Specious Theoretic Chatter (The free-for-all shared-RMT thread 19/20)

[Stemania]

Vardy, Abraham, Jimenez defo seems to have developed into the big-point-getting striker trio of choice in the last couple of weeks. Ironic(/dodgy?) that I've been spending this period of the season transitioning to 451 (and will complete the transition in the upcoming week via Pukki to TAA).

No short term plans to get Vardy, though may become more temped by Leicester for the (WAT, avl, NOR) run. So, finally get to the squad structure I want (451), before ripping it all apart again. I suppose the aim is to dismantle the Liverpool trio in GW18 for a couple of gameweeks (-blank-, lei), before wildcarding them back in again for GW20.

[Stu255]

No short term plans to get Vardy, though may become more temped by Leicester for the (WAT, avl, NOR) run. So, finally get to the squad structure I want (451), before ripping it all apart again. I suppose the aim is to dismantle the Liverpool trio in GW18 for a couple of gameweeks (-blank-, lei), before wildcarding them back in again for GW20.

I was thinking of shipping one Liv and benching the other two for GW18. So many options around base price that you can get a decent bench without allocating much money there.

I don't think GW18 will be very high scoring as Liverpool are missing and it also includes... MCI v LEI and TOT v CHE.

I think a lot of managers will completely mangle their teams trying to move 3 Liverpool in and out either side of GW18, so I see the greater potential for gaining rankings in GW16-17 and GW19-20.

[Archy]

It’s good still having the wildcard up the sleeve to help deal with the forthcoming blank. My immediate concern however is what to do with Man City assets now their fixtures are turning sour. Liverpool’s, by contrast, look outstanding.

So there’s Sterling Mane as a tempting straight switch for the next 5 weeks

Or there’s Sterling/KDB Leicester/Chelsea Mid
and
Rico Robbo/TAA (Or another move to spend the cash if you already have both these)

to consider to take best advantage of the fixtures. Thoughts?

[Zimmerman]

Surely likes of Sterling and de Bruyne are fixture proof?

They come in to their own against better opposition and less chance of rotation (Aguero coming in to the mix too).

[Sutter Kane]

Surely likes of Sterling and de Bruyne are fixture proof?

They come in to their own against better opposition and less chance of rotation (Aguero coming in to the mix too).

Yes this was my exact thought process too. Chelsea, Newcastle, Burnley, Man U and Arsenal is on paper, not an easy run. But is it difficult for a side like Man City, now desperate to close the points gap? I can see Man C conceding a few goals in those games, which does make them a bit tricky, but they can score plenty against all of them. I suppose the question is, and this pertains to Sterling in particular, because of those fixtures, lack of captaincy makes their value diminish so I can understand people getting rid because it's difficult to get some of the current hot properties without removing Man C attackers.

[Gambit]

Surely likes of Sterling and de Bruyne are fixture proof?

They come in to their own against better opposition and less chance of rotation (Aguero coming in to the mix too).

yes but for me it's currently about what you can get with your money elsewhere.

Sterling and KDB are what? £22+m depending on buy/sell price etc, for around the same money people are looking at Salah and Pulisic or Mane and Son, and given the fixtures it's easy to make a case for going without one or both for the next month.

Or maybe "downgrading" a City mid to a high performing mid price and then using the money to bring in TAA/Robertson or one of the Leicester full backs, or both.

It's not a case of thinking they will flop, but more a case of there are potentially better options - especially for the C - with some of the other players.

Looking at the coming fixtures for Liverpool/Chelsea/Leicester and for me their key picks have a great chance of beating the premium City picks over the next month or so, and it's hard to see any GW's where I'd really want to captain a City player, the next GW v Cheslea is probably the only one for quite a while.

[blahblah]

Citeh need to score to stay anything like close to Lpool...

Lpool could well be more conservative ie defensive due to the congestion.

Leicester I expect to continue as they are.

Chelsea are my unpredictables due to their ages and manager. But I'm happy having Pullasicky and Tammy.

[Bobby Fetta]

Aguero's chart looks a bit odd? He started this season with goals in every game (scored 8 goals in the first 6 games, scoring in all 6) and then went quite, whereas his chart here suggest the opposite??

The thing about a 10 week average is that it moves as much with what drops out the other end as the new data going in. So at the start of the season Aguero's numbers were pulled down by his lean spell at the end of last season. Most of the goals from the start of the season are still in his 10 week average now but they will fall out over the next few weeks. Unless he scores well, his average will fall.

[Stu255]

Aguero's chart looks a bit odd? He started this season with goals in every game (scored 8 goals in the first 6 games, scoring in all 6) and then went quite, whereas his chart here suggest the opposite??

The thing about a 10 week average is that it moves as much with what drops out the other end as the new data going in. So at the start of the season Aguero's numbers were pulled down by his lean spell at the end of last season. Most of the goals from the start of the season are still in his 10 week average now but they will fall out over the next few weeks. Unless he scores well, his average will fall.

How about a diminishing weighting?

When doing the average use 10 samples of the last game, 9 samples of the previous game, 8 samples of the one before that... right down to 1 sample of the game 10 weeks ago.

This way the tail doesn’t wag the dog and any anomalous results fade with time, rather than flatter the chart for a full 10 weeks before reminding you with a sudden cliff edge.

P.S. I think your charts are super useful as they are.

[Bobby Fetta]

How about a diminishing weighting?

When doing the average use 10 samples of the last game, 9 samples of the previous game, 8 samples of the one before that... right down to 1 sample of the game 10 weeks ago.

This way the tail doesn’t wag the dog and any anomalous results fade with time, rather than flatter the chart for a full 10 weeks before reminding you with a sudden cliff edge.

P.S. I think your charts are super useful as they are.

Yes, I did consider that and can see the merit. The advantage of an equally weighted 10 week average is that it smooths out fixture difficulty. But 10 weeks is a long time... Another way round the problem is to use the median rather than the mean as the average but I'm not sure that completely works either as you lose any representation of the really high scores. All in all, I think don't worry too much about the precise numbers. It should only be a general guide.

Glad you find them useful. Gives me something to play with whilst my FPL team slides further and further down the rankings.

[Bobby Fetta]

Fascinating stuff, thanks Bobby. Would love to see the Lundstrum graph!

I changed my mind and decided to include Lundstram in the midfield dataset for fun. Here is his chart. On total xG over the last 10 gws, he slots into 14th place - just ahead of De Bruyne, McGinn and Maddison

[Spreadsheet]

These are the sorts of posts and discussions I come here for. I absorb it all and vow to utilise terms like models, stats, projections and trends in my transfers.


Then, when I analyse the data of my own behaviour, I find nothing but words like punt, hunch, differential and next year.

[raoul]

Don't look now but there's a Raheem kerfuffle of some sort kicking off...

[TheRumourMill]

Fascinating stuff, thanks Bobby. Would love to see the Lundstrum graph!

I changed my mind and decided to include Lundstram in the midfield dataset for fun. Here is his chart. On total xG over the last 10 gws, he slots into 14th place - just ahead of De Bruyne, McGinn and Maddison

Thanks Bobby! I knew he was good but the graph suggests he may well be getting even better, and with fantastic attacking fixtures to come too! First name on my teamsheet now. Just gutted he hit the post at the weekend, I had him starting and it would have been a nice reward for that faith.

And yeah Sterling dropped for the next England game due to a fight with Gomez in the canteen! Not sure if this is good or bad for him in fpl. One one hand he gets a rest and shows the passion he has to haul in Liverpool, in the other, this may well be the sort of behaviour that sees him pick up his 5th yellow card imminently. As an England fan, I'm disappointed. Although we should still have enough to get past Montenegro!

[hancockjr]

Does anyone else look at those graphs and wonder whether xG is actually a useful tool? By useful I mean a better predictor of future goals than actual goals? Has anyone done any analysis to prove or disprove it any way? Seems 100% over performance is not unusual over a sustained period.

[Sutter Kane]

Seems 100% over performance is not unusual over a sustained period.

Not sure about this. Playing the percentages, I think it is unustainable in all but the most freaky circumstances. (e.g. Yaya scoring 7 free kicks in a row or whatever it was) Also depends how long is meant by sustained period. I think XG is useful but has many limitations but combine it with other stuff such as eye-test then it's a decent predictor in my opinion. And we've seen it work before when a player has underperformed when compared to their underlyings, then they explode into points.

[math!]

Does anyone else look at those graphs and wonder whether xG is actually a useful tool? By useful I mean a better predictor of future goals than actual goals? Has anyone done any analysis to prove or disprove it any way? Seems 100% over performance is not unusual over a sustained period.

I think you should just use it as a flag before then doing the visual test on whichever players caught you eye.

[dino1980]

Does anyone else look at those graphs and wonder whether xG is actually a useful tool? By useful I mean a better predictor of future goals than actual goals? Has anyone done any analysis to prove or disprove it any way? Seems 100% over performance is not unusual over a sustained period.

On a player level, there’s a really long article here: https://fplreview.com/2019/09/how-accur ... potential/

I’ll be honest my eyes glazed over about halfway through it but here’s one of the key takeaways:
“Something that stands out quite quickly is the superiority of xG90 data to G90 data. 40 matches of xG data appears to provide a similiar confidence level as 100 matches of goal data.”

And on a team level I found these two pieces:

https://cartilagefreecaptain.sbnation.c ... cted-goals
http://11tegen11.net/2015/01/05/the-bes ... ted-goals/

[Stu255]

Does anyone else look at those graphs and wonder whether xG is actually a useful tool? By useful I mean a better predictor of future goals than actual goals? Has anyone done any analysis to prove or disprove it any way? Seems 100% over performance is not unusual over a sustained period.

What do you mean over a sustained period? This is a 10 week average so if someone scores a hattrick against Southampton that will boost their chart for the next 10 weeks. These charts cannot capture 1 week spikes.
Also xG is a measure of the opportunities a player gets to score a goal. Some players (typically strikers) are better at converting their opportunities than others.

If a player gets more opportunities this generally leads to more goals, but there are other factors too. But xG is as much a measure of the team as it is of individual player.
xG was originally designed for team v team analysis and xG charts work quite well to illustrate the run of play between two teams over 90 minutes.


But comparing individuals from team A to individuals from team B is a bit more nuanced.

[TheRumourMill]

Does anyone else look at those graphs and wonder whether xG is actually a useful tool? By useful I mean a better predictor of future goals than actual goals? Has anyone done any analysis to prove or disprove it any way? Seems 100% over performance is not unusual over a sustained period.

It hasn't been particularly useful this season, no. But that's because a lot of statistically odd or unpredictable events have happened which have resulted in a skew over the first 10 or so gameweeks of the season. Its why when you look at analysis of the top 250 managers or so, a hell of a lot of them are languishing compared to where the odds or xG says they "should" be.

Capture.JPG

10 games though isn't much of a sample size and I fully expect things to regress to the mean and calm down a bit over the remainder of the season. Its easy to lose faith in stats and models when stuff like this happens but I'm going to put it down to statistical noise and variance. That's what I'm comforting myself with down at 1.5 million rank anyway

[Stu255]

Thant tank distribution chart is interesting in that it clearly has 2 populations.

The larger majority population but also a smaller minority group who are doing better than the normal distribution would suggest.

What is the major difference there I wonder?

[hancockjr]

What is xG rank and odds rank?

[hancockjr]

Credits to Understat.com and OfficialFPL. GW12 Playerdata.

FPL - GW12 - PlayerData - Excel.rar

Hi,

Do you have this data in this form from last season?

[hancockjr]

On a player level, there’s a really long article here: https://fplreview.com/2019/09/how-accur ... potential/

They seem to do an awful lot with the data except prove it's actually useful: i.e. they don't use xG data to predict future actual goals and compare that to using past actual goals, and show that xG is a better predictor.

If xG is not a better predictor than actual goals then what is its actual use?

I'm pretty sure it should be but for all this talk of xG there's very little evidence backing it up.

I can't get positive results for xG, but I only have this season's data (Thanks FrankKessie) and there's a lot I can do to improve my model (it's only very rough draft at the moment). I did expect it to be a slam dunk though (as would showing that adjusting for fixture difficulty, both for the matches "model build" data and the matches the model is being tested on).

I do think this season has been "odd" though, which may explain why it's not easy to prove anything.

One thing that's bugging me is some of my best transfers (Vardy especially) have been the ones I was least convinced about, and certainly weren't supported by xG data.

[FranckKessie]

Credits to Understat.com and OfficialFPL. GW12 Playerdata.

FPL - GW12 - PlayerData - Excel.rar

Hi,

Do you have this data in this form from last season?

I am an idiot at making links historic posts, but I did upload this zip in the summer. I think it suits your needs.

1920 FPLAnalysis.rar

[math!]

Don't think you can use xG on Vardy. He's a national treasure! He will dry up when his teammates stop performing.

[Sutter Kane]

You just have to adjust it for him, factoring in a high conversion rate. Just not as high as it is right now.

[Finisher1]

Surely likes of Sterling and de Bruyne are fixture proof?

They come in to their own against better opposition and less chance of rotation (Aguero coming in to the mix too).

They would be truly fixture proof only if we could expect them to score similar (or better) points against good teams than against poor teams. I don't think that's the case, even if we factor in their higher chance of rotation against poor teams.

I don't think any player is fixture proof. The whole term "fixture proof" is a myth, until we finally find a player who consistently scores similar (or better) points against good teams than against poor teams.

Funnily enough (and take this comment with a pinch of salt), a certain Norwich striker is close to this definition as he has scored 10 PPG against Man City, Liverpool and Chelsea, but only 3.44 PPG against other teams

[Stu255]

I don't think any player is fixture proof. The whole term "fixture proof" is a myth, until we finally find a player who consistently scores similar (or better) points against good teams than against poor teams.

**MYTHBUSTERS**

From the small sample of 19/20 data...

KDB is 6.9 ppg and 6.5 ppg v's top6.
Sterling is 6.1 ppg and 4.0 ppg v's top6.

Leicester famously raise their game v's big sides, and this does bear out in the data...
From 18/19...

Maddison 3.8 ppg and 3.9 ppg v's top6.
Vardy 5.1 ppg and 5.8 ppg v's top6.

This "fixture proofness" repeats for Leicester every season for which there is data.

Aguero is a 100% starter v's top6 sides going all the way back to 17/18.
18/19 he was... 5.9 ppg and 6.7 ppg v's top6. He was the same goal threat in both but has higher xA and more assists v's top6. Although that doesn't repeat this season, with KDB back in the side.

[math!]

Anfield is Aguero proof however.