I'm building a logistic regression model to predict which team is going to win a soccer match. I'm looking for ideas to broaden the number of predictor variables. I need before as well as after match variables. The variables that came to my head:

*ELO-ratings of the teams
*Total shots & total shot on target
*Market value of the teams
*Free kicks
*Total pass
*Pass percentage

I would greatly appreciate for any help you can provide.

5 Answers 5


A fairly important variable is home advantage. Last season in the Premier League, there were 172 home wins and 115 away wins, which is a significant difference. Though this will vary from league to league, it is still commonplace, and if you do the calculations on other leagues you will come up with similar results. Whilst there is no conclusive proof of this, teams with more compact stadiums often have a greater advantage. I support Gateshead FC and we do not have a particularly large home advantage, which is probably due to the large distance from the pitch to the stand, and the low attendances.


Over the last handful of years a great deal has changed in the analysis of likely outcomes in football. In no small part this has been influenced by approaches developed in American sports, hockey probably more than others.

Probably the best-known of the 'new' stats is Expected Goals (ExpG or xG). At its very simplest, ExpG tells us how likely a player or team is to score from a single shot. Suppose two teams from last weekend both had 5 shots on goal during a game, one team (A) scored 3 goals and the other (B) none. That suggests that shots on target are a poor predictor of scoring probability. However, if we know that team A's shots were all from inside the 6-yard box while B's were all from outside the penalty area, then we have a bit of context.

ExpG puts a quantitative quality score on each shot based on location and many other factors. As this is still a science in its infancy, there are many slight variations in the model that decides on the ExpG of a shot. That notwithstanding, if you know a team's average ExpG and average shot count per game then you can make a prediction of their likely scoring output.

This video by Dan Altman explains better than I.

One of the earliest mentions of ExpG was this blog post by Sam Green on the OptaPro blog.

A slightly later post introducing the concept is this one by Colin Trainor on StatsBomb.

ExpG is becoming more widely accepted with MLS probably leading the way, but UK newspapers and TV channels are starting to mention it.

Another is PDO -

The sum of a teams shooting percentage (goals/shots on target) and its save percentage (saves/shots on target against).

(From James Grayson's blogpost on the subject.)

I don't understand any of that so I won't even try to add anything useful.

A much simpler concept is Per90 -

Normalization for time, usually used in player stats. Instead of dividing a player’s rate stat by games or appearances, you take all the minutes played and divide them into 90 minute chunks. Then take THAT number and use it as you would games played.

(From StatsBomb's Stat Definitions page.)


Goals scored and goals conceded in various forms (eg. think about using all goals scored in the season so far, goals scored against the specific opponent, goals scored in the last x games etc etc).

  • 1
    Do you have any evidence to show that these are significant factors?
    – Philip Kendall
    Oct 19, 2015 at 15:04
  • Well, ELO ratings (at least for the international soccer calculation) do take into account goals in at least one way - I think it's goal differential, but don't remember the exact formula. Also, the post doesn't seem to be concerned with whether there's evidence that they're significant factors. Presumably Ubel will try to figure that out when building the logistic regression.
    – Duncan
    Oct 19, 2015 at 18:34
  • All answers on Stack Exchange should should be evidence-based when possible. Opinions are fine, but the entire answer generally shouldn't be primarily opinion based. Oct 25, 2015 at 1:31
  • @studro: you have misunderstood the original question. At this stage he's brainstorming inputs to his model. He can take my suggestion and do the statistical verification on his own, which might reject or confirm my idea. Oct 26, 2015 at 9:44
  • 1
    If that is the case, this is not a suitable question for this site. However, I think the question can still be answered with evidence, as research has already been undertaken in this area. Oct 26, 2015 at 23:30

I think you have covered most of the important bases. But one thing to remember is current form which is very important in today's game. For example, just a few days ago, chelsea fc, a highly rated team who are the defending champions in the premier league but are going through a bad patch, lost to west ham united fc, a team which finished in 12th position last season, 40 points behind chelsea. So maybe include a way to take into consideration the results in the last 5 games.


I think the defense quality of certain teams plays a big role. For example, Juventus in the season 2015/16 is a team that did not create very many chances in most games compared to teams like roma(seria A), Barcelona(laliga), Arsenal(epl) or Real madrid(laliga) just to name a few. By this I mean, in most cases their ELO ratings ranged between 30-50, whereas teams like arsenal and barcelona it would range mostly between 40-70. However, they were solid in defence, and even now I see them win so many games by single goal diffences especially away from home. To strengthen this fact, the season 2015/16, Juventus had an average conceeding rate on 0.53 whereas they scored 1.97 averagely, matches with odd goals where 21 while even goals were 17, but they only drew in goalless matches twice both away. This shows that their ability to defend was a major key factor in them winning matches. refer these statistics from Punterslounge.com

  • Do you have any quantitative evidence to back up the assertions made in this answer? i.e. can you show that, overall, without cherry picking stats of one team, teams with strong defences do better than their ELOs would predict?
    – Philip Kendall
    May 2, 2017 at 21:01
  • @Steve Thanks for your comment! I think your idea is getting more and more publicity lately, I see your point. For Example, Atlético Madrid or Juventus can compete other teams with higher net worth (Barcelona, Bayern). Can you elaborate this hypothesis with some data or analysis? May 5, 2017 at 21:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.