- Better Haul Salah – How To Shake Up Your FPL Decision Making - 17 Aug 2022
- FPL Value: The £8m Midfielder Question - 4 Aug 2022
- FPL Value: Which Forwards Are Worth Their Weight In Goals? - 29 Jul 2022
The new season is barely two weeks old and there is already an arm-twisting difference of opinion breaking out in FPL that rivals Conté versus Tuchel as a flare up between members of the same guild strongly believing borderline decisions are relying too much on imprecise evidence. Thankfully no content creators have been banned as a result and handshakes look likely at the next fixture, but the rise to prominence of statistical modelling for Fantasy Premier League is steadily eroding faith in the eye test as the deadline decision tie-breaker.
I value the tactical and observational insight I glean from podcasts like Black Box, Planet FPL, FPL Wire and the Scoutcast, but I also like to pore over the predictive statistics you can find in places like FPL Review and the Jumpers for Goalposts site itself as well as an increasing variety of ‘homebrew’ concoctions that emphasise one ingredient or another and often appear in Twitter threads 24 hours before a Gameweek deadline.
I have a major source of frustration with such models though. The ones I’m familiar with generally predict Mohamed Salah to get between 6.5 and 8.9 points every single Gameweek. FPL is scored in whole numbers and most of the Liverpool winger’s predictions would see him relentlessly awarded a rounded score of 7 with occasional wild fluctuations to 6 or 8.
What none of the models ever predicts (for a standard Gamweek) is any player scoring 10 or more FPL points. Or -2. And yet we know it happens every Gameweek. Research I did last year showed that in the 23 ‘standard’ Gameweeks of the 2020/21 season an average of 12 players scored 10+ points each time and 3 more went into minus numbers. Salah got between 1 and 3 FPL points eight times last season, registered a 0 (Gameweek 20) and reached double figures on ten occasions including a 19, a 24 and a 28. I can virtually guarantee that no published model predicted any of those scores. And yet it’s become almost received wisdom that ‘going against the models is folly’.
This led me to an online discussion with The FPL Kiwi, a fellow Jumpers for Goalposts contributor and doyen of statistical approaches to FPL who put forward several defences for the models. Kiwi pointed out that they should be really taken much longer term and what they were predominantly presenting was average outcomes. I acccept that in general, but they then can’t resist the temptation to go on and predict that Salah will get 6.9 points this specific week but 7.4 points next week, despite the fact that, with the rates of variance on show, the difference between the two doesn’t mean anything and FPL doesn’t score in tenths of points. Presumably they do it because that’s what FPL managers want to see as they’re making decisions week by week rather than once a month.
Kiwi then played a trump card:
And taking the longest, most OR-conscious view possible he definitely has a point. I was still uneasy, though:
Kiwi was unimpressed:
I read this, smiled and slowly typed back:
VERY MUCH ON THE CARDS
After a little more thought I realised that what I really wanted was dice and cards. I’ve said in articles before that I’m a big fan of the sporting ‘replay‘ games created by Terry Goodchild. Some of his best efforts rely heavily on rolling two ordinary six-sided dice (2d6) and looking for a result against some kind of chart or a card representing a player. There is a 2.77% chance of rolling a 2 (or a 12), a 5.55% chance of rolling a 3 (or an 11) and so on.
My pre-season articles on midfielders and forwards were based on the percentage of Gameweek appearances last season where players registered a Return (5 to 9 FPL points), a Haul (10 or more FPL points) or a Blank (4 or fewer FPL points). For each FPL asset I already had the percentage chance of them getting a Blank, a Return or a Haul in any random Gameweek last season where they actually played (I leave the injuries stuff to Ben Dinnery…). If I simply translated those percentages into chances of rolling certain numbers on 2d6 I could make individual player cards, like this one for Mohamed Salah:
Suddenly we’re a world away from “6.5, 7.3, 7.2, 6.8…”. Most FPL managers will be familiar with rolling 2d6 in boardgames and will have a natural feel for what the chances of rolling a Haul, a Return or a Blank are based on that simple player card. And the card is based on a whole season’s worth of statistics for a star player continuing on the same team which hasn’t changed much. A card made for Salah at the end of the coming season will probably have most of the results in the same place.
But how does the card for Salah compare to those of some of his rival FPL assets?
You have to look quite hard for differences between the cards for Salah and Son Heung-min. Roll thirty times each against those and you will probably get very similar results. Kevin De Bruyne’s 2021/22 card is still pretty good, but you can see at a glance that the other two are better both at avoiding a blank and grabbing a Haul. Over a short period, however, you might not notice much difference in outcomes.
Harry Kane is somebody else who is regularly in the captaincy discussion alongside this trio of midfielders, so his card probably looks very similar, right?
The chance of a Harry Haul is certainly competitive, but the rest of his card makes clear the stark situation if you don’t get that outcome. A lone (if prominent) Return result and Blanks here, there and everywhere. A Kane captainer might respond by listing a number of favourable circumstantial factors that show it’s definitely Kane-over-Son for the next round, but we’ve already seen this season that Harry can blank when Tottenham score four at home.
It’s too early to make a card for him yet, but we’ve also seen that Erling Haaland can be reduced to a mere Return (from one of his only two passes) even when Manchester City are cruising past Bournemouth and he plays most of the game. Variance gets everywhere. And on these cards, it’s built-in and front-and-centre.
PARTY OF ADVENTURERS
The difference between the elite captain options and the riskier but still plausible ones becomes clear when you make cards for popular FPL picks like James Ward-Prowse, Ivan Toney and Riyad Mahrez:
Overall, the chance of a precious Haul goes right down. There’s a decent possibility of a Return, but more often than not the outcome is going to be a Blank. Furthermore, the first 6 results on each card are the same. Toney’s card is the same as that for Ward-Prowse but for the position of the Haul and the same as that for Mahrez but for the Return instead of a Haul at 11.
BACK TO FRONT
If you’re looking for safety in captaincy and want an alternative to premium attackers then the new breed of marauding full-back could provide just what you’re looking for:
Stand back and admire the safety (usually) on show from Anfield’s finest! Admittedly, so far this season Jurgen Klopp has been rolling far too many 6s for Trent Alexander-Arnold and Andy Robertson. When you look at these cards (based on last season when they were the top performers at their position) you can still see that a blank or two is entirely possible in the grand scheme of things. But, unless there has been a fundamental change to the general fortunes of Liverpool FC over the summer, regression to the mean dictates that the 7s will start tumbling in and, in Robertson’s case especially, almost any roll gets you something rather than nothing.
Of course, you can’t have an FPL squad defence entirely composed of Liverpool players (outside of Draft leagues), so what does the landscape look like for Joao Cancelo and Reece James?
Thomas Tuchel has also been rolling 6s, but that works out just fine for Reece James. He’s actually slightly more likely to rack up double-figures than Robertson, but the dropoff to Blanks is a lot steeper. The Blanks on Cancelo’s card could be spread around a bit more, but their placement emphasises that, even in Manchester City’s title-winning defence, Blanks can be harder to avoid than you might think. Similarly, the drift of Hauls from the centre show that, although Cancelo does get them quite often, he’s only about as likely to get one as the non-KDB City midfielder of your choice.
And while we’re here, that Reece James card reminds me of something……
So there you have it. Reece James is basically Harry Kane except for a Return at 3 rather than a Haul and significantly less robust leg muscles.
Somebody who didn’t get a Haul all last season is City’s goalkeeper Ederson. Somebody who did is his Liverpool counterpart and Brazilian squad-mate Alisson:
Plenty of nice, steady returns on display here, too, but given that these two and Hugo Lloris are the statistical leaders amongst FPL custodians you can see why goalkeepers don’t get the armband very often.
But what if I want to generate actual FPL scores, not just Blank, Return or Haul?
Let’s be clear. I am not, repeat NOT, suggesting that anybody generates enough cards and rolls enough 2d6 to recreate a sort of artifical version of the 2021/22 FPL season. Well, I might if this unceasing sunshine is suddenly replaced by endless wind and rain, but even then… Let’s always keep in mind that these cards are not simulating actual football (like Subbuteo or Logacta Chart Soccer). They are simulating an online game that is based on actual football.
The real value in these cards for FPL managers is that they give you a readily understandable idea of the magnitude of difference (or not) between various prominent FPL assets and their likely range of outcomes for the purposes of helping with expensive transfers or captaincy decisions.
Nevertheless, if it would feel more tangible to have some actual numbers to look at (or if you’re sitting in a holiday caravan with no signal and think it might be fun to generate enough cards for a draft FPL knockout tournament because everybody is already fed up of dominoes) then my article researching the statistics for a standard FPL Gameweek gives us what we need to make cards that can give us simulated FPL scores.
Returns is the easiest of the three:
There are only five possible outcomes and lots of data points to generate a card guaranteed to give a realistic range of Returns, the classic “Centre Back’s Six” being the most likely. Blanks and Hauls are more tricky because you can get MINUSes or MEGA Hauls, so you really need a table and <gulp> a sub-table for each:
Rolled a 2 on the main table? Then roll again on the MINUS table! Simulate an own-goal contentiously awarded after tedious deliberation by a remote panel or a new star striker who misses big chances before decking somebody!
Jan Bednarek’s -7 was so statistically remote that it is doomed to remain forever beyond the reach of mere dice-rolling. Or you could toss a coin (best of three) if you roll a 12…
Meanhile, the same two-table approach will do for Hauls:
I realise only two readers made it down to here, but thank you to both of you and happy gaming. 🙂
And if anybody would like a card generating for a specific FPL player based on LAST SEASON’s data (so not Haaland, Perisic, etc) drop me a line on Twitter or in the Comments below and I will do my best to oblige.