Fantasy Football

FPL Underlying Metrics: Incorporating Team Statistics

Recently I have been asked on Twitter and in the comments section of this blog whether the Fantasy Premier League (FPL) underlying metrics formulas I have written about the past can be expanded upon in order to replace the ‘clean sheets’ metric. The problem with the Goalkeeper and Defender formulas I created is that ‘clean sheets’ are an effect rather than an underlying cause. A preferable metric for player assessment would be what happened on the pitch, such as ‘shots against’ or ‘passes conceded in the final third’. However, the source for these statistics, Opta via the Fantasy Football Scout website does not provide many ‘against’ metrics such as these.

What follows is the process I have gone through (for those interested, if not skip ahead to the interesting bit!), followed by my efforts to neutralise clean sheets in my analysis and incorporate team metrics in order to create an improved method of assessing players.

Why ‘Clean Sheets’ cannot be replaced with underlying metrics

A proposed solution I considered was monitoring the ‘for’ statistics for each team’s opponent on a given gameweek as a method of understanding the said team’s against column. For example, if Burnley play Watford and I know that ‘shots on target’ for Burnley stands at 3 and for Watford it’s 5, we can easily deduce that Burnley’s ‘shots on target against’ total is 5 and Watford’s is 3. The main obstacle I encountered with this was, when conducting retrospective analysis, stats from double gameweeks are bundled into one. For example, Opta’s stats may show that a double gameweek yielded 800 passes for Sunderland, but there is no way of discerning how many were against opponent A and how many were against opponent B.

It appeared that there could be some partial relief from, which provided at least some of the metrics – such as shots, shots on target, passes – on a game-by-game basis. This could allow me to fill in the blanks for double gameweeks, albeit via a laborious manual process.

However, it dawned on me soon after that the process of collecting team stats for individual performances was flawed because not every player will play 90 minutes. For example, if Kyle Walker plays 60 minutes for Spurs in a game where they concede 5 shots, how many ‘shots against’ should be attributed to Walker? Maybe all 5 occurred after the player left the pitch, or before.

For this reason, I have decided not to disrupt the formulas because I cannot find a satisfactory method for replacing clean sheets at this stage. I will keep thinking about it though, and hopefully a solution will present itself in time.

Introducing team stats anyway

I have been attempting to use my formulas to identify the potential of players in the opening weeks of the season, but these have been a little frustrating because they have not taken into account fixture difficulty. The work I have done trying to find a method of replacing clean sheets has resulted in me believing I should be putting an additional dimension into my assessments, so I have used the potential of the team to create and concede ‘big chances’ in the coming six gameweeks.

Below is a table of big chances (defined by Opta) created and conceded over the last four weeks (gameweeks 2-5).

So let’s consider Arsenal’s defenders; how many big chances are they expected to concede in the next six games based on their opposition? We can calculate a rough estimate by looking at the chances created of their next six opponents:

  • Chelsea: 2.25 per game
  • Burnley: 0.25
  • Swansea: 1
  • Middlesbrough: 1.75
  • Sunderland: 1.5
  • Tottenham: 2.5

Therefore, we can conclude that Arsenal will be expected to concede 1.54 big chances per game, on average, over the next six considering the form of the opposition. If we then combine that with the formulas, which measure the expected adjusted points per game, we can visually demonstrate an interpretation of the state of play. The following chart covers defenders (note: the clean sheets here have been neutralized by setting all players values to zero. This allows us to see which players would have the greatest potential with all things being equal).


Combining the metrics

As can be visually seen in the chart above, there are some players who are performing well individually – getting forward, having shots, etc – whilst others have a greater potential for clean sheets by facing opponents who are creating fewer big chances per game. The ideal next step is to combine the metrics into a single figure so that we may rank the players in order of potential. However, it isn’t that simple because the two figures we have are fundamentally different: one is individual performance per minute, the other is team performance per game. Smashing these to figures crudely together will not yield anything particularly insightful. What is required is a single unit of measurement, so I converted the figures into ranks:

  • 1 for the highest expected adjusted points per minute down to the nth value for the lowest
  • 1 for the lowest expected big chances conceded (a rank shared amongst all the members of the team) down to the nth value for the highest (note: the value nth value may not be 20 if two or more teams share the same rank).

By combining the average of these two individual and team metric ranks, we can produce a list of players in order of preference. The following table is the top defenders’ data above reimagined with ranks (gameweeks 2-5).



The data shows that Jose Holebas of Watford appears to be a key player to watch out for in the coming weeks based on his expected adjusted points per minute, where he is ranked 3rd, and Watford’s favourable fixtures which are ranked 1st. This leaves him with a Rank Average Score of 2. Manchester United’s returning Chris Smalling had two attempts against Watford in gameweek 5 – his first 90 minutes of the season – and so pushes himself to near the top with a Rank Average Score of 3 (2nd for expected points per minute and 4th for expected big chances conceded) despite a dodgy overall performance.

I have mentioned in previous blog posts the dangers of looking at such a limited sample size to make decisions, so caution is again urged and I encourage anyone reading these tables to use them as an informative guide rather than a decision maker.

What follows will be the tables for goalkeepers (like the defenders the clean sheets have been neutralised), midfielders and forwards. Rather than posting lengthy analysis on each list, I will simply highlight some names that I find interesting either because I feel they are being over- or under-valued by the FPL community. If you have any questions or want to discuss player potential further, you can find me on Twitter @artemidorus_1.


Players of interest:

  • £4.0m options Pickford (SUN) and Jakupovic (HUL)
  • Lloris (TOT) and Bravo (MCI) deceptively tricky fixtures upcoming which could limit his clean sheet potential

Please note that for midfielders and forwards, the opposition rank is defined by the expected big chances conceded by the opposition


Players of interest:

  • Sanchez (ARS) appears to be a must-own at the moment
  • Mkhitaryan (MUN) and Son (TOT) are misleading because they are not guaranteed starters and have played fewer minutes than others on the list. Check the minutes also of Stanislas (BOU), McClean (WBA) and Iwobi (ARS) before plunging on them as differentials
  • Mane (LIV) appears to be poor value when compared to club-mates Coutinho and Firmino
  • Hazard’s (CHE) star is falling fast, including a couple of recent price drops.


Players of interest:

  • The price tags of the heavy-hitters – Lukaku (EVE), Sturridge (LIV), Ibrahimovic (MUN), Aguero (MCI), Costa (CHE), Kane (TOT) – appear to be justified (although Kane’s recent injury will have an impact on his appeal).
  • Southampton forwards seem like they should be scoring more goals than they are, although the minutes for Long and Austin are a concern
  • Ighalo and Deeney (WAT) are coming to prominence
  • Rashford (MUN) will be a bargain if he can nail down a starting place
  • Defoe (SUN) and Benteke (CPL) might be overvalued by the FPL community

2 thoughts on “FPL Underlying Metrics: Incorporating Team Statistics

  1. Nice work, but I think it is a shame you go from absolute to relative estimation. This makes it impossible to compare players potential on diffrent possition and to decide whether it is worth to take a hit. Did you test created and conceeded big chances against actual goals scored and Clean sheets (CS) on a team basis? Did you test other underlying statistics against CS and goals scored on a team level?

    A second dimension is how tough / easy was the last 4 games from which a team created and conceeded big chances?



Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s