Exploring Expected Goals with StrataBet Data: Part One – Under Pressure

Expected goals has quickly become the main metric in measuring both team and player performances over recent seasons, but it still isn’t without its flaws. One big flaw that is usually referred to is that it doesn’t take into account the actions of the defensive side.

A shot from the penalty spot is treated the same whether the shooter has 5 players around him or if they’ve broken free of the defence and only have the ‘keeper to beat.

There are proxies to try and avoid this – what the shot was preceded by being one, if the team is on a counter-attack the assumption of less defensive pressure can me made or if the player has completed a take-on you could assume they’ve created more room for the shot.

None of these are concrete though, you assume the state of the opposition defence rather than having it be quantifiable. StrataBet however look to solve this collecting data that measures the number of both defending and attacking players between the ball and the goal as well as having a more subjective measure of pressure on the shooter, ranking it on a scale of 1-5.

Having got my hands on this data the first thing I was interested in was seeing how defensive pressure would impact the xG of shots and how it compare to the xG values I have using just the distance and angle to goal, as well as seeing if it would have an effect on any teams xG performances in last season’s Premier League.

Using StrataBet’s data I was also able to incorporate what preceded the shot (this will probably be part two) so that will also have an impact on the xG values that will follow.

Players In Front Of The Shot (featuring Burnley)

As expected, the number of players between the ball and goal has a huge effect on the expected goal value of a shot. The graph below shows how the probability of a goal decreases the more players are behind the ball:

Click to Enlarge

A shot is roughly half as likely to result in a goal if around 3 defenders are behind the ball than if none are, this means teams who pack the box in defence may concede lots of shots – possibly even lots of shots in what could be considered a dangerous location – but the value of these shots may not be as high as previously thought given how many bodies are between the shooter and the goal.

Take Burnley for example, in my xG model using just distance and angle to goal their xGD puts them 2nd from bottom with only Sunderland faring worse. While with the new xG model it’s not as though they suddenly find themselves on the verge of Europe but they do jump to 16th, their actual finish in 2016/17 – with the actual bottom 3 and Swansea all below them.

While there’s still room for improvement for Burnley this shows they might not be as doomed as people think judging on last years xG numbers and the may not be defying xG numbers in the way previously believed.

In fact, with the StrataBet xG numbers Burnley even had the lowest xG per shot against last season.

The graph below plots the xG per shot against and average number of players behind the ball for Premier League teams last season – it’s pretty clear that having more players behind the ball dramatically reduces the xG value of shots conceded and Burnley are huge outliers when it comes to getting players behind the ball.

Click to Enlarge

An example of a shot against Burnley that took a hit in terms of xG value was Eric Dier’s goal in the match at Turf Moor last season. In the previous xG model it had a value of ~0.3 however this halved to ~0.15 using StrataBet data. While it still resulted in a goal it shows the difference in expectancy between the two models.

If you just take into an account the angle and distance it seems like a great chance, but factor in all the bodies in front of the shooter and while it still may be a good chance the expectancy of it resulting in a goal is half that as before. Take a look at the goal below:

Another interesting case from above is that Liverpool had the 2nd highest xG per shot against last season as well as having the 2nd least players behind the ball, this is presumably because Liverpool conceded high quality chances from opposition counter attacks – Hull’s 2-0 victory over them is an example of this.

It’s not to say Liverpool’s solution is to congest the box but more how can they try to give more protection to the defence to avoid conceding high quality chances from counter attacks. Liverpool’s problem isn’t reducing the number of chances the opposition has – StrataBet data has them conceding the least per game – but the quality of the chances they concede.

Liverpool’s opening game against Watford typified this, Liverpool only conceded 9 shots but 4 were inside their own 6 yard box.

So it seems players between the ball and the goal have a pretty big effect on the expectancy that a shot results in a goal, but what about defensive pressure?

Defensive Pressure

Defensive pressure is defined as how much pressure the shooter is under when taking a shot, for example if a player is trying to block the shot or is in close contact with the shooter possibly grabbing their shirt. The effect of this metric on the probability of a goal can be seen on the graph below:

Click to Enlarge

Interestingly this doesn’t have as big an effect on the expectancy of a goal as the number of players behind the ball does.

While the probability may generally fall the more pressure the shooter is under it doesn’t fall as quickly or as low as it does with the number of players behind the ball. There’s also little correlation with defensive pressure and xG per shot against.

Click to Enlarge

Burnley are again a pretty big outlier but this time at the other end of the spectrum while Tottenham and Arsenal lead the way for putting pressure on the shooter.

While the results of defensive pressure alone may not be as interesting as the number of players behind the ball there are still some interesting points if you combine the two.

  •  Burnley put a huge number of players between the ball and goal but not much pressure on the shooter. This suggests they’d rather congest the space in front of the shooter giving them either less to aim at or making them shoot from a bad position rather than putting them under heavy pressure.
  • Arsenal were one of the worst teams for getting players behind the ball on shooting opportunities but one of the best for defensive pressure. This seems to suggest that opponents tend to break fairly free of the defence but there’s usually a player hassling them when they’re taking the shot.
  • Liverpool and Manchester City don’t perform particularly well in either categories suggesting opponents are not only breaking free of their defence but are also having shots with relatively little pressure on them too.
  • Southampton do a good job of both hassling the shooter and having bodies behind the ball.

Conclusion

As expected what the defensive side does has a big effect on the likelihood that a shot produces a goal, with the number of players behind the ball having a much greater effect than the pressure the shooter is under. This leads to some interesting observations, particularly involving Burnley, Liverpool, Manchester City and Southampton.

It’ll be fascinating to see if these observations continue into the 2017/18 season, will Liverpool and Manchester City give the defence more protection on counter attacks and were Burnley defying xG numbers or is the model taking into account the defence a better metric for assessing them?

I’m sure I’ll come back to these defensive numbers throughout the season both in the Premier League and the other top European leagues, while the other parts of exploring StrataBet’s data will be looking at creativity (both 1st and 2nd assists) and entries into the box.

Finally, here’s the xGD table ranking for the Premier League with the two expected goal models, xG Method 1 only takes into account the distance and angle while xG Method 2 takes into account the distance, angle what the shot was preceded by and defensive metrics.

Click to Enlarge

This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations.

Leave a Reply

Your email address will not be published. Required fields are marked *