Statistical model of football scoring

[UKBoy1977]

Sorry, I'm a Brit so that's soccer to most of you!

I'm interested in deriving a prediction equation for the number of goals scored in a match based on the two teams' average scoring and conceding rates.

Using the match results of the whole of last year's Premiership season I am going to find the average goals conceded and scored per match by each team simply by dividing total goals scored/conceded in the season by the number of matches.

Then I want to try and find a correlation which would tell me, say, if Team 1 (scoring rate S1/Conceding rate C1) plays Team 2 (Scoring rate S2/Conceding rate C2) what is the expected number of goals scored by each team?
Thus, if one were plotting it, you would obtain a 3-D plot with scoring rate and conceding rate on the x- and y- axes and number of goals scored on the z-axis.

But from here I'm stuck. How does one derive the best (maximum likelihood?) prediction equation? Will it even be linear? What if I plot the points and it's not a straight line? I can't believe I did Statistics at Uni and it's (nearly!) all gone.

 

[Robin]

Sorry, I'm a Brit so that's soccer to most of you!

I'm interested in deriving a prediction equation for the number of goals scored in a match based on the two teams' average scoring and conceding rates.

Using the match results of the whole of last year's Premiership season I am going to find the average goals conceded and scored per match by each team simply by dividing total goals scored/conceded in the season by the number of matches.

Then I want to try and find a correlation which would tell me, say, if Team 1 (scoring rate S1/Conceding rate C1) plays Team 2 (Scoring rate S2/Conceding rate C2) what is the expected number of goals scored by each team?
Thus, if one were plotting it, you would obtain a 3-D plot with scoring rate and conceding rate on the x- and y- axes and number of goals scored on the z-axis.

But from here I'm stuck. How does one derive the best (maximum likelihood?) prediction equation? Will it even be linear? What if I plot the points and it's not a straight line? I can't believe I did Statistics at Uni and it's (nearly!) all gone.

We did a whole thread on this recently (called "Help with Excel" - it was actually more about stats than Excel).

I would warn you that if this is for betting then you probably can't do this better than the betting agencies do it themselves.

Essentially you have to bet on the long shots where you calculate the odds as better than those offered. So it means that even if you have got it right, you still lose most of the time.

I have a model that generates about the same odds as the betting agencies do, I will find it if you are interested.

 

[UKBoy1977]

That would be great thanks. Do you think all British bookmakers use computer models for all divisions? You think there is literally no edge to be found here?

 

[Jaggy Bunnet]

That would be great thanks. Do you think all British bookmakers use computer models for all divisions? You think there is literally no edge to be found here?

If they use a model for the premiership, how hard do you think it would be to adapt that model for other leagues?

If you think it is worth your while to develop a model to enable you to win, why do you think it would not be worth it for bookies who take part in far more bets than you do, to do the same?