Putting Fantasy Football Strategy To The Test

Having been away from Fantasy Football for a few seasons, I returned last season following the tantalising prospect of winning a £30 wager from a colleague with whom I discuss football frequently. I’m happy to say I put him in his place by a commanding 414 points, although from a strictly monetary perspective my £-per-hour return from the effort invested throughout the season put me firmly in the ‘slave labour’ category on the economic ladder. The other sobering fact of my season was that I finished a full 371 points behind the global winner and outside the top 50,000. For many of you on this site, that qualifies me as someone to be firmly ignored when it comes to Fantasy football. But to those people I say: at least it’s not my colleague writing this.

In the close season, I’ve been deconstructing the final point tallies of the 530 players who played in games for at least one minute throughout the 2014/15 season to help me understand why my performance was, by the standards I was aiming for, average. I realised early on that there are certain assumptions that I make when picking players that govern my strategy throughout the season, and I have been testing these against the data to see whether they are valid hypotheses which will form the basis of a successful season. Below are the key findings I have made regarding three of my assumptions. Whether this will help me become a better Fantasy player, we’ll have to see in 2016.

Assumption 1: Follow the form

My first port of call when creating a shortlist of players to bring in is the ‘form’ filter in the Fantasy game. The theory goes that if a player is in good form, then that is who I will bank on to continue to bring in points. There are anecdotal examples of this throughout the season, from Harry Kane to Charlie Adam. However, it seems just as frequently there are isolated examples of players securing large hauls seemingly from nowhere, such as the season-high 24 points from Crystal Palace’s Yannick Bolasie, who scored just two points the week before.

Simon March, aka Fantasy Football Scout community member Dufflinks, won the Fantasy Premier League (FPL) this season with 2,470 points. To achieve this, he scored an average of 65 points per Gameweek, or 5.4166 per player (10.833 for a captain). Therefore, because indefinite transfers are not an option, it is wise to look for runs of form where the players in your squad are scoring an average of 5.4166 (rounded up to 6) per week in order to compete at the top. Six or more points was achieved on 10.40% of occasions.

This means that there were 1,988 instances – an average of more than 52 a week – where players were scoring ample points to achieve the points target. However, runs of form where this total is achieved on a consistent basis (averaged over six consecutive games) occurred only 650 times throughout the season. In other words, it’s a pretty rare occurrence, so identifying when they are going to happen is tricky.

The question becomes what is the trigger for identifying these runs of form? I had been hoping it was form. Therefore, I looked at what happened in the six games after a high points total was scored (six or more points were scored on 1,890 occasions) and compared these scores with what happened in the six weeks before the high score.

I was expecting to find that the average points per game after the high point hauls was higher than before, but in truth this was rarely the case to a significant degree. For example, there were 694 instances when a player scored 6 points in a game; the average of the 4,164 games that preceded these events was 2.64 points per game, and the 4,164 games which followed these events generated an average of 2.68. This represents a barely perceptible increase of 0.04 points per game. In only 171 (9.0%) of the 1,890 instances whether 6 or more points were scored did the player go on a six-week run which returned more than 5.42 per week.

Conclusion: The data suggests that large point hauls happen largely in isolation; in 91% of the instances following a large point haul the player will not go on to produce form worthy of a championship-winning team. Therefore, it is statistically unlikely to be worthwhile bringing them into your team after the point haul. That said, there were only 650 championship-quality runs in the season; 171 of these were after a significant point haul (26%). Considering the unpredictability and scarcity of the events, a large point haul appears to at least provide a fighting chance of identifying the runs, although the odds are still stacked against.

Assumption 2: Play the fixtures

The conventional wisdom in Fantasy football is to play the fixtures; target the strong and avoid the weak. Sometimes this flies in the face of form; Boaz Myhill, the West Brom goalkeeper deputising for the injured Ben Foster, had scored 17 points in two games before his team went to Old Trafford to face Manchester United. I am as guilty as anyone in thinking he needed to be shipped out, despite his good form and despite Man Utd not scoring in their previous two. My happiest stroke of luck in the season was being forced to stick with Myhill as he made save after save, rebuffed a Robin van Persie penalty and kept a clean sheet on his way to a 17-point haul; it was the highest single game points total of any goalkeeper throughout the season.

As a way of testing whether the fixtures are a good method of picking players, I considered that there were two major factors I would use to assess teams: form and reputation. I created a form index, which assesses the reputation of a club (1-5, based on pre-season perceptions and final league position) and form (0 for a loss, 1 for a home draw, 1.5 for an away draw, 2 for a home win, 2.5 for an away win) to put a value on a club’s performance; for example, in the Myhill example above, the Form Index registers 4 (Man Utd reputation) * 2.5 (away win) = 10. The average score over the previous six games contributes to the overall Form Index score.

Using this method, I discovered that the average points per player increases the higher the Form Index of the player’s club, although once the Form Index reaches above six, it begins to decline again. Furthermore, the average points per player decreases as the Form Index of the opponent increases.

The evidence is reasonably compelling and my original assumption of ‘play the fixtures’ appears to hold true. The data shows that if a team is in good form, the average number of points scored by its players is high, whereas poor form correlates to a low potential for points. Accordingly, an opponent in poor form is more likely to concede more points than a strong opponent.

However, the data above has looked at the average strategy over the course of the season, but as with the player form explored above, identifying the individual explosions of points that appear almost spontaneously is more of a challenge.

Of all the occasions (67) when a player has scored 15 points or more in an individual game, 37% were scored by teams who finished the season in the top four, whereas only 12% came from teams in the bottom four, indicating that picking players from high-reputation teams improves the chance of finding one of these explosive events. However, the reverse is true when we consider the opponents; 7% of the 15+ hauls were scored against top-four teams, whereas 33% were scored against the bottom four.

Conclusion: The data tells us that my instincts were correct; go with players from the in-form, big teams, especially when they are playing the weaker teams.

Assumption 3: Pick defenders with assist potential

On Gameweek 1 of the 2014/15 season, my team lined up with a four-man defence of Gary Cahill (Chelsea), Glen Johnson (Liverpool) Mathieu Debuchy (Arsenal) and Aleksandar Kolarov (Man City). Cahill’s inclusion can be put down to my belief in the impenetrability of a Jose Mourinho defence. The other three were there as representatives of big clubs with a potential for assists.

My philosophy can be summed up as follows: goals for defenders are rare, therefore they are so unpredictable they are not worth chasing; clean sheets are fragile, with a solitary error from any defender costing the whole backline the points; assists are more common than goals, and more predictable than clean sheets, and so are worth investing effort in chasing. To explain this last point, I believed that the proactive, repeatable approach of continually delivering crosses into the penalty area would reap better rewards than hoping for no errors to secure a clean sheet. So my strategy (at the start of the season at least) was to chase the assists.

Turns out, this was nonsense. The data shows that the high point scorers have high assists sometimes, but it is by no means a guarantee. Perhaps the biggest folly of my strategy was the misapprehension that assists are far more frequent than goals for a defender; they are more common (143 vs. 91), but not by a big enough margin to be significant when we consider that goals for defenders are worth 6 points against the 3 for an assist. In this event, we discover that there were more points awarded to defenders for goals (564) than assists (429).

Assists only contribute 4.5% of the total points awarded to defenders throughout the season, and despite the fact that 40% of defenders to complete at least one minute throughout the season gathered an assist, only 32 of the 168 players (19%) gathered more than one (three points’ worth).

This begs the question: what is the key metric for defenders? The answer, perhaps obviously when we consider the primary role of a defender, is keeping the ball out of the net. Clean sheets contributed 36% of all defender points and was the most strongly correlated (measured via linear regression analysis) with total points throughout the season. Although they are more fragile (e.g. lost by a single error), even the worst teams in this discipline (QPR) provided a total of 26 clean sheets across seven players, whereas the same club’s attacking returns totalled just six events (four assists and two goals). The data reveal that clean sheets are the greater predictor of a player’s performance throughout the course of the season.

The other thing to consider at this stage is whether it is worth investing heavily in defence at all. The price range of 6.3-6.8 achieved a clean sheet every 2.48 matches on average, however 2.0 lower than this (4.3-4.8) achieved one every 3.66 matches. It suggests that if you can find a guaranteed starter you will get 10 clean sheets a season (approximately) for 2.0 less than 15 clean sheets a season, a difference of around 20 points. Of course, these are crude estimations, but it does seem to imply that a cheap defence might be worth the sacrifice.

Conclusion: There were defenders who supplied plenty of assists (the dependable Leighton Baines and Branislav Ivanovic of course, as well as the surprising Daryl Janmaat of Newcastle contributing eight), but on the whole, depending on assists for a defensive points strategy was not a sound idea. Clean sheets are the undisputed key indicator of points success when picking a defender.