This repository contains the data and code used in a manuscript Estimating the change in soccer's home advantage during the Covid-19 pandemic using bivariate Poisson regression by Luke Benz and Michael Lopez. A pre-print of our manuscript is available here.

• league_info.csv: csv of general information on 17 European Leagues used in this analysis. Most important are the restart_date (the date the league returned to play following the pause of its season to the Covid-19 pandemic) and fbref_league_id, the league's unique id on Football Reference.
• fbref_scraper.R: R script for scraping game level statistics for each league from Football Reference
• fbref_data/: Folder containing all data used for this project. Each of the 17 leagues has its own folder, containing 5 csv files of game level statistics for games played that year.

• models/cards/ R scripts for fitting yellow card models (Model (4) in paper)
• stan/cards/ Stan files for fitting yellow card models (Model (4) in paper)
• models/goals/ R scripts for fitting goal card models (Model (3) in paper)
• stan/cards/ Stan files for fitting goal card models (Model (3) in paper)
• models/empirical_baselines.R: Saves models/empirical_baselines.csv for empirical Bayes priors in Models (3) and (4) with no-zero correlation.

.rds files of posterior draws are available for goals and yellow card models for each league.

• bvp_goals_no_corr/: Folder of posterior draw .rds objects for Model (3) in paper
• bvp_goals_lambda_3/: Folder of posterior draw .rds objects for Model (4) in paper

.rds files of model objects are available for goals and yellow card models for each league.

• bvp_goals_no_corr/: Folder of .rds model objects for Model (3) in paper
• bvp_goals_lambda_3/: Folder of .rds model objects for Model (4) in paper

• simulations/simulation.R R script for running simulations described in Section (4).
• simulations/biv_pois.stan STAN file for running simulations described in Section (4).
• simulations/paired_comp.stan STAN file for running simulations described in Section (4).
• simulations/sim_files/: Folder of saved simulation results (specifically, see v2_sims/ for the most up to date versions in version 2 of the paper).

• paper_figures/: R scripts to produce all figures and tables we present in the manuscript.
• eda/: Folder of old or draft versions of analysis.
• helpers.R: R script of useful helper functions

In order to replicate the entire model fitting process simply run:

• models/goals/bvp_goals_no_corr.R Model (3) presented in our manuscript.
• models/cards/bvp_yc_lambda3.R Model (4) presented in our manuscript.

Such a script takes the following steps:

1. Creates model-specific directories for saving both model objects and posterior .rds files.
2. Reads in data for a specific league from fbref_data/ folder, via read_league_csvs() helper function in helpers.R.
3. Filters data to relevant games.
4. Prepares data for use by Stan.
5. Sources corresponding Stan file and fits model.
6. Saves league specific model .rds object into model-specific folder.
7. Saves posterior draws .rds obeject into model-specific folder.
8. Repeat 2-7 for each of 17 leagues.

Model (4), our yellow card model which assumes correlation > 0, relies on empirical baselines as priors from a version of the model fit with no correlation.

In order to fully reproduce the results in our paper, one would run the following scripts in order:

1. models/goals/bvp_goals_no_corr.R: Model (3) presented in our manuscript.
2. models/cards/bvp_yc_no_corr.R: For yellow card priors when fitting Model (4).
3. models/empirical_baselines.R: Extracting posterior means of no-correlataion versions of models for use in priors when fitting verions of models with correlation.
4. models/cards/bvp_yc_lambda3.R Model (4) presented in our manuscript.
5. models/goals/bvp_goals_lambda3.R: Verion of Model (3) w/ corrletion; not presented in manuscript.

Note the rstan package is required to work with model objects and/or run the modeling scripts. For assistance installing Stan, please refer to the official documentation.