TL:DR?
Changes in fantasy performance over the season correlates with early season xG+xA over/under-performance
Trade out Pogba and Doucoure
Trade in Lukaku and Kane
Sitting pretty at 7 and oh, expecting the season to be a walk in the park? Or drowning your sorrows at 1 and 6, hoping that your fantasy fate will take a considerable turn in the positive direction? Whatever your position, how much does the first stanza of the Premier League season foreshadow what’s to come? With zero football being played until October 16th, I thought I’d take a look at data from the 2020-21 season and find out.
The Analysis
Points per 90 (PP90) was calculated for all players that: a) played a minimum of 100 Premier League minutes from gameweeks 1 to 7, and b) played a minimum of 300 Premier League minutes from gameweeks 8 to 38. These values can be seen in the graph below. The difference between these two periods was then calculated, with positive numbers indicating a player improved their fantasy performance (in terms of PP90) from gameweek 8 onwards, and negative numbers indicating that they declined.
Not surprisingly, PP90’s from gameweek’s 1-7 correlate well with PP90’s from gameweek’s 8-38. But the relationship is far from strong. We see numerous examples of players improving drastically, and multiple instances of players falling considerably after their initial strong starts. The top 20’s for each are shown below (PP90 decliners in red and PP90 improvers in green).
There’s a chance that some players do have a tendency to start a season slowly. Or conversely, make a flying start before running out of steam. More likely, though, is that this data reflects strength of schedule, outlier performances, general team improvement/declines, and randomness. So whilst this is no doubt interesting (at least to geeks like me anyway), is there anything we can actually take from it that will help us in our quest for fantasy glory? Well, I’m glad you asked.
The Curious Case of Son
The sight of Heung-Min Son’s name at the top of the fallers list triggered memories of Twitter discussions about the South Korean superstar this time last year. He was in the middle of a purple patch that even Marcos Alonso would have been proud of. 8 goals and 2 assists in the opening 7 gameweeks had given Son 148.5 fantasy points – a tally bettered only by his teammate Harry Kane who had scored 6 and assisted 8 himself (and who also happens to sit 4th on this list!). This Messi-like production wasn't what generated the Twitter debates though. It was the expected numbers behind them. Son’s 8 goals had an expected value of just 3.09, whilst he had also overperformed on the 2 assists, albeit only marginally. The best players do tend to outperform xG (Messi has done so by an average of 8.4 goals per season). But Son’s current rate, at least according to the xG advocates on Twitter, was unsustainable.
Based on comparative fantasy performance, if you had tried to trade out Son after 7 gameweeks last season you might have been able to get absolutely anyone (bar probably Kane). KDB, Bruno, and Salah may have been slightly less likely, just purely based on reputation, but I’m sure some owners would have seen Son’s much superior FPts and FP/G and probably agreed to it.
Son’s performance post- gameweek 7 hardly fell off a cliff. In fact, he ended up the 6th highest scoring player on the game (albeit with “only” the 12th highest WAR and 16th highest FP/G*). But his value at that time was close to it’s peak, and arguably the most important skill (or fortune!) required in fantasy football is knowing when to jump off or on a bandwagon. Was Son’s xG+xA a crystal ball into his fantasy value for the remainder of the season? Let’s take a look at some more data…
*min 1,300 minutes played
Change in PP90 and xG+xA
The below graph shows the relationship between a player’s change in PP90 (from GW’s 1-7 to GW’s 8-38) and their xG+xA performance post- gameweek 7. The strength of the correlation is similar to the previous graph. Of the top 20 players who were over-performing their xG+xA, only one ended up improving their PP90 (Marcus Rashford, whose PP90 went from 11.17 to 13.28, despite at the time, underperforming by 2.05 xG+xA). At the other end of the scale the assertions are a little messier due to the floor effect of underperforming xG+xA after just 7 games, nevertheless, 3 of the 4 players who were significantly underperforming ended up improving their PP90 in the remainder of the season. This included Trent Alexander-Arnold, who went from a PP90 of 6.01 to 13.29, and Tomas Soucek who went from 7.79 to 11.59.
Note: xG+xA numbers in this instance refer to how many goals and assists players HAVE had, compared to what we would have expected. So a +5.00 means that they have scored or assisted 5 more goals than we would expect, whilst a -2.00 means that they have scored or assisted 2 fewer goals that we would expect.
Some percentages…
To perhaps put this into a more meaningful context, consider the following statistics based on this data:
89% of players who were over-performing their xG+xA by at least 1.0 goal after gameweek 7 saw a decrease in their PP90 for the remainder of the season. The average change in PP90 was -4.83.
When expanding this to those players who were over-performing their xG+xA by at least 0.7 goals after gameweek 7, it was still a massive 83% of players who saw their PP90 fall.
The top 11 over-performers all saw their PP90 decrease post- gameweek 7, with the average change in PP90 of these 11 players being -6.42.
Looking at this from the perspective of PP90 changes, of the 34 players who saw a PP90 drop of more than 4 points, just 27 (79%) had been overperforming their xG+xA. Of the 29 players who saw a PP90 increase of more than 4 points, just 25 (86%) had been underperforming their xG+xA.
The top 5 PP90 fallers had all been overperforming, and only 1 of the top 12 PP90 fallers had been underperforming (Mohamed Salah, who’s xG+Xa was 0.09 higher than his actual 7 goals and 0 assists).
Finally, ZERO of the top 17 PP90 increasers had been overperforming their xG+xA after gameweek 7, and only 1 of the top 38 PP90 increasers had been overperforming by more than 0.6 xG+xA (Aaron Wan-Bissaka).
Just tell me what this all means!
Okay, okay – enough with analysing last year. Let’s look at how we apply this knowledge to this season. It’s simple enough: we exploit the expected numbers. There are a number of sources for this – FBRef probably being the best – but because I used Understat for the initial analysis (they allow you to easily split the data by date), we will continue with their numbers here.
Below are all the players who have played a minimum of 100 minutes in the Premier League this season and are currently either overperforming or underperforming their xG+xA by at least 1.5 goals.
Overperformers
Whilst there are no players currently reaching Son’s astronomical levels of overperformance from last year, Paul Pogba and Abdoulaye Doucoure have been producing far more in front of goal than we would expect based on their actions on the pitch. If we compare them to last season, they would sit 4th and 5th on the list and it is notable that the players that would have been either side of them (John McGinn, xG+xA overperformance of 3.37, and Danny Ings, xG+xA overperformance of 4.13) both saw their PP90s drop by between 4.35 and 4.53 points. For Pogba, this is not so much of an issue. His PP90 would still be a very respectable 13.1. But for Doucoure, a ~4.4 point decrease would leave him at 7.8. That’s in the Jordan Henderson-James McCarthy range. Historically both players have tended to overperform (Pogba more so), but not to this level. If you roster them, the smart move would be to trade them out now.
Jamie Vardy, Danny Ings, and Gabriel Jesus are also candidates for this, though Ings may warrant a bit more thought for owners. For the past two seasons he has far exceeded his xG+xA – both times to a much greater extent than the 2.65 which he is currently at – so maybe his is sustainable. Names like Demarai Gray and Andros Townsend have maybe the biggest red flags over them. These two are doing things that nobody predicted at the start of the season. And there’s a reason for that. Can they keep it up? Of course. Would I put money on it? No chance. Right now Townsend is the fourth highest scoring midfielder in the game – there will be no better time to cash in.
Underperformers
The list of underperformers includes some big, big names and some of them (Lukaku, Traore, Mane) have still been putting up decent fantasy points. Whilst it doesn’t seem that the relationship between underperforming xG+xA and PP90 change is as robust as the relationship between overperforming xG+xA and PP90 change, I still think the data makes a strong and logical case for trying to trade in, well, pretty much all of the list on the right.
The Full-Time Whistle
Wherever you are in your fantasy league right now, there's a helluva long way to go. Yes, the first 7 gameweeks give us an idea of what to expect going forwards, but it's more of a ballpark prophecy than an exact foreshadowing. Lots can, and will, change. And one of the clues in predicting those changes comes in the form of expected goals and assists. Don't ignore them. Use them to your advantage. Find the players whose fantasy value is about to rise, and avoid the ones who are on the verge of decline.
For all the latest from The Inner Geek, follow @the_innergeek on Twitter!
And for more in-depth and exclusive resources, become a member of The Inner Circle.
