Last gameweek was another frustrating one for my spreadsheets, even though predictions performed favourably in comparison with other better known predictive models and algorithms. My spreadsheet proved more accurate than fivethirtyeight.com for both match result predictions (see appendix I) and correct score forecasts (see appendix II). It also performed better than “the world’s most powerful predictive fantasy football algorithm” with regards to the number of goals predicted, albeit only matching bookies best odds in the process (see appendix II). Once again though, they outperformed bookies best odds for clean sheet probabilities, even with the effects of their margins removed by FPL Review (see appendix III). These positive results were in spite of giving other models a big headstart by going out on a limb in fancying CHE so strongly to beat TOT.
The frustration arises from the fact that my spreadsheets’ superiority over bookies odds and other models is still only marginal so far this season, and frankly not yet justifying the time and energy I expend on them. The main causes of instability in my spreadsheet’s predictions continue to be the dizzying rate at which penalty kicks are being awarded, and the overly optimistic, best guess weightings that the promoted sides started the season with. These wrinkles are being incrementally ironed out though, and my spreadsheets are still counting down to the lift off expected after sixteen gameweeks of season data have been collected.
Onwards and upwards then, five teams are deemed more likely than not to score at least 2 goals in GW11.
All five are expected to win their matches this weekend, albeit just by the odd goal based on the most likely number of goals expected to be scored by each team.
I wrote last week about how reductive presenting a single correct score forecast was though, and how much more meaningful it would be to present scoreline predictions in a matrix. For example, the table above shows LIV 2 WOL 1, but the 9.1% probability assigned by my spreadsheet is not the highest, as the probability for 2 – 0 is reckoned to be 0.7% higher (see table below).
This is a much better way to visualise the likeliest scorelines of matches, in my opinion. Note that the probability of the Reds winning 3 – 0 is higher than doing so 1 – 0, and winning 3 – 1 more likely than drawing 1 – 1.
The fly in the ointment here though is that these probabilities are based on Liverpool’s last 8 home games, and Wolves’ last 8 away games, but Nuno Espirito Santo evidently took on board the part in my blog last week when I highlighted his team’s away attack strength weighting was worse-than-average and trending in the wrong direction. Here’s how it looked prior to their GW10 trip to the Emirates Stadium.
So, what did Nuno do for the first time in a away match during his time managing Wolves? He played with a back four! For the second match in a row actually, following a switch prompted by Coady‘s absence in their previous game. The ever-evolving nature of the sport is one of the things that makes life so difficult for predictive models in football. Point being here is another example of why we should always be wary of taking my spreadsheet’s predictions at face value.
All of which needs to be borne in mind when we see Salah at the top my players predicted points for GW11 (see table below), because we don’t yet have enough data to assess whether his prospects will be enhanced or diminished if WOL continue to forego their trademark back five. Also remaining to be seen, is the effect Jota has on the Egyptian’s attacking output and minutes.
The table below shows how many points players are expected to get on average (excluding bonus points, and yellow/red cards). The red ‘P’s indicate where predicted scores have been inflated by the effects of recent penalty kicks, and the numbers in the column to the right of expected points show what these scores would have been if the distorting effects of penalties are removed. In other words, these are the average expected points assuming there are no opportunities to score from the penalty spot.
KDB, Fernandes, Salah and Vardy were the top ranked attacking players in this table last week, and are all in the top half dozen this week (see table above). They were predicted to score 8.5, 8.1, 6.7, and 6.1 points on average, before bonus, and scored 9, 10, 6, and 5 points respectively, making for a very respectable mean absolute error of only 1.05 points for that quartet.
Salah and Vardy have a higher points prediction this weekend, while Fernandes‘ projected score remains the same. Surprisingly, the average expected points for KDB is over 2 points lower vs FUL than it was last week vs BUR, and both Jota and Ziyech are ranked higher than the City playmaker.
Kane returns to my players points table after missing the cut for the first time this season last week, which was vindicated by him blanking for the first time since GW1. Assuming his yellow flag is of the Mourinho mischief-making variety, he looks a good bet to maintain his good record in North London Derbies.
With so many players to choose the best captain from again, I will turn to My Stats Tables in Fantasy Football Scout’s Members Area in hopes of narrowing down our options. Of the 9 teams with representation in my Players Points table above, the only 3 not in the top half of my Team Offence table are unsurprisingly CRY, WBA, and BUR, which should probably put paid to any interest in Zaha, Pereira, and Wood as differential captains.
That said, all 3 will be facing defences among the 7 worst in the league over the past 6 gameweeks (see table below). The case for Wood rests on him facing arguably the worst defence in the league currently in the form of EVE.
Significantly, the only captaincy options not facing weak defences in GW11 are those drawn from LIV and CHE, albeit the latter’s opponents (LEE) conceded only 0.22 less non-penalty xG than tenth worst WHU. Discouragingly, for those thinking of giving Salah, Jota, or Mane the armband, WOL have the fifth best defence by this metric, as well as joint-sixth best going by Big Chances conceded.
Vardy was a popular captain pick last weekend, and encouragingly, for those willing to back him again, he now faces the second worst defence by this metric. His captaincy claims are strengthened further by topping my Players xG table sorted by Minutes Per Shot On Target, which features only 4 players after removing Wilson and Watkins. Salah is the other notable inclusion in the table below.
Both Fernandes and KDB retain their position in my expected assists table, with the MUN playmaker outscoring his MCI counterpart on every metric in the table below.
As stated last week, the one area where KDB might have the upper hand is with set plays, especially as Telles has taken 9 corners in his 2 appearances to date. Ziyech marks his debut appearance in the table below with a chart topping position.
Through the process above I have whittled down the 19 players who made the cut in my predicted points table down to a shortlist of 5. These are Salah, Bruno, Vardy, Ziyech and KDB.
A key finding of my review of last season’s rolling data was that LEI were a much better team on the road, and that pattern looks to have continued this season. Monday night’s disappointing showing vs FUL notwithstanding, Vardy looks the best armband option to me in GW11.
Scout is the only fantasy football service I’ve ever paid for over the past seven seasons, and I’ve always believed their membership subscriptions offer great value for money. Anyone considering joining can find out more here.
A season-high 6 defenders appeared in my players points table last week (Chilwell, Masuaku, Furlong, Bartley, PVA and James), with only the WHU and CRY defenders blanking. The WBA pair and Chilwell are joined this time around by yellow flagged Castagne and Robertson.
Bartley, Furlong and Castagne owe their place in my players points table primarily to underlying goal involvement stats, whereas Chilwell and Robertson can back those up with good clean sheet prospects also (see table below). MCI, MUN and BUR are the only other teams deemed more likely to concede zero goals than (exactly) one. No team is reckoned more likely than not to keep a clean sheet in GW11.
My doubling up on MCI defenders ahead of a much heralded run of 6 good clean sheet chances in a row was all well and good, but was counterfeited by Pep choosing the worst moment to rest Cancelo, and by Mendy scoring his first goal for City whilst playing his best game in years!
Looking ahead, LEE catch the eye with 3 good clean sheet chances in a row (GW13–15), while the Martinez owners who hold onto him despite AVL‘s blank this weekend look likely to gain compensation in back to back fixtures (GW13 & 14). Tellingly, AVL are ranked the fifth best for clean sheets over the next 6 gameweeks despite playing one game fewer.
Cancelo played the full 90 minutes vs Porto midweek, and Mendy played well last weekend, so there has to be a question mark as to him starting vs FUL on Saturday, which is exasperating because he is one of only two players who qualify for inclusion in my Defenders expected Goal Involvement table (see below). PVA is the other, and his owners have grounds for optimism going into GW11.
Compounding the grievance felt by Cancelo owners is the fact that he is also one of only two defenders to feature among the players expected to score above 20 points in the next 4 gameweeks (see table below).
SOU were projected last week to be the third highest scoring team in the 3 gameweeks between GW11 & 13, and I touted a possible move for Walcott, but he failed his audition against MUN, and LEI have edged SOU out now anyway as predicted third best scorers over the next 3 gameweeks. Given the strong case made for captaining the Foxes’ talisman, I plan to transfer Werner and Podence out for Vardy and Jota instead.
That’s all for now, folks. May the GW11 flop be with you.
Coley a.k.a. FPL P0ker PlAyerFollow @barCOLEYna
Appendix I – mean absolute error of result probabilities
Appendix II – mean absolute error of predicted goals
Appendix III – mean absolute error of clean sheet probabilities