lbenz730 / world_cup_2022

Modeling and simulations for the 2022 FIFA World Cup

FIFA World Cup 2022

---

Modeling and simulation of the 2022 FIFA World Cup

Model Script

• fit_model.R: Estimates Bayesian bivariate Poisson model. For more details, see Equation (2) of Benz and Lopez, 2021.

Simulations

• run_sim.R: Run 10,000 simulations of the tournament
• helpers.R: Helper functions for sims.
• game_preds.R Save out predictions for indivdual games.

Data

Data (internation_soccer_score.csv) courtesy Kaggle.

Flag figures courtesy of Flaticon.

Pre-Tournament Predictions

Previous Tournaments

• Euro Cup 2021
• World Cup 2018
• Women's World Cup 2019

world_cup_2022's Issues

unaccounted confounders

Good work. But I've seen your previous predictions for Euro 2021 and also seeing the current progress in Qatar 11/2022 I believe that your model is incomplete and there are other confounding factors that you should be accounting for. What those are, I cannot really tell you, it needs exploration.

would it be faster and more stable if using poisson_log_lpmf( ) ?

Great work! Thanks for sharing.

I have a small question, if using poisson_log_lpmf( ) in model, will it be faster and more stable?

 for (g in 1:num_games) {
    lambda1[g] = mu + home_field * ind_home[g] + neutral_field * (1 - ind_home[g]) + alpha[home_team_code[g]] + delta[away_team_code[g]];
    lambda2[g] = mu + neutral_field * (1 - ind_home[g])  + alpha[away_team_code[g]] + delta[home_team_code[g]];
  }

  for (g in 1:num_games) {
   target += poisson_log_lpmf(h_goals[g] | lambda1[g]) * weights[g];
   target += poisson_log_lpmf(a_goals[g] | lambda2[g]) * weights[g];
  }