Predicting the outcome of the 2018 FIFA World Cup using Elo ratings

Introduction

The 2018 FIFA World Cup starts in a week’s time. The first game is between Russia and Saudi Arabia in the Luzhniki Stadium in Moscow.

In the most recent World Football Elo Ratings (8 June 2018) Russia is rated 44 and Saudi Arabia 63. (The  FIFA Rankings place Russia 70 and Saudi Arabia 67.)

The Elo rating system for international football is calculated from the results of international A matches. The rating takes into account the kind of match, home team advantage, and goal difference in the match result.

I used Elo ratings in my analysis of the 2014 FIFA World Cup tournament.

Gracenote‘s predictions for 2018 and a paper written by Lorenz Gilch and Sebastian Müller use Elo ratings to propose outcomes for the tournament.

Gracenote

Gracenote uses the Elo ratings “to estimate the percentage chance of each match at the 2018 World Cup being a win, draw or loss for each of the teams”. These percentages are fed into a simulation of the World Cup “to produce estimates of the chance that each team reaches a particular stage of the competition”.

Gracenote will be run their simulations after each day’s matches to update their forecasts during the tournament.

Before the tournament:

  • Brazil has 21% chance of winning the tournament.
  • Spain, Germany and Argentinawill be Brazil’s main challengers.
  • 47% chance that the World Cup winners will be from a country other than Argentina, Brazil, France, Germany or Spain.
  • European and Latin American teams should dominate the knockout phase.

Lorenz Gilch and Sebastian Müller

Lorenz and Sebastian’s paper appeared on arxiv.org this week (7 June). It is titled ‘On Elo based prediction models for the FIFA World Cup 2018‘.

Their paper favours Germany as the tournament winners.

The abstract of their paper is:

We propose an approach for the analysis and prediction of a football championship. It is based on Poisson regression models that include the Elo points of the teams as covariates and incorporates differences of team-specific effects. These models for the prediction of the FIFA World Cup 2018 are fitted on all football games on neutral ground of the participating teams since 2010. Based on the model estimates for single matches Monte-Carlo simulations are used to estimate probabilities for reaching the different stages in the FIFA World Cup 2018 for all teams. We propose two score functions for ordinal random variables that serve together with the rank probability score for the validation of our models with the results of the FIFA World Cups 2010 and 2014.
Lorenz and Sebastian use a single Sankey diagram (Figure 4 in their paper) to visualise the probabilities of their models.

Conclusion

It is fascinating to observe the approaches used to predict tournament outcomes. The two examples shared here illustrate the range of expertise brought to sport analytics.

Gracenote started its football rating system in 2002. I will be following Simon Gleave’s tweets during the tournament to see how a dynamic prediction system works:

Lorenz’s Twitter account is sharing links to datatreker.com. I will be following Lorenz’s tweets too.

Photo Credit

New construction in St Petersburg (Ninara, CC BY 2.0)

Postscript

After writing this post I found Kiko Llaneras and Borja Andrino’s article in EL PAIS (5 June 2018) ‘¿Quién ganará el mundial?‘. Their analysis has “La base de nuestro ranking es un método Elo, inspirado en el que se emplea en ajedrez y otros deportes”. There are three weighted Elo components:

  • Ranking Elo clásico (peso 50%)
  • Ranking Elo esperado (peso 30%)
  • Ranking Elo de jugadores (peso 20%)

The Impact of Managerial Change: EPL 2017-2018

I noticed this tweet earlier today

By coincidence, Ron Smith had sent me a link to the paper too. It is available online at this link.

The paper encouraged me to think about the data I have been collecting from the EPL.

I have been tracking the impact of the eight managerial changes in the EPL this season with a very basic ‘momentum’ metric. This indicates that, after week 28 of the season, Leicester has benefited most from a change (+3) whilst West Brom has struggled (-7).

My maps of routes to safety or relegation, in order of managerial changes, are:

Crystal Palace

Leicester

Everton

West Ham

West Brom

Swansea

Stoke

Watford

Simon Gleave added a fascinating dimension to this conversation in his consideration of how many points are required to remain in the EPL.

He shared this graphic (14 February 2018) of the % chance of relegation this season:

Photo Credit

BFCvFCUM_ENPLPD_041114 (Matthew Wilkinson, CC BY-ND 2.0)

 

On the ball … in 1935

Simon Gleave and Jurryt van der Vooren have been tracking down the earliest example of football statistics.

There have been some Twitter exchanges

In response to:

This encouraged me to write a blog post about the game.

Today Jurryt came up with two new leads, one from a Holland v Belgium game in 1935:

and this from De gronwet on 15 January 1936

This second source refers to some French journalists at the Jour newspaper. My brief enquiries suggest this might be a newspaper published in 1933.

I do need to follow up on these leads but I am immensely grateful that Simon and Jurryt are sharing their treasure hunt.