How important is the time of the first goal? | Stats Analysis

How important is the time of the first goal? | Stats Analysis


I thought it would be interesting to review the games which took place on 29th and 30th of December, focussing on the time of the first goal. In six of these games the first goal was scored by the away side within the first 20 minutes; of the other four games, three of them saw a home goal within the first twenty. The odd one out was Sunderland v Spurs which ended 1-2, with the first goal scored By O’Shea for Sunderland on 40′.

The question is, does the time of the first goal change the expectation of further goals? In this research paper by Dixon and Robinson, published in the Statistician, they looked at data from over 4000 games from the main English competitions. They found,

[quote]clear evidence that the rate of scoring goals changes over the course of a match. This rate tends to increase over the game but is also influenced by the current score. We develop a model for a soccer match that incorporates parameters for both the attacking and the defensive strength of a team, home advantage, the current score and the time left to play.[/quote]

As shown in the graph below, Dixon and Robinson concluded that  there was a continuously increasing rate of goals for both the home team and the away team, and they thought that maybe this was caused by an increase in defensive mistakes due to fatigue.

They also found no evidence to support the idea of immediate ‘fightback’, that is, the home team equalising soon after conceding.

Dixon is also known for looking at the effect of an early away goal, which has been recently been re-examined by Nevo and Ritov (2012). Nevo and Ritov examined the effect of the first goal on the second goal using survival analysis methods.

Survival, in terms of football, is the ability or inability of a team to prevent fightback.

Nevo and Ritov concluded that

[quote]a first goal occurence could either expedite or impede the next goal scoring, depending on the time it was scored. Moreover, once a goal is scored, another goal becomes more and more likely whether the goal was scored or conceded.[/quote]

Nevo and Ritov concluded that there were limitations to survival analysis, such as goalless matches, which cannot be incorporated into their research. In addition, their model could not account for the “fast goals”, as it was inaccurate for the first 15 minutes of a game because these events may be more random than others.

Dixon argued that expectation of a goal is dependent on the current score and when an early away goal is scored, expectation of further goals is increased (more then the original expectation of goals before the game) with a bias to the home team having their goal expectation accelerated.

We can certainly see if this is true or not by adding all the goals to an excel sheet for the season thus far to see if the total goals per game are greater then 2.86, which is the average per game this season. We can also identify the home goals as well to see if they have accelerated or not.

Looking at the table below we do indeed see that when the away team score an early goal which in this sample is the first goal of the game (and I have not added the games where the away team score the second early goal for ease of reference) that further goals are accelerated. The average total number of goals in the sample of 44 games this season is 4.11, so much higher then the current EPL average of 2.86, confirming that early away goals increase goal expectation.

I want to take this further and try to apply this to betting, so lets have a look at what percentage of games below show both teams scoring (BTTS), as we know that the home team should score at least one goal or more after conceding an early away goal.

At the moment the percentage of BTTS in the sample is 88%, a very high percentage.

I agree that expectation of a goal is dependent on the current score and that from a betting perspective a good time to enter the market is after a goal is scored. Say there is an early away goal, for example. If you have backed the home team then your bet is in danger of losing but if you have not had a bet and now back over 3.5 goals then you are in a much better position than the person who had a bet before the game.

I could not  name more than 3 players in the Reading team but I do know their:

  1. Goal to shot on target ratio
  2. Their goal to final third pass completion ratio
  3. Their one shot per possession %
  4. The average number of shots they have and concede per game
  5. Their expected goal expectation, shot on target and goal to shot on target expectation

When Shawcross of Stoke was suspended for Stoke and did not play against Southampton I was fully aware that this was a blow to Stoke’s chances of “survival” against the Saints and people agreed, as Southampton were heavily backed before the game and again the backers lost because of a Stoke “fightback”. The question is whether you can link Shawcross not playing to the early away goal.

I personally see no evidence to rely on the strength of a team during a game. Imagine that the ‘power’ players in the team have a bad game, as can happen because players are not machines. We are not privy to information such as personal problems which would effect a player’s ability to perform at his highest level. When, for example, Sunderland won 3-0 at Chelsea, was it because Chelsea underperformed or because Sunderland improved for that game?

What we do see is that there is an average of just under 0.3 in terms of the goal-to-shot-on-target average for the EPL and that when a team such as Sunderland and WBA overachieve in the early part of the season, historical evidence suggests that they will revert to the mean, having some bad games to lower their ratio. This happened to Sunderland already and WBA look likely to lose a few games in the next few weeks.

A team like Liverpool are very interesting in terms of predictive modelling. There is a general opinion that there is a correlation between goals and final third passing. The problem is that Liverpool have had 383 shots, 107 shots on target, 2495 successful final third passes, and only 31 goals.

Liverpool only score 1 goal per 80 final third pass completion (FTPC), when we should expect a goal around every 55-65 FTPC. In the early part of the season Liverpool’s FTPC stood at around 94 so progress is being made. If Liverpool had a goal-to-FTPC of 60 then this would mean that they would have scored around 41 goals. In effect, then, they should have scored 10 more goals this season which, if we distribute the goals evenly over games, is another 0.50 goals per game. This would have resulted in another 14 points and Liverpool would be in joint second with Man City on 42 points.

My overall conclusion is to read the research and try and keep an open mind. I fully appreciate that the idea that football is a science mixed in with some random events is counter-intuitive.

[table id=140 /]

Reference: Dixon & Robinson (1998), Birth Process Model Football