Document Any Deviations or Observations

• I noticed a major deviation in the Non-parametrical Wilcoxon Signed-Rank test results. The original study reported a constant p-value for xG of 0.000515 for all leagues. However, my results showed variations across leagues. It might be an error in the original study. Apart from this, most results were consistent.

Variability Factors

• Source of Data: Understat's league data from 2014–2020.
• Version of Data: Potential inclusion of a larger dataset for extended analysis by using an other source.
• Statistical Tests: Alternative tests for robustness.
• Environment: Use of Python 3.9 for enhanced performance and compatibility with newer libraries.

Draw Conclusions: Is it Reproducible?

Yes, the study is reproducible with the provided tools and data pipeline. The results consistently support the hypothesis that COVID-19 affected home advantage in football.

Replicability

Variability Factors

• Version of Data: Potential inclusion of a larger dataset for extended analysis.
• Statistical Tests: Alternative tests for robustness.
• Parameters: Different parameters for statistical tests.
• Environment: Use of Python 3.9 for enhanced performance and compatibility with newer libraries.

Variability Factors

• List of Factors: Identify all potential sources of variability (e.g., dataset splits, random seeds, hardware).
  Example table:

  Variability Factor	Possible Values	Relevance
  Version of Data	2014->2024	Impacts consistency of random processes
  Statistical Tests	Test t, permutation_test	May affect computation time and results
  Parameters	Significance Level (α), Alternative Hypothesis (alternative)	Ensures comparability across experiments
  Version of Python	3.9	Compatibility and performance considerations

• Constraints Across Factors:
  To ensure consistent and reliable results, the following constraints were applied to variability factors:

  • Data Consistency: All data used was sourced from the same database (e.g., Understat) to maintain uniformity across analyses.
  • Statistical Test Parameters: Parameters for statistical tests (e.g., significance levels) were standardized to avoid discrepancies.
  • Environment Configuration: The replication was performed in a controlled environment using Docker to mitigate hardware and software-induced variability.
  • Python Version: Python 3.9 was chosen for the replication to align with the updated libraries and ensure compatibility.

Replication Execution

1. Instructions
   Use the provided Dockerfile to create a replication environment (you can rename the image to keep both environments available on your machine):

   docker build --build-arg MY_PYTHON_VERSION=3.9 --build-arg WORKFLOW=replicate -t covid-home-advantage .
   docker run -it -p 8888:8888 covid-home-advantage

   Open the jupyter-notebook : covid_home_advantage_analysis_replicability.ipynb

2. Presentation and Analysis of Results

   • The analysis of replicated results, as shown in the notebook, highlights a notable trend: during the COVID-19 season, teams gained more points away than in previous seasons across most championships.
   • The seasonal comparison of home and away statistics reveals significant differences during the 2020 season, contrasting with other seasons. Additionally, the disparity between xPoints (expected points) and actual points at home was more pronounced during the COVID season for most championships.
   • These results strongly suggest that the absence of crowds during the pandemic negatively affected home advantage in football across multiple leagues.

Does It Confirm the Original Study?

• The replication successfully confirms the findings of the original study. Additional statistical tests and extended datasets further substantiate the conclusion that the COVID-19 pandemic reduced home advantage in football.

Conclusion

The study is reproducible using the provided data and tools. It is also replicable with extended datasets and alternative statistical tests, consistently arriving at the same conclusion that the COVID-19 pandemic negatively impacted home advantage in football.

---

Contributors

• Aidan Gallagher, Student at INSA Rennes in 4th year of computer sciences
• Noé Bourmalo, Student at INSA Rennes in 4th year of computer sciences