Rate My Fantasy Premier League Team

Fantasy Premier League: Analysis of Underlying Metrics

This summer I had planned to spend time with Pago researching the key indicators of Fantasy Football points. My aim was to identify the most important statistics that help us identify players likely to earn the most FPL points. And then another blog published an article along these lines!

So we republish this work with the permission of, and thanks to, author Peter Blake of Mathematically Safe.

Fantasy Premier League: Analysis of Underlying Metrics: The original article republished with thanks.

It was supposed to be simple, but it wasn’t. Following an unsatisfactory finish to the 2014/2015 season in which I finished a full 371 points behind the global champion, I entered the 2015/2016 season brimming with confidence. Armed with the conclusions from wide-ranging analysis of the previous season which I conducted through the summer of 2015, the theory was that I should be able to easily rise above the more than 3.7m Fantasy Premier League (FPL) managers and crack into the top 10k at the very least. Or maybe higher, after all I had just finished around the 54k mark with only my ignorance, and now I had a powerful headwind of a data-led strategy.

For the uninitiated, the FPL is an online fantasy football game provided by the Premier League for free. The premise is that you pick a squad of 15 players, playing 11 each gameweek and accumulating points related to their successes. Goals, assists, clean sheets, bonus points for the best players in a gameweek are all up for grabs, and really the only pitfalls are points conceded for yellow and red cards, infrequent events like own goals, and the risk of your charges being dropped by their real life managers. There are of course some curveballs to contend with, such as blank and double gameweeks (when a team doesn’t play leaving your man with an unavoidable score of zero, or playing twice with the opportunity to accumulate points over 180 minutes rather than the standard 90), free transfers (one per week), budgeting (£100m only), wildcards (as many transfers as you like for free, twice a season), and the introduction in 2015/2016 of The Chips as a way of boosting points on a selected gameweek. But on the whole, it’s an easy game to play. But it’s also bloody hard to master.

A disastrous second wildcard in gameweek 33 ruined my season. I plummeted back down the rankings, and although I was already well on my way to finishing a demoralising 248 points behind the eventual champion, the misstep meant my ambitions of the top 10k ranking were extinguished with alarming speed.

However, I can say with some justification that a rise of nearly 21k places and 111 points counts as progress, and up until the point that Jurgen Klopp started trolling me with his rotation policy the analysis I conducted before the start of this season had served me well. What I hadn’t paid enough attention to in previous work though was the underlying metrics. Goals, assists, etc. – the actions which generate all those lovely points – are Effects, the result of a confluence of circumstances, but the underlying metrics can be thought of as the Causes that lead to the Effects. A leading FPL fan site provides for a fee access to Opta data on player and team performance, and so in the wasteland between seasons (yes, I know Euro 2016 happened, but really who cares about that other than to monitor for form and injuries?) I have decided to look back at data from the past two seasons to see which if any of the underlying causes have an effect on the FPL points totals, whether they should be monitored, and to what extent.


For this analysis, I have assessed 568 players over two seasons – a total of 825 unique records – for their FPL adjusted points total (more on this in a moment) as the dependant variable and 54 performance metrics, providing a total of 44,550 data points. Only records where a player has recorded more than four games’ worth of minutes (360) in a season have qualified for this analysis.

A point to note is that in the forthcoming analysis, I will refer to ‘Adjusted Points’. The ubiquitous points-gathering method that transcends all players regardless of on-field position or in-game activity is appearance points, two for completing 60 minutes and one for 0-59 minutes. In analysis which is looking to identify independent variables relevant to points scoring ability, it is not necessary to include this information and so I have removed the total points accumulated by appearance for all players in the analysis. Therefore, taking two examples from the season just gone, the leading point scorer in the game was Riyad Mahrez with 240 points. But by virtue of 34 appearances of 60+ minutes and three more of under 60, he would have scored 71 points regardless. So, his Adjusted Points total is 169. At the opposite end of the spectrum, a moment of appreciation for Aston Villa’s Alan Hutton. The full back was awarded 54 points in the game just for playing a role in 28 fixtures, but such was the nature of his season, laden with cards of both colours and a phenomenal capacity to concede a vast amount of goals meant that he finished the season on 31 points, making for an Adjusted Points total of -23, which is several kinds of terrible combined into one hapless season.

So, what should be monitored?

The Defenders (n=283)

In an attempt to discover whether there is a correlation between the relevant metrics and the Adjusted Points total for the players I have used a statistical technique called regression analysis. The first thing to note is that there are not any sufficiently strong, stand-alone relationships to speak of. The closest we get to a ‘silver bullet’ metric that will explain perfectly the variations in the Adjusted Points totals is Touches Of The Ball, but with a R2 value (that is the strength of the relationship between Adjusted Points and the independent metrics, valued between the weak -1.000 and the perfect 1.000) of 0.495 the variation in points adjusted can only be explained by Touches half the time, which is not solid enough to form a conclusion.

Figure 1: Defenders’ Adjusted Points totals vs. Defenders’ Potential Adjusted Points totals, Touches as independent variable

In order to find a formula that explains the collection of points from defenders, first we must accept that the team plays a role an individual defender’s Adjusted Points total. Indeed, Clean Sheets is the single most important metric, with an R2 value of 0.767. This means that the success of the team in shutting out the opposition accounts for more than three-quarters of the variation in an individual defender’s total. This is more relevant than individual pursuits such as goals (0.321) and assists (0.240).

To improve the accuracy of the model to 82%, we can add in metrics that explain the individual’s performance from a defensive (CBI, or Clearances, Blocks and Interceptions, and Tackles) and an attacking perspective (Shots, Attempts from Set Plays, and Touches in the Final Third).

Adjusted Points =

(5.029*Clean Sheets)

(0.176*Attempts from Set Plays)

(0.015*Touches in the Final Third) 


– (0.104*Tackles)

– (0.032*CBI)

– 3.860

In simple terms, this means that for every Clean Sheet a team achieves, an extra 5.029 points can be added onto the player’s end of season Adjusted Points total, and for every Attempt from a Set Play add an extra 0.176 points, etc.

Mapping the Points Adjusted totals for the last two seasons with the total the model above predicts looks like this:

Figure 2: Defenders’ Adjusted Points totals vs. Defenders’ Potential Adjusted Points totals, multiple independent variables (Adjusted R2 = 0.820)

As this chart visibly demonstrates, there is a relatively strong correlation.

The most interesting thing to note about the equation above is that whilst it shows the incremental benefit of attacking activity (e.g. there is a positive benefit for shots, etc), points are deducted for defensive activity: for every tackle a player makes, 0.104 points are lost, for every addition to his CBI total, 0.032 points are lost. This means, in flippant terms, that defenders accumulate more points when they don’t have to bother with actual defending.

The data suggests then, perhaps obviously, that it is most profitable to pick an attacking-minded defender from a team with a good defensive record. This explains the names we see at the top of the Adjusted Points table over the last two years.

These players all represent defensively solid teams, and most have an eye for goal or attacking intent either as a set piece taker, set piece target man, auxiliary winger, etc. Indeed, the players who most exceed the projections of the model (see Residuals) are the ones which contribute a disproportionate number of goals.

This highlights that the prized assets are certainly those players which can score goals, but also that a solid team defence is the primary factor in the model and the basis for success.

Defenders: Next Season’s Strategy

  • Two Premium Defenders
  • Three Budget Defenders (rotating)

There is an argument to made for spending high on defenders from big teams who will guarantee clean sheets. In the 2015/2016 FPL season, the top 10 points-scoring defenders contributed between 10 and 18 clean sheets, which is as prolific as all but the very best strikers’ goals output, and for a fraction of the cost. However, an alternative view is that only two of the top 25 points scorers in the same season were defenders, whereas midfielders and forwards who performed only sporadically outscored consistently strong defenders, so there is an argument for going cheap and hoping for clean sheets on occasion whilst relying on the form of forwards and midfielders.

In reality, as with so many things, my preferred option lies somewhere between the two. Budget certainly comes into play here as it is obviously unfeasible to spend big across the whole team. Analysis of the 2014/2015 season that was conducted last year shows that premium defenders priced between 6.3-6.8m achieved a clean sheet achieved around 15 clean sheets a season, however for just 2.0m lower you will get 10 clean sheets. The incremental difference here is only a matter of 20 points. However, when we consider that a premium defender will score points comparable with an above-average midfielder for slightly less money, it suggests that having a couple of premium defenders in your team is preferable for the consistency; for example, an ~6.0m-valued Koscielny was outscored by the similarly priced Arnautovic by 12 points, however Koscielny with his 15 clean sheets was a more reliable source of consistent points compared to the explosive Arnautovic, with six or more points being scored 16 times to ten. Whilst this is an anecdotal example, it does illustrate that a player from a consistently solid defence can be placed into an FPL and essentially left to it. The absence of risk that we associate with an explosive attacking player means that there is security in a premium defender and, crucially, having a player who is a consistent source of good points reduces the number of players within the squad who are required to be transferred on a regular basis.

With this in mind we can, with foresight, build a defensive line that can be left to run with minimal changes (although clearly, there needs to be some flexibility required in order to respond to injuries, loss of form and trends; nobody imagined that Leicester would be such a reliable source of clean sheets in 2015/2016, and it would have been foolish to ignore it). In order to maximise the potential for clean sheets and attacking returns next season, I intend to field two premium defenders from top teams with some supplementary attacking potential (Bellerin and Alderweireld are proven examples of these type of players from the season just gone), with the third defender in a 3-4-3 or 3-5-2 formation a budget option coming from a lower-ranked team and part of a rotation strategy with two other similar defenders. With planning, the three budget players in the defensive line can rotate so the player with the most favourable fixture plays and every week the potential for three clean sheets is enhanced, improved by the occasional attacking return from the two premium players.

Whilst this approach will remove at least 2.0m budget that could be spent on more glamourous and explosive attackers, it will help in the hunt for clean sheets which we have seen is the primary indicator of FPL points and will also reduce the need for rotation in the transfer market.

Analysis: The Goalkeepers (n=65)

The analysis of the goalkeepers is much simpler. Like the defenders, they rely on a good team effort and Clean Sheets are the main predictor of points, with a 0.890 relationship. This relationship can be strengthened if we add in Saves and Touches to 0.955 (a very strong relationship approaching the perfect 1.000) with the following formula:

Adjusted Points =

(4.792*Clean Sheets)

+ (0.519*Saves)

– (0.027*Touches)

– 2.901

Figure 3: Goalkeepers’ Adjusted Points totals vs. goalkeepers’ Potential Adjusted Points totals, multiple independent variables (Adjusted R2 = 0.955)

This tells a simple story: points can be accumulated with Clean Sheets and to a much lesser extent Saves, but some points will be lost the busier the goalkeeper is (Touches) because goals are then inevitably shipped. However, as the table below shows, some goalkeepers from teams which attract shots will occasionally have successful seasons and rack up the points, although what separates a Tom Heaton from a John Ruddy is more difficult to comprehend, much less predict.

Goalkeepers: Next Season’s Strategy

  • One Premium Goalkeeper
  • One Budget Goalkeeper

Fundamentally the choice for FPL managers when selecting goalkeepers often comes down to this: one premium that will play and one budget who will sit on the bench and never get on, or two mid-range goalkeepers rotating. The data suggests that premium goalkeepers (that is, goalkeepers from the ‘top six’ sides) will keep between 11-16 clean sheets a season with a few exceptions (Chelsea’s 2015/2016 implosion, Arsenal’s 2014/2015 dropping of Szczesny, Lloris’ reliance on some very suspect Spurs defenders in 2014/2015). This means a solid pick like Hart, De Gea, Cech will collect a decent number of clean sheets, and if they are brought in and left, then an FPL manager can expect a 130+ point haul and hopefully never need to call upon the reserve.

The alternative is to pick up two mid-range goalkeepers and hope that with the correct rotation you can match or beat 11-16 Clean Sheets. Individually, they are unlikely to do it. Even the best mid-range goalkeepers over the last two years have relied on Saves to elevate them because their Clean Sheet count was not high (the exception being Watford’s Gomes in 2015/2016 with 11). Relying on Saves is a riskier business because you’re dependent on finding a goalkeeper in the form of his life (Heaton in 2014/2015, Butland in 2015/2016, etc), and proportionally few of the mid-range ‘keepers elevate to this level. So it a risk-laden strategy to pick two mid-range goalkeepers which rotate well with the hope of catching their 6-9 Clean Sheets a season each to match the performance of a Premium goalkeeper.

Having spent last season rotating mid-range goalkeepers, the security of points is an appealing one even when considering the implications of a tough fixture or a run of poor form. Therefore, next season I’m switching strategy to favour a single goalkeeper over two rotating ones.

Analysis: the Forwards (n=137)

The main sources of points for attacking players are individually-accrued through goals and assists. Therefore, whereas with defenders and goalkeepers we need to focus on the context of an individual player within the defensive unit they are playing in, with forwards this is reversed; it doesn’t matter who they play for, providing they can score or make goals.

Goals are a potent indicator of Adjusted Points: 0.952 (95%) of the variance in Adjusted Points can be explained by a player’s Goals count. By contrast, Assists account for only 0.480 (48%). This means that a striker’s ability to lay on chances for his team mates is a less apparent indicator of FPL success than the ability to score goals. This is reflected in the underlying metrics: a player’s Shots (0.791), Shots On Target (0.848) and Shots Inside The Penalty Area (0.783) are all better indicators of Adjusted Points than any of the ‘Proactivity’ metrics, the best of which is Touches in the Penalty Area (0.734).

Exploration of these metrics shows that it really isn’t much more complicated than a striker’s ability to hit the target. Combinations of metrics do not yield any formula much more conclusive than the relationship being Adjusted Points and Shots On Target; we can improve the formula slightly though by adding Touches in the Final Third, from 0.848 to 0.873. However, there aren’t any deeper causes that I can find. I theorised that in the modern game a striker needed to be combative, chasing down defenders and intercepting passes out from the back, but whilst this may be a fashionable style of play the relevant metrics do not improve our calculations here. I also thought that striker’s would need to be at the centre of attention to be successful, that Passes made or Passes Received would be indicative of more points, but that does not account for counter-attacking strikers like Jamie Vardy who have achieved success in the last couple of seasons.

The numbers reveal that, over the course of the season, the principal metric to monitor in order to predict points is Shots On Target, supplemented by Touches in the Final Third, as evidenced below. The formula for this relationship is:

Adjusted Points =

(1.697*Shots on Target)

+ (0.034*Touches in the Final Third)

– 5.670

Figure 4: Forwards’ Adjusted Points totals vs. forwards’ Potential Adjusted Points totals, Shot on Target and Touches in the Final Third as independent variables (Adjusted R2 = 0.873)

Considering the prominence of Shots On Target, it obviously pays to monitor this metric throughout the season; if a player is having plenty of shots on target but is not picking up points, then it is a reasonable assumption that he is owed some points.

Of course, there are outliers; players who have fewer shots on goal but score more than the data suggests they should, and vice-versa, but the chart above shows that these occur most frequently at the lower end of the Adjusted Points total. Examples include West Ham’s Jelavic in 2015/2016, who scored an additional ten points despite recording only one Shot On Target, or Mario Balotelli in the previous season, who scored only an additional six points despite 19 Shots On Target. It should be noted though in both these examples playing time and total points were low (Jelavic: 25 points from 408 minutes; Balotelli: 32 points from 939 minutes). With a lower sample size there are more likely to be outliers, but these are not the type of players who you would want if you’re looking to build a successful FPL campaign, where minutes on the pitch is important for racking up a strong total. Many players will burn brightly for a short space of time, as Jelavic did in this example, but predicting where and when this will happen is not sustainable. The further we move up the ladder, to more Adjusted Points and more Shots On Target, the more consistent the numbers are in line with the predictions.

More consistent that is, with the exception of Harry Kane. In the 2015/2016 season registered a scarcely believable 21% more Shots On Target than the nearest competitor (Aguero 2014/2015). However, his Adjusted Points total if we take Shots On Target as the only independent variable of 164.7 did not live up to expectations and he fell short of this by almost 30 points, making him in the strictest sense of the phrase one of the most inefficient strikers in the last two seasons. Regardless, such was his productivity in creating scoring opportunities he remained a prize FPL asset, highlighting the importance of Shots On Target as an absolute measure rather than as part of a relative efficiency equation, for which Jelavic would come out near the top. The following table details the leading forwards with the two independent variables factored in, Shots On Target and Touches in the Final Third.

Forwards: Next Season’s Strategy

  • Flexible

The difficult thing to acknowledge about forwards is how unpredictable it is to pre-empt who will be a success. There is an assumption that the leading strikers from the big teams will receive plenty of opportunity to register Shots On Target, however for every Aguero and Kane there is an underperforming counterpart such as Rooney or Benteke in 2015/2016. Equally, it is assumed that promoted teams will not have the ability to dominate games or fashion chances, and so their strikers will be starved of opportunities, but this doesn’t explain the success of budget players like Austin in 2014/2015 or Ighalo in 2015/2016.

My strategy for next season therefore is to remain as flexible as possible with the forwards in the early stages, monitoring Shots On Target closely to see who has been an active goal threat on a week by week basis and using the weekly free transfer to keep the forward line in rotation when needed. Too often FPL managers (myself very much included) will fall into the trap of thinking that a good or bad run of form “can’t carry on much longer”, and so are guilty of holding onto players who are not scoring or not buying players who are.

Shots On Target is a much sturdier metric by which we can assess the form of a striker which enables us to see past the Goals, which are a precious and rare commodity. If a forward scores two Goals from just two Shots On Target, it is reasonable to assume that he is not worth picking, because the shots on target total is so low and therefore the good form is unlikely to continue. By contrast, if a player scores a hat-trick in a game but has achieved four Shots On Target for each of the last four weeks, it is more likely that the hat-trick will not be a one-off.

The FPL game over the last couple of seasons has been kind to us in terms of strikers, because it has provided a limited number of stand-out candidates from a range of prices, and the consistency of these players has made retaining them for weeks at a time an easy decision. However, predicting where these players will come from is difficult. So, counter-intuitively for someone who plans as much as I do, the theory goes that it matters less who you start the season with, providing you retain agility to get in the form players quickly, and retain them only when they show the consistency of an Aguero, Kane, Ighalo, etc.

Analysis: the Midfielders (n=340)

The assessment of the key underlying metrics for Midfielders is immediately apparent to be more complicated than the forwards. The R2 values for key metrics to explain the variance in Adjusted Points show that Goals again are the primary indicator of points, however the relationship (0.835) whilst strong is not as clear cut as the forwards. The principal cause of points is again Shots On Target, but with a R2 value of 0.754, it is suggests that around a quarter of the variance is explained by other factors. This means that we can be less confident of a midfielder’s points total by looking at his Shots On Target than we would be of a forward’s.

To discover the other causes required to improve the relationship, we must consider the other primary source of points of an attacking player in FPL: Assists. Whereas the importance of Assists towards a forward’s Adjusted Points total was less than half (0.480), for midfielders it significantly increases to 0.588. Therefore, this leads us to a hypothesis that the prominence of a midfielder in the FPL points-scoring universe is dictated by the joint ability to make and create goals.

However, the introduction of Passes (and variations of) into the equation does not improve the strength of the relationship notably, if at all. So, the hypothesis needs revision. But if we think about Passes made and Assists as by-products of a midfielders’ ability to put himself at the centre of the attack, then we start to consider things in a new light: a midfielder’s points-scoring ability is related to his possession of the ball in key attacking areas.

Considering this new hypothesis, we strengthen our formula if we add to Shots On Target not the midfielder’s end product, but the amount of the ball the midfielder sees in the attacking third:

Adjusted Points =

(1.988*Shots on Target)

+ (0.019*Touches in the Final Third)

+ (0.034*Passes Received in the Final Third)

– 5.854

The adjusted R2 value of this formula is 0.818, which is by no means perfect but does represent an improvement on the simple Shots On Target relationship.

Figure 5: Midfielders’ Adjusted Points totals vs. Midfielders’ Projected Adjusted Points totals, multiple independent variables (Adjusted R2 = 0.818)

What is notable here is how much more distributed from the trendline the top performers are when compared to the forwards. This indicates that there are some other factors also at play, possibly dictated by the role each player has in the team. For example, the significant overachievers here are Mahrez (2015/2016) and Hazard (2014/2015), who were both in the form of their lives and direct, auxiliary forwards for title-winning sides, Ozil (2015/2016) who is a master at finding and exploiting space on the pitch and who subsequently accumulated a vast number of assists, and Payet (2015/2016), whose accuracy from dead-ball situations was unnervingly consistent. By contrast, another playmaker, Coutinho, underachieved in both seasons due to a combination of mis-firing strikers and a lack of consistency in team selection which meant he was on the ball a lot but the team were not able to make the most of his positions.

Midfielders: Next Season’s Strategy

  • X3 Premium Midfielders (‘Top 6’ clubs)
  • X2 Mid-Range Midfielders

Whilst the forwards are an unpredictable, high-risk group that demand flexibility, the midfielders are by contrast a lot more predictable. Of the top 30 midfielders ranked by Adjusted Points totals over the last two seasons, 18 have come from the six ‘biggest’ clubs (Manchesters City and United, Arsenal, Chelsea, Tottenham, Liverpool – apologies to Leicester fans). Therefore, whilst the personnel may change throughout the season, it is safe to say that significant budget is required to be spent on the midfield. The players chosen, as we have seen, will need to be played high up the pitch because their ability to find space, get amongst the play and shoot when necessary is a good predictor of points.

The prospect of picking five such players is unrealistic, so as with the forwards, a season can turn on the jokers in the pack. Last season budget ‘enablers’ such as Mahrez and Alli came to the fore, which helped many FPL managers, but as with the forwards the capacity to predict who these are going to be is limited. Therefore, being alert to the best fourth and fifth midfielder options is key; the first three, if selected from the big six clubs, will be relatively consistent over the course of a season, but finding the cheap differentials before they rise to prominence will enable budget to be spent elsewhere.


  • A Goalkeeper’s success is dependent on the defensive unit he is a part of, making the acquisition of a premium goalkeeper for consistency throughout the season a desirable option over the high-risk forecasting of when a mid-rank goalkeeper’s team will keep a Clean Sheet
  • Defenders also depend on the team’s defensive performance, so Clean Sheets must be the primary target, with attacking returns secondary. Ideally, the defender should engage in the least amount of defending activity possible
  • Midfielders who see a lot of the ball in the attacking third should be prized above all others. Consistent points can be found amongst the attacking midfielders of the big clubs, and FPL managers should be hyper-alert to the emergence of ‘enabling’ budget midfielders as these are difficult to predict
  • Forwards who hit the target most often will score more points, so Shots On Target should be monitored closely. However, predicting where the most effective players will come from is difficult in the early stages of the season, so flexibility from FPL managers is important until a consistent pattern emerges
  • The strategy put forward here facilitates ten positions which favour consistency over the course of the season with occasional changes as trends develop, and five positions (x2 midfielders, x3 forwards) which will be subject to heavy rotation, chasing form and searching for high-point scoring consistency.

So there you have it: a FPL strategy born out of data analysis. It simultaneously reveals the complexity inherent in the game that makes it so hard to master on the one hand, and on the other provides the tools to see through the murkiness and succeed. The only thing left is to go forth and conquer, right? That’s what I thought last season before it all went so horribly wrong.

Forecasting based on past performance can be a useful addition to an FPL manager’s tool box because it partly removes subjectivity from the equation. We will of course make subjective calls on a week-by-week basis based on recent form of the opponents, teammates, forthcoming fixtures and rotation policies of managers; these can be effective additions to our models which help us to predict where points will be forthcoming. However, we can bias decisions based on what we see with our own eyes; an edited set of highlights on Match Of The Day may show a player in a much deeper position than we would expect them to see, and though they may have registered touches in the final third, shots on target and other indicators that they will have a successful FPL season, we may ignore them because of what we have seen and a media-driven narrative around an upcoming English forward that has just broken into the team that could threaten our player’s role, for example.

Statistics are a way of seeing past our own prejudices, and any statistician will tell you that long term analysis is more robust than short term. But here we encounter our flaw; the sample size for this analysis is still quite small. I have taken care to assess any model I have presented here for chance, and none of the values attributed to the independent variables or intercepts, when explaining the fluctuations in dependent variable, are likely to be there by chance. For example, if we add Attempts From Set Plays into our midfielders model, the R2 doesn’t change significantly, meaning that the data can be accommodated, however the p-value (or chance that the metric’s relationship to the model has been generated randomly) is at 82%. Therefore, there isn’t any significant relationship between Attempts From Set Plays and our existing model. None of the inputs into the models presented here have a p-value of more than 0.01 (1%) and most are substantially lower.

However, in spite of this, we still need to approach the models with caution for the following reasons:

  1. None of the models have an R2 value of 1.000, which means that despite the best efforts to explain what happens to the points to the underlying metrics that are generated on the pitch, there will always be innumerable seemingly random, unforeseen acts that will affect the outcome of a game, and subsequently a player’s FPL points.
  2. We are still dealing with a small pool of elite players. If we were analysing the season’s performance of thousands of players we could be a lot more confident in our models’ predictive abilities. As it is, we are still limited to at most 340 season stats of midfielders, and only 65 goalkeepers.

Regarding the first of these points, figures 2 (defenders), 4 (forwards) and 5 (midfielders) illustrate this point well. To take an example, Mahrez in 2015/2016 was projected to have scored an Adjusted Points value of 109, but he scored 169. There are inevitably unforeseen acts that determine a player’s points, not least bonus points, which raise the idea that a player’s performance is determined in part by their relative performance in a game in addition to their absolute performance as described here. However, it does show that there are some players – extremely important players from a FPL perspective – which are being undervalued by these models because the projections are being ‘dragged down’ by underperforming players. Sole dependence on a model such as this would have rated Mahrez’s early season form as an overestimation and the logical interpretation would have been to avoid him because his numbers would eventually correct downwards to align with the model. That drop never arrived.

Therefore, subjective interpretation and analysis of the numbers is still required. This analysis is not intended to provide a definitive answer, a set of rules that can be followed to the letter and success will follow. If football was predictable the bookmakers would have gone out of business decades ago and the rest of us wouldn’t watch. Rather it is intended as a guide that allows FPL managers to understand the longer-term trends and focus on the metrics which are underpinning long term performance whilst we navigate the vagaries of the short-term. In short, the analysis here is one of many signal indicators, one which I personally have never had before, but is not the definitive answer.

Republished with kind permission of, and thanks to, author Peter Blake of Mathematically Safe.


  • Bio
  • Latest posts


Site founder, first used stats to try to win his work mini-league. Now helps other managers win theirs.

30 thoughts on “Fantasy Premier League: Analysis of Underlying Metrics

  1. My favorite part is “subjective interpretation and analysis of the numbers is still required”. This is why the RMT Bot combined with the opinions of experienced managers should deliver the best outcome.

    • @ChorleyRocks Saw this article before & thought it was a thorough analysis of past data to giv eus an insight into which stats to monitor at each position. Thanks for making it part of the site.

  2. well written, needed two cups of coffee to read it twice and wrap my brain around some of it. thanks for all the effort

  3. This article is interesting for sure; but what about the idiots like me that always get the butterfly effect?

  4. Bravo to the author, what an amazing piece on statistics in FPL. :bow: :bow: :bow:
    Would have never imagined that I would see econometrics applied on fantasy football league. I love that it is a bit simplified after taking consideration the target group. :thumbup:

    Just wanted to ask, what was the decisive criterion for choosing the “best” model, the R-squared? And what kind of “regression” have the author used – simple OLS, some weighted least squares regression or something a bit more advanced to deal with some possible problems the linear regression models might present?

  5. Chorley, Do you have contact with the person who wrote this article ? I would be interested to know if he knows how to do principal component analysis or PCA. This would allow him to enter all the pertinent statistics he was interested in and determine the best correlation line through the data in that n-dimensional space.

    Thanks in advance for a reply…. Beautiful article.

  6. Great article Chorely, the level of detail and thought put into the piece is inspirational!

    For those of us that would like to replicate a similar ‘moneyball’ approach in trying to use data to buy the most points for our buck, can I ask how you are scraping the data to run your analysis? I’d very much like to attempt this myself using my own model.

    Kind Regards

Comments are closed.