Syzygy Analytics

This week I’m just looking at some more data on games states and xG from understat.com.

10 League Analysis

I want to further investigate the question of whether goals are converted more readily than xG would imply in different game states.  Adding up xG data from understat.com and using data from whoscored.com (Opta), I made the following table:

Table 1

• I am looking at performance for teams that play in leading game states (same story for trailing teams, you just need to flip the signs)
• The third data column is the performance we expect from leading teams because leading teams will on average be stronger teams.  It’s a somewhat involved calculation I (attempted) to explain in this post: https://syzygyanalytics.co.uk/2025/08/01/soccer-team-rating-iii-game-states-i/

The red numbers on the right show leading teams underperform expectation.  I believe this is likely because of loss aversion (the losing team can find more motivation) and explains the increased prevalence of draws in football compared to what standard Poisson distributions would predict. 

I am somewhat surprised to see that goals have been converted at a slightly lower rate for teams that lead than would be expected (leading teams are underperforming their xG numbers shown by a +0.27 GD but a +0.37 xG difference).  This is because it seemed possible to me that chances while teams lead are likely to be better chances than the Understat model calculated (e.g. less defensive coverage). 

This difference is nearly all down to the Italian league but we are looking at nearly 3,800 matches so the fact teams underperform their xG here could imply loss aversion is even playing a role in finishing ability. 

Game state adjustment

My current perspective for a game state adjustment is as follows:

1. Teams that lead suffer a degradation in performance (likely due to loss aversion).
2. If teams perform worse when they lead it will continue to happen in the future as well.
3. To correct for game state, we want to look whether a team has led/trailed for a significantly different amount of time than would be expected for a team of that level.
4. For example, consider a strong team that has led for seven 90s (and tied for eight 90s) in the first 15 matches when we would only expect them to have led for a nett three 90s.  Using the averaged -0.18xG per game figure from the table above, the xG difference for this team has been damaged by the leading states by around 1.26 total xG (0.08 xG per game). 
5. However, in the next 15 games on average we expect it to be damaged by 0.54 total xG (0.035 xG per game).  The game state adjustment is then the difference between these figures – our game state adjustment will add 0.045xG per game onto this teams’ performance. 
6. xG is one of the least game-state affected metrics, we would have larger adjustments for something more shots/passes based. 

A bit of a slapdash article again but I am doing a lot of football betting as well so it’s difficult to dedicate a lot of time each week.  More to come in the future…

 Thanks for reading, please subscribe using the box in the top left on the home page and get in touch if you want to talk about anything.