The coverage debate is a story that seems to go on and on. In truth, it is a tough topic to analyse in great detail because it plays on so many different aspects. You can argue for or against coverage based on risk versus reward and on expected returns, but also out of completely different aspects such as that having players feature in our teams can make an otherwise dull game interesting to watch.
This article aims to contribute to this debate by crunching some numbers and presenting a little bit about the range of results that can be expected when applying a coverage tactic compared to stacking on players from one or more teams. Please note that it this is exclusively targeted at offensive coverage, meaning midfielders and strikers.
Preparations
In order to examine the influence that coverage has on our expected returns we need to set up a controlled environment. The point of doing this is to rule out other factors interfering with our results.
First of all, we assume that an offensive player has an average Goal Involvement rate of 33%. For those that are not familiar with this stat, it means that they have a hand in 33% of their team’s goals, either by assisting or by scoring. For reference, two of last season’s gems, Dele Alli and Kevin De Bruyne had a Goal Involvement rate of 36.7% and 39.7% respectively. While the exact number will not matter, it is nonetheless a reasonable assumption.
We further assume that each Goal Involvement brings an average net benefit of 5 points, without trying to distinguish between goals and assists. Pitch time, yellow cards, clean sheets or any other factors affecting points are not included and will not be treated.
Combining the fact that players are involved in 33% of their teams goals and that each involvement brings 5 points, we can say that on average each goal scored by a team is equivalent to 1.65 points per player featuring from that team.
Picking four offensive teams, such as Tottenham, Manchester Utd, Manchester City and Liverpool, we estimate the following table of probabilities that the teams score a certain amount of goals in a given game. Each team has been given identical probabilities, so the table is valid for all teams.
Probabilities of goals scored
Over a season this equates to 83.6 goals per team, which is a high but still reasonable number for a title contender.
An example of what the table tells us is that in a given Gameweek the probability of all four teams scoring exactly three goals each is 0.25*0.25*0.25*0.25 = 0.4%.
We are now ready to define our different coverage-models which we will evaluate against one another. The base condition is that we have four players and we may split them up in any way we want between the abovementioned four teams. Price has been completely disregarded, and all players are assumed to be equally potent with a goal involvement rate of 33% as previously mentioned.
Coverage features one player from each team and which will be denoted as (1-1-1-1).
Mix features one double up, and thus is forced to miss out on covering one team (0-1-1-2).
Double up features two double ups (0-0-2-2).
Triple up features one triple and a single player (0-0-1-3).
Results
Based on the table defined in the previous section, the number of possible combinations of goals scored in a single gameweek for four teams is 1296. In order to evaluate every single combination a small script was written to take care of the calculations. One example of such a combination can be Spurs scoring two, Man Utd three, Man City no goals and Liverpool one. For each combination the average points returned is calculated together with the probability of that specific combination occurring.
The results of running through every possible score combination for each coverage model can be seen in the charts below.
Point ranges
The charts show the different spreads of points for the different coverage models that we have set up. Interestingly we see that they differ quite a bit from one another. The blip at 16 points is a result of that particular score requiring some pretty uncommon circumstances to happen.
Starting with Team Coverage we see that their curve is fairly symmetrical around the values in the middle of the points range. Team Mix is showing similarities but is a little more spread. Team Double Up is the one that perhaps stands out the most. They tend to get points in more coarse increments than all the other teams, due to the fact that they have two double ups and two blanks in terms of coverage. Finally, Team Triple up sees the most widespread results, reaching all the way to the top of the points range but also featuring scores at the very bottom.
Analysis
We can use the above results to create the following table, which shows how likely the different coverage models are of scoring below 10, between 10 and 20, or above 20 points.
We immediately see that Team Coverage is by far the least likely to do really poorly. They only fall below 10 points in our model 9.85% of the time, whereas Team Triple up do so 22% of the time. It is somewhat of an expected behavior considering that Team Triple up have put almost all of their eggs in one basket. On the other hand, Team Coverage is the least likely by some margin to do really well, and Team Triple up is the most likely to do really well, doing so in 17.75% of the time which is more than twice as often as Team Coverage. The results indicate that by opting for coverage, high-end scores are sacrificed in order to gain some stability in avoiding the lower scores.
Team Mix and Team Double up report very similar numbers across the board but there are minor differences in their approach towards the extreme values 0 points and 30+ points which can be seen in the graphs.
Is this enough to decide which strategy is the best one?
Not by any means, but there are some trends here that suggest that going coverage may be a way of somewhat reducing the variability in which scores you do see at the end of a Gameweek. If you seek to do big hauls every now and then, this study shows that you are more likely to do so by stacking your players rather than spreading them.
Conclusion
It is interesting to see that it is possible to find some merit for the coverage theory being a safer option compared to stacking players from certain teams.
However, considering that the results are not overwhelmingly in the favour of coverage, it remains unknown if it is worth it from a points perspective to choose a player solely based on coverage. When faced with a decision of individual player quality versus team-coverage, I believe that player quality is the way to go, even if that means doubling or tripling up.
As far as applying this to the live game goes it is important to remember that many factors were excluded from this study. We cannot determine how much the coverage spread influences the outcome compared to other factors that were controlled for this study. In reality, all teams do not score equally many goals, and players most certainly do not have equal goal involvement. While a good player may average 33% goal involvement over a season, it is likely to swing from much lower figures to much higher figures due to form and fixtures.
6 years, 7 months agoThanks for this.
Think this really highlights the mentality of managers.
Covering offers security, so arguably is for the dullards.
But tripling and doubling is for the mavericks - all eggs in one basket and all risk and reward. Eg Aguero and Jesus or triple Saints defence -although this article covers attacking roles only.
Interestingly my most successful seasons were when I doubled in attack - Sanchez and Sturridge, Ronaldo and Rooney, Lampard and Drogba.
Currently I have Eriksen and Kane and Firmino and Salah so am continuing that.....trouble is as the article says - that can either do really well or really poorly.