Using Goal Expectation Data to Predict the Outcome of a Game

I built a predictive model a few years ago with the help of another chap in terms of goal expectation, and I wanted to show you a few of the games to see how the model performed. Spurs v Newcastle Spurs 0.29*6.54, which means that my model predicted that Spurs would have 6.54 shots on target and a goal to shot on target ratio of 0.29 to give total goal expectation of 1.90. Newcastle 0.30*3.92= 1.18 goals and I advised this on twitter giving goal expectation of 1.90 + 1.18 for the game so 3 goals. If we move to the game then Spurs had 5 shots on target so 0.4 goals to shot on target ratio and scored 2 goals in line with expectation. Newcastle had 3 shots on target so a goal to shot on target ratio of 0.33 and goal production was 1 goal, which was also in line with goal expectation. Defoe did not play but we can see that this had no effect on the ability of Spurs to reach their goal expectation although their shot on target production was down 24% on expectation. Newcastle expectation was almost spot on. The key is did anything happen in the game to impede goal expectation or accelerate goal expectation and the answer is no. There was not an early away goal or a red card or any other variable to define a change in the expected outcome. If we look at the game at HT we see that before today we can see that 52% of games that are 1-1 HT in the EPL have ended a DRAW FT. When Spurs scored to make it 2-1 I was aware that expectation of a further goal(s) to Spurs was very unlikely and this is simply because just 4% of 1-1 HT have ended 3-1 FT this season and BET365 were aware of this because at HT they advised on TV, back Spurs to win 3-1 and I think they offered 17/2 when the price to gain any advantage needed to be much higher then that. I can offer a further edge which is I profile games after HT to see what the likely FT shot on target count to each team is likely to be, and model the full time expected score around that information. The average home shot on target figure is 5.33 when 1-1 HT and FT 2-1 so we see that figure is in line with SPURS who had 5 SOT. The away average is also 5.33 so higher then Newcastle who managed 3 SOT but we can mirror the WEST BROM v Norwich game which was 1-1 HT and 2-1 FT and 6SOT v 3SOT and goal on 82 minutes against a goal today on 78 minutes and nothing gives me more pleasure then finding a game that mirrors another game. Chelsea V Wigan Chelsea 0.37*5.21 = 1.92 goalsWigan 0.27*3.58 = 0.97 goals Total goal expectation of 2.9 goals. Chelsea had 12 shots on target so 0.33*12= 4 goalsWigan had 6 shots on target so 0.17*6 = 1 goal We can see that in terms of goal expectation that Chelsea out performed and Wigan goal to shot on target was lower then expectation but this was countered by their higher then expected shot on target production. The key is why did Chelsea score four goals against the expectation of2 goals and the obvious point is that Chelsea had 12 shots on target against an expectation of 5.21 but does 1-0 HT and FT 4-1 model the shot on target production of both teams? I am unable to answer that question as have no data to compare as before today there were 0% 1-0 and 4-1 FT games this season but if we look at FT 2-1 when 1-0 HT in the EPL this season we see that the home full time SOT is in the range of 5-8 so Chelsea at 12 did not fit the 2-1 Model as time decayed. 2 late goals and was there expectation of the late goals i.e. did the second Chelsea goal accelerate goal expectation? The game does not even fit 3-1 FT so at 2-1 should you have pulled the lay 2-1 trigger and I see no evidence to suggest that you should have. I would have said the same thing at 3-1. We see that Chelsea did have a number of late shots on goal but there was certainly not expectation of this and 2 goals in the 80-FT time band to one team is a very rare event. Sunderland v Arsenal Sunderland 0.26*4.13 = 1,07 goalsArsenal 0.25*4.63= 1.15 goals Total 2,2 goals that was in line with the bookmakers expectation of a low scoring game and I would imagine many people had 0-1 and 0-2 as a popular FT score although my model was keen on 1-1. Sunderland 0.00*6= 0 goalsArsenal 0.125*8= 1 goal We see that both teams out performed in terms of shot on target production but both their goal to shots on target ratios average were very poor. I have not had time to watch this game but at HT 0-1 we see that 0-2 has only happened once this season so 0-2 is an unlikely FT as a second goal to Arsenal should trigger further goal(s) or a goal to Sunderland would take 0-2 out of the equation. In terms of the game ending 0-1 Ft the SOT production (14) does not look anything like a 0-1 HT 0-1 FT so during the second half there was expectation of a goal(s). The game in terms of SOT does not fit a 0-1 >1-1 FT. The closest fit is 1-2 FT. Average Home SOT is 5 and away5.25 SOT I am going to have to watch the Arsenal game later to see what was going on in terms of there being goal Isofair online expectation in the second half but no goal(s) I was watching Chelsea today and then I moved on to the Norwich game, which was a poor game.