Expected Goals (ExG) and Shot Data Explained

We talk a lot about shot data and expected goals on an almost daily basis within our picks columns. We get plenty of questions about these as well on the social media channels, so we thought it would be worth a quick column to explain a few of the metrics we use and why they should be important to you when having a bet.

Goals in football are rare. Very rare In fact, and it’s one of the reason’s we love the game so much. There are, on average, just 2.68 goals per game in the Premier League, so when you hear commentators rolling out their clichés about the first goal of any given game being “crucial” they are rather stating the obvious. With low goals, also comes disproportional variance, which means that if we price up football games based purely on league tables or goals scored we are likely to make poor decisions. There really is a thing called “luck” and it has a significant impact on a teams results compared to what they might reasonably have expected an outcome from a game to be. We’ve all been to a game in the past where one side have dominated the whole match, hit the post, the bar and had a clear goal ruled out, only to end up losing 1-0 haven’t we? It just happens.

Relying on league tables and goals is particularly dangerous early in the season, when said “luck” may not have had a chance to even itself out over the course of a season.

So rather than using results or goals to generate our predictions of a football game, we instead try to consider how many goals each team might honestly have expected to score, had those narrow margins not conspired for or against them, and this is where shot data and Expected Goals (ExG) come into the equation.

Shot Data

Let’s start with one of the quickest and easiest pieces of data available to us. www.football-data.co.uk is a particular favourite for mining our Shot Data, with around 20 divisions offered on a weekly basis, all completely free. Their data downloads allow you to see, for every game, the HT score, the FT score, the referee, corners, cards and crucially shots and shots on target for every match. This is one of the main drivers into our model at StattoBets.com.

The chances a team creates are far more prevalent than the goals that end up being scored, and this allows us to eliminate some of the disproportionally high variance in our data. We know that in the Premier League for example, over large data sets, you average team will score one goal for approximately every 9.3 shots they manage. They will typically get one goal for every 3.2 shots on target. Most leagues produce similar sorts of numbers to these, and long term trends show that most teams will converge towards this sort of conversion rate as the season progresses, and is therefore an excellent metric to track when assessing the quality of two sides.

Shot Ratio

Very similar to the above, we don’t always want to simply look at how many shots for and against a team have managed, but rather the ratio which they achieve these. Most sides in most divisions will typically sit somewhere in the range of 40-60% in terms of ratio of shots at goal, and similar in terms of efforts on target. Teams towards the top of this range will invariably challenge for promotion, while those at the bottom will end up in a relegation scrap once all the ups and downs of a season have ironed themselves out. Any side performing outside of these ranges is well worth another look – (such as Man City who are achieving an out-of-this-world 76% shot ratio so far in 2018/19, while Bury and Barnsley look to be very good sides this season.

Expected Goals (ExG)

At it’s most basic, you can assign ExG values to teams based purely on their shot data. Obviously we need to factor in WHO they have played to judge whether this is a fair reflection, but it’s certainly more useful than goals alone.

Unfortunately, not every shot was made evenly. We might all remember that screamer for 35 yards from your midfield general, but the reality is that for every 35 screamer that hits the back of the net, around 20 would end up in row Z or getting blocked. No, instead we should be getting excited when our sides work the ball into the area and get their shots away from around the 6-yard-box, where conversion can be as high as 30%.

Knowing a shot location, and how it was struck allows us to assign an “expected goal” value to every chance a team creates. We might assign 0.05 of a goal for that speculative long punt, a little more for a free-kick from 20-yards, and significantly more for an effort from inside the area – perhaps 0.3 goals

Weaknesses

There are a couple of things you should be aware of when using Expected Goals. Like any model, understanding weaknesses is imperative.

Firstly, game state is very important. When a team leads, they inevitably sit back and protect their leads. This generally means they will appear to have performed more poorly than they perhaps have in reality if they have led a number of matches from early in proceedings.

Another important consideration is team style. Counter-attacking sides almost always convert a greater proportion of their chances, and will often be under-rated by Expected Goals models. That was certainly the case with Leicester City’s title winning campaign, when chances from decent long ranges were often in one-on-one situations, where Expected Goals suggests they should not expect a good chance of scoring. In fact the opposite was true.

Resources

www.football-data.co.uk

www.understat.com

www.infogol.net

www.statsbomb.com

Ben Mayhew Twitter - @experimental361

@MarkOHaire - Twitter Judge

@Caley_graphics ‏

#Football