This post is an accompaniment piece to Player Ranking posts that I will be producing throughout the 2017/18 season, explaining the methodology behind each of the ranking tables produced.
Methodology: Assessing Player Form
In the post-2015/16 summer period I performed regression analysis on the underlying metrics supplied by Opta (via the Fantasy Football Scout members area) to see whether there was a correlation between the actions that players take on the pitch and their FPL points. The objective was to focus not on – for example – the goals scored which are rare events, but rather whether a striker was getting himself into the position to get goals.
I have since re-done this analysis to include the 2016/17 data, which extends the data set to include three seasons’ worth of data. The correlations (the adjusted R2 values) for each position have changed very marginally downward in three cases and upwards in one, but the changes have been so minimal that I retain confidence in them. The formulas have changed too, to:
- Adjusted Points = (Clean Sheets*4.40731) – (Touches*0.01038) + (Saves*0.29405) – 2.44533
- R2 = 0.94 (-0.02 from last season)
- Adjusted Points = (Clean Sheets*5.25853) + (Touches Final Third*0.01195) + (Shots*0.53035) – (Tackles*0.12006) – (CBI*0.03539) – 2.9066
- R2 = 0.82 (-0.0)
- Please note that ‘Attempts from Set Pieces’ has been removed from the original formula as the latest data reveals there to be no impact on the overall correlation
- Adjusted Points = (Passes Received Final Third*0.04914) + (Shots on Target*2.04348) + (Touches Final Third*0.00985) – 6.1331
- R2 = 0.84 (+0.02)
- Adjusted Points = (Shots on Target*2.00862) + (Touches Final Third*0.02409) – 5.19914
- R2 = 0.86 (-0.02)
Please note that ‘adjusted points’ is the FPL points that the player was expected to score minus the appearance points which were removed because they are not related to on-pitch actions. With regards to the player rankings I will use these formulas in two ways:
Column: Total Adjusted Points as % of Expected Points, Season to Date
Compare the season-long metrics against the total FPL points scored by an individual player (minus the appearance points, described above as ‘adjusted points’).
- Those with better underlying stats than FPL adjusted points can be seen as ‘underperforming’ and, theoretically, due a haul of points to revert towards their expected points.
- Those with more FPL adjusted points than the expected points would suggest they should have got can be viewed as ‘over-performing’, with the implications being that they are either essential assets or are due a dry spell to revert back towards their expected points.
Column: Form Rank: Adjusted Points per Min, Last 4 Games
Looking at the underlying metrics of the last four games only to assess ‘form’. These calculations will take the form of ‘expected adjusted points per minute’, which shows what the actions on the pitch would have been expected to return in terms of points every minute. Please note however, that this can lead to ‘impact’ subs being high in the rankings despite low minutes which make them unviable FPL assets.
- Note that for defenders and goalkeepers here the clean sheets have been neutralised to zero for all players. This is because we want to assess the individual performance of players rather than the impact of the team on his performance. The potential for future clean sheets and its effect on the rankings are compensated for by assessing the ‘fixture strength’ of the following six games, see below.
- Note also that players must have played at least 90 mins in the last four to qualify.
Methodology: Assessing Fixture Strength
Column: Fixture Rank: Team Attacking Opportunity, Next 6 Games
Here I look at the big chances created and conceded per game of the player’s next six opponents over their last four games. For goalkeepers and defenders, the lower the big chances created by the forthcoming opposition indicate a greater potential for clean sheets (as mentioned above). For midfielders and forwards, the greater the number of big chances conceded by forthcoming opposition is an indication of potential attacking points.
Methodology: Overall Player Rankings
Column: Overall Rank, Fixtures + Form
For the final step I rank each player by ‘form’ over the last four games, and then by the ‘fixture strength’ of the forthcoming six games. The overall positon of the player in the list is generated by ranking the score you get when you take the average of these two ranks. For example, if a forward is ranked no.3 for form and his forthcoming fixtures are ranked as no.11, then his overall rank score is 7 (3+11=14/2=7). This rank score of 7 is then ranked against other players’ rank scores.
Methodology: Next Gameweek Rankings
Column: Overall Rank (Next GW Only)
This is the same as the Overall Player Rankings, expect the ‘forthcoming fixture strength’ looks at the next opponent only.
If you have any questions, I can usually be found on Twitter @mathsafe_fpl