Below are advantages and disadvantages of xG data sources one can use for FPL. Why xG and expected assists (xA) are better stats than goals and assists has been widely discussed. For example, Michael Caley (@MC_of_A) has done a lot of work on the benefits of xG:
This twitter thread is another explanation on xG from @FPLSeles:
Fbref has the best publicly available expected goals data. Their expected goals data comes from Statsbomb. Statsbomb’s xG model is the best publicly available model as it’s the only model which takes into account defensive pressure on the shooter and the number of defenders between the ball and the goal.
Fbref also publishes statistics from Statsbomb’s post shot expected (PSxG) goals model. This is similar to their normal expected goals model but based on ‘how likely the goalkeeper is to save the shot’. Therefore, PSxG statistics can be used to evaluate the shot-stopping ability of goalkeepers.
Fbref gives users the options of exporting its tables to excel. Another way their tables can be obtained is by writing a program in R/Python to scrape its website.
Understat is the other main website with publicly available xG data. The xG model used is worse than Statbomb’s, as it doesn’t take into account pressure on the shooter or the number of defenders between the ball and goal. There’s no easy way to export Understat data into Excel but there exists an R package which scrapes data from the Understat website: Understatr. Whilst the xG model isn’t as good as Statsbomb’s, its much quicker to access Understat data via UnderstatR than it is to scrape Fbref data from its website. Furthermore, it could be argued that slightly more inaccurate xG statistics has little effect on accurately predicting FPL outcomes.
Whilst Fbref publishes more statistics that Understat, the website can feel old-fashioned and not user-friendly. Another useful feature of the Understat website is that it’s easy to get breakdowns of xG and xA by position, which can be useful for a player who plays multiple positions.
Infogol is another website with publicly available xG data. Whilst this data isn’t as accessible as from Fbref or Understat, it contains xG stats on championship players and teams. This could be useful for making predictions for newly promoted teams and players.