Projecting the World Cup Group Stage

This article originally appeared in The Shin Guardian

Predicting the outcome of soccer matches, and World Cup matches in particular, with any confidence is an exercise for the foolhardy.  One fateful bounce, wonder strike, mistake, penalty, or offside call can change a nation’s entire trajectory.  But, what fun would this event be if we could not dissect and over-dissect all the matchups and possible outcomes?  We have rounded up some of the best regarded international soccer rating systems and played out W-L-D probabilities for every match of the group stage.  Let us meet our contenders:

Elo: Originally devised as a method to rank world chess players, it is one of the most robust international soccer rating systems.

SPI: Developed for ESPN by famed political prognosticator Nate Silver.  Get used to seeing their ratings thrown around a lot during ESPN’s World Cup coverage.

Oddsportal: An aggregator of 10+ online betting house odds.  Reflects the opinion of the betting public.

And…

One way to decide it…

EA Sports FIFA Video Game: Ok, so using video game player ratings is not a statistically rigorous method, but this still seems a step up from Paul the Octopus.

—

The Predictions

(Note: all figures represent approximate expected points)

Group A

Group B

Group C

Group D

Group E

Group F

Group G

Group H

Ranking Biases

Each one of the four rankings (EA Sports, Elo, SPI, Oddsportal) have relative biases.

In many instances, these biases follow along geographic lines (see table below).

For example, many of the online betting houses are based in Europe, so there is a noticeable bias against lesser known international sides from North/Central America and Asia.

Similarly, EA Sports player ratings are noticeably biased towards players and teams that feature in the major European leagues.  In SPI’s case there is a favorable bias to South American sides which is likely due to the heavy weight they placed on CONMEBOL World Cup Qualifying matches (as far as I am aware the raw SPI ratings I used do not take into account continental bias for the event taking place in Brazil, which might be a plausible explanation).