A Soccer/Football Ratings System - Home Advantage

#1
Hi guys, this is my first post. (So do lower your expectations)

I'm currently writing a thesis that deal with sports probabilities and part of the project will require me to find a indicator for the performance of national soccer/football teams.

However, I need to adjust this indicator to account for home advantage.

The ratings system that I briefly describe below have some drawbacks.
Does anyone know of another soccer/football ratings system for national teams? If not, is the proposed adjustment for the ELO rating below intuitive?



http://www.constantinou.info/downloads/papers/pi-ratings.pdf (Pi-ratings system)

A ratings system known as the pi-ratings system provides both a separate rating for teams playing at home and away. However, its determination appears a bit too complicated for my liking. For eg. there is the existence of a learning rate i.e. essentially the weight placed on newly acquired information. Although, it could be set up. On the otherhand, it appears to be quite comprehensive.

I much prefer the ELO ratings system with the algorithm provided on eloratings.net. Various journal articles have reaffirmed its ability to forecast matches. I proposed to my supervisor that I could adjust the ELO ratings to account for home advatange by adding K*We = K / (10^([-dr+100]/400) + 1), where dr is the average difference across teams participating in any tournament and 100 adjusts for the home advantage that was initially subtracted in the algorithm. However this appears unreliable and has limited intuition (as my supervisor mentioned.)

http://eloratings.net/system.html
 
Last edited:

Englund

TS Contributor
#2
Have a look at Stefani (1997). The model he describe takes home advantage into account. It is tricky to set the data up for the modelling process though. Just out of curiosity, do you in any way take the bookmakers odds into account? They, who does this for a living, set the best odds for sport outcomes. You can find implicit probabilities from their information, if that is of interest.

[math]P(\text{Home team wins}) = P(H) = \frac{1/Odds_H}{1/O_H+1/O_D+1/O_A}[/math]

Since bookmakers are the best in this field, i'd definitely make use of their information. If you suspect their odds to be biased (they often set up biased odds on purpose to minimize their risk), you can always run a multinomial logistic regression with outcome as dependent variable and the bookmakers' odds as independent variable.

Stefani, R. (1997). Predicting the outcome of soccer matches: Variations. Proceedings of the Section on Statistics in Sports, August 10-14, 31-37.
 
#3
Thankyou Englund for the well researched and insightful response.

Unfortunately, myself and my supervisor have settled with conducting a (non-parametric) kernel regression using a ratings system (undecided) and match data from a hybrid knockout-round robin tournament. As inflexible as that may sound, I will be utilsing it in my paper and thus need a ratings system that adjusts for home advantage.

The proposed method is based on the following paper.

Geenens G, 2013, 'On the decisiveness of a game in a tournament', European Journal of Operational Research, http://dx.doi.org/10.1016/j.ejor.2013.06.025

I also emailed the innovator of the pi-rating system to see whether he could assist in the matter.

My supervisor advised in our last meeting that I form my own ratings system that combines elements of the pi-rating system and ELO rating system.