Although I've been told I hide it well, I am latino in my heart of hearts. I love the beisbol, I am passionate about boxing, but once you throw in my italian heritage to my cuban and puerto rican background, my passion for futbol was completely inevitable.
Given that passion, my unmitigated glee at what I foresee as a fantastic upcoming world cup, and my love for the math behind the sports, it's only natural that I would want to crunch the world cup. But Futbol is not an easy game to figure out, and a lot of the studies and literature that exist for the predominant sports in North America don't exist for it.
So my passion will only get me so far. Given the lack of data out there, I needed a lot of creativity to actually get something done. First, I needed to figure out how to compare the teams most efficiently. Then I needed to figure out how that comparison translates to goals, and I also needed to work out how to deal with ties. Finally, I needrf to simulate everything.
It's like christmas came early this year.
The end result looks something like this:
It's Brazil(Auriverde), then Spain(La Roja), Germany (La Aplanadora (for hispanics, translates as
"the demolisher"), Argentina (La Albiceleste), Holland (Oranje), Portugal (Selecção das Quinas), England (The Three Lions), Colombia (Los Cafeteros), Uruguay (La Celeste), France(Les Bleus), Italy (Azzurri) as our top 11 teams. All those teams have a greater than 1% chance of winning per the sim, and one of them wins in 95% of all the simulations.
I got goosebumps typing all the nicknames.
Let's talk about the journey for a bit, shall we? We'll start with the data and the rankings. I am a big fan of Elo based ratings systems for international and club futbol (You can go read that link for full detail, it's fascinating). The two main pages I use are the Football Club Elo page and the World Football Elo page. For our purposes, the World Football ELO ratings based on data from Advanced Satellite Consulting is best. However a lot of the math was inspired by the math model discussed on the Football Club ELO page as well as the Wikipedia page for World cup Elo. This answers the question of how to rate teams and how to work out win probabilities. As a baseline assumption, I treated every team as playing on a neutral site, except Brazil, who gets the typical host homecourt advantage (If you want to see a different approach, I recommend the 538 piece that did this differently).
If you start reading all that literature you will notice a glaring problem: all the math is based on projecting a match as a win or a loss. Futbol features a third outcome, the tie, at least in group play (EDITOR'S NOTE: especially because a 50% win probability does not equate to a tie, given the 3pt/1pt/0pts scoring structure). I had to figure out a way to model it.
A good starting point is to tabulate and figure out that about 23% of international matches end in ties. A better piece of data is this:
It runs higher in the World Cup, which tracks with the fact that the level of competition is generally higher. A key point is that it's only in the group stage that we have ties, in the knockout stage we must have a winner and we also have extra time driving down the number. Armed with this piece of information and some handy projection models, I started building my sim and tuning it to match the data. I ended up with a similar distribution.
You can see the full fixture sim here:
I've kindly worked out the odds of the different goal differential scenarios. Keep in mind that this assumes no biases in all the matches but there will likely be some real biases in the final group matches, as teams may be playing for ties to guarantee advancement.
So that's the full tournament sim, the math behind it, some cool links to look at, and the simulation in detail for every fixture. Are we missing anything?
That's a breakdown of the relative strength and odds of every group in the competition. Group B is your group of death with Group G getting it's average driven down slightly by Ghana.
Enjoy the cup!
-Arturo