We check in with the Opta data next and roll out our usual post-Gameweek assessment of the weekend’s standout performers.
Huddersfield Town’s target man and a certain Liverpool winger both up their attacking potential ahead of a reduced Gameweek 31.
Meanwhile, big-hitters from the Leicester, Spurs and Arsenal midfields stake a claim for our upcoming Wildcards, while we also check out the impact of Chelsea’s new frontman at Stamford Bridge.
Looking at the expected value of Butland's score this week:
- 2 expected points for playing.
- 35% chance of a clean sheet based on betting odds and 4 points for a clean sheet, so 1.4 expected points from a clean sheet.
- 30% chance of conceding 2 or more based on betting odds and negative 1 points for conceding two, so minus 0.3 expected points from a clean sheet.
- Averages roughly 0.5 bonus point per game (and even more so recently) over the course of the season, so 0.5 expected point from bonus points.
- Averages roughly 1 points from saves over the season, so 1 expected for for saves.
So his expected score = 2+ 1.4 - 0.3 + 0.5 + 1 = 4.6
This indicates that he's worth a hit. Have I made any error in any of these assumptions?
all that effort to gain 0.6
that's how you play fpl
Haha took 2 minutes and it may be more than 0.6 if you apply it to other players.
I also didn't take into account YCs, RCs or injuries so it is slightly biased upwards (even though all these things are unlikely for goalkeepers).
I also didn't take into account penalty saves so could even be biased downwards.
A Baines penalty miss with a Butland save would be sweet.
Impressive work ! Would you mind doing that for all the players please 🙂 i still wouldn't get him
Not for a hit anyway
Well if you trust my methodology it is worth a hit in the long-run since 4.6 is greater than 4.0.
Minus the clean sheet and recalculate...
What does that mean?
"35% chance of a clean sheet based on betting odds and 4 points for a clean sheet, so 1.4 expected points from a clean sheet"
- I don't understand this calculation. He either gets a clean sheet or he doesn't. So it's either 0 or 4 points. So how is adding 1.4 points to his score helping?
Even assuming your calculation is valid, 4.6 means he either gets 4 or 5. If he gets 4, you gain nothing. If he gets 5, you gain 1 point.
So you're willing to risk losing 2-3 points just to gain 1 point?
That is just an expected value, which the probability of getting a certain outcome multiplied by the payoff of getting that outcome. It is effectively the same as calculating an average score but it attaches a different 'weight' to each observation whereas an averages attaches the same 'weight' to each observation (since in an average calculation you are simply dividing by the number of observations).
The probability of a clean sheet is 0.35, and the payoff of a clean sheet is 4. So 0.35 multiplied by 4 is 1.4. Look up more about what an expected value is if you're still confused.
4.6 is just what he'll get if we played or simulated this gameweek 1000 times - it is just represents the what he'd get on average in the long-run. Obviously in any individual simulation he could get 0, 2, 6, 11 or any other score. This just represents an average.
So we're not risking 2-3 points to just gain 1 point! The gain could be 10 points and of course the loss could be 2-3 points, it's just that on average, after simulating this gameweek 1000 times, we would expect to gain 0.6 points from it. If you have a move that is going to get you points in the long-run, why would you not do it?
The beauty of this approach is that - provided all your assumptions are correct - even if you move loses you points you actually made the correct decision!
We can't be results-orientated, we have to make the best decision based on the information that is available to us at the time, without the benefit of hindsight.
Great explanation, and possibly the best definition of the FPL game right there in your last paragraph.
Yes, it is something that is often said in poker and applies equally in FPL.
You may go all in with pockets Aces before the flop and lose the hand, but this doesn't mean it was the incorrect decision to do so. In the long-run you will make money from doing so, you just got unlucky in that one particular situation.
I've been thinking a lot about poker theory this week. Sometimes, it's super-relevant to fpl. And certainly helps with mental health when things go badly. If I know I made the most rational decision, it's easier to accept a bad-beat.
I've been playing two years, and apply probability theory to fpl, for me its the single most relevant aspect of decision making. However, eye test is crucial also, but not to pick out good players, more so to see patterns and likelihood of rotation and tactic changes
Often statistics like this don’t tell the full story, that’s why the shared knowledge of this site is so valuable.
But in a week like this where:
- The player you take out is guaranteed to score 0.
- It is a one week move and not a long term move assuming you’re WCing in GW32.
This makes this week in particular really well suited to the analysis above, whereas in other weeks it is not quite as clear cut (but still useful).
..also sounds related to backgammon (and life) strategies. great post!
well, there's nothing to deny your theory and calculations, i agree completely and play also that way. but i just can't put that "gut feeling" into these calculations. how can we relate those unknown sourced pulses into our decision-making process? how can we justify them with a scientific background? maybe adding an unknown coefficient for both positive and negative cases? if so, how do we calculate it?
Great question. Two points to make there:
1. Firstly, where the expected value is very close (for example between 3.7 and 4.3), then there is more scope to make a 'gut feel' call since we have to allow some margin of error in my calculations. But where the expected value of one move is significantly higher than another move, then we really shouldn't be making a gut feel call. We see this all the time where players go for a worse player just a differential - this is a bad move and will lose you points in the long-run.
2. As a practical matter, my analysis is better suited to very short-term decisions. Where it is longer term decisions we could theoretically calculate the expected value of two different players (as a means of comparing them) over a 6 week period, but it begins to hold less weight then and the reliability of our assumptions (i.e. the betting odds) becomes more tenuous.
great reply. this has been really helpful!
But why are we looking at averages? This is an individual simulation as you said.
It's like having 100 balls, 35 of them have 'CS' written on them, the other 65 have 'No CS'. And you pick a random one and you get 0 or 4 points for it. The chances of picking a CS ball are lower. And this is for one single match, we're not taking a hit for him to keep him for more games to get that 'average' score and possibly pay back the hit.
So according to your logic we would take a hit if there is a 51% of a clean sheet and we would not take a hit if there was a 49% chance of a clean sheet? Surely you can see why that is flawed and why using an expected value is a more robust means of decision-making.
Remember we only care about the mean and not the median provided that we're aiming for overall rank and not mini-leagues (remember again that the expected value is just another word for the mean/average). The implication of your method is that all we care about is the median. In this instance the mean is likely to be higher than the median, meaning the distribution will be right-skewed (https://www.google.com.au/search?q=right-skewed+distribution&source=lnms&tbm=isch&sa=X&ved=0ahUKEwjwjfKht-vZAhVKObwKHUXQBE8Q_AUICigB&biw=1440&bih=761#imgrc=Tir_0GM9fP6vOM🙂
We play FPL over the long-term and over many seasons. If we consistently base our decisions on averages we will do well in the long-run. Short-term fluctuations mean nothing.
jasong123 one of the few here with a brain
People here have much better football knowledge than me, which is reflected by my ranks in the last 4-5 seasons once I started following this site.
But I appreciate the compliment nonetheless :p
No, my logic is simple. I would risk losing 2 points if I can gain at least 3 points. The upside has to be better than the downside. Otherwise, I'm happy just getting 0 points with no risk. To gain 3 points after the hit, Butland has to score at least 7 points in this single match. Unless I think there's a high chance he gets 7 points, I won't take a hit.
Butland needs 4+ points to repay the hit, he doesn’t need 7.
You’ve confused the mean with the variance now.
Repaying the hit gets me nothing. It's 0 points. I can get that without getting him in with zero risk. If the likely outcome is him just repaying the hit, I'd rather not get him and avoid the risk of losing 2-3 points.
The likely outcome is you gaining 0.6 points, not breaking even. I agree that if the expected value was 4 rather than 4.6 then if you were risk averse you wouldn’t take the hit.
This doesn’t make any sense, you have to consider probability....if there is a 90% chance of losing two points and a 10% chance of gaining 5 points you’d be mad to go for it, flip it round and you’d be mad Not to go for it.
JasonG - wise words
So are you saying that you have a 90% chance of gaining 5 points and a 10% chance of losing 2 with Butland for a hit?
No he was just using hyperbole to illustrate the point.
Shouldn’t be interpreted literally.
Surely this chap is winding you up. Fairplay to you for being so patient with him and giving him such detailed explanations
I don't think he is, he's a member so he's more dedicated to this site.
It's not just for the benefit of him, but other FFS users out there since I'm sure many don't fully appreciate the use of statistics in FPL.
The expected return of a fair die is 3.5 even though it's not possible to get that on any single roll. This is the same idea. It's not a bad way of looking at it, though I think at 4.6 it's too tight to justify the hit.
Betting odds have nothing to do with how many goals a team will score
It is one way of measuring the probability of a clean sheet - it reflects the 'free market-value' of a clean sheet.
I agree they may be biased but it represents a good proxy, if you have a better measure for the probability of a clean sheet I'd be interested and I could update the numbers.
I'm just saying you are trying to correlate random events by the use of gambling statistics, I agree it works as a rough measure.
Yep it’s not ideal, my post was more just illustrating the importance of using expected values and was a very rough measure!
I would argue it is a lot better than a "rough measure". If it was just that, the bookies would be losing money every day
Fair point - but my calculation of bonus points and saves was 'rough' (no bookie odds available for that).
It would be a more accurate measure if bookies were not for profit. Since they are interested in making money, the betting odds are skewed in their favour, ever so slightly, so that (if you don't mind me borrowing your catchphrase) 'in the long run' they make money. This means there is in fact slightly lower probability of a winning result coming up than the odds imply. Or am I missing something? So 4.6 imo is an estimate slightly on the high side. I'm pretty high hope that all made sense
Yes you are correct that the bookies create a small margin for themselves so that they can always profit in the long-run.
However, I think this actually makes my estimate of 4.6 even more conservation.
The 35% chance of a clean sheet is based on clean sheet odds of $2.85. The bookies are likely to make odds worse so they can profit. Thus if the real odds were $2.65, the implied clean sheet odds would be 38%. Thus this actually works in my favour.
Are the bonus points and save points correlated to the chances of a clean sheet? If so the calculations are different. He's probably more likely to get bonus if he keeps a clean sheet and in a low scoring game? More likely to get save points if he doesn't keep a clean sheet, perhaps?
Yes of course.
A better analysis would involve a quick regression model.
But I think this rough calculation serves our purposes equally well.
mate, a good general FPL regression model would be extremely handy. I guess you'd need to do one for each position, and choosing the features would be an interesting challenge, but it wouldn't be too hard.
I think I could develop something like this quite easily - probably a few days of work but I'd need all of FPL's data including historical data. Don't have the time at the moment unfortunately but maybe in the off-season.
ah, just saw that you beat me to this. But I think this is an important point, since the correlations seem to be strong enough to change the expected value pretty significantly (as I noted below).
Nice read! You know your thing.
But then again, why do you own Kane?
Because he’s the best player in FPL after Salah of course!
Must've been quite something if he was worth paying over 2m coins for every point he scored.
Cheaper players will always be better value just because of the way prices are structured (the relationship between cost and player returns is not a linear function but rather a logarithmic function). But there is no point finishing with a team that cost 60m and has the best value players but not the highest scoring players.
Expensive players are also useful as captaincy picks.
I see you're very good at maths until it comes to Kane.
Team of 60m? I know hindsight is a great thing but if you were to pick the highest possible scoring team so far for 105m, Kane the top scoring forward wouldn't be nowhere near it.
Captaincy...? A fixture can't get better than home against Huddersfield and most didn't captain him that week.
My guess is that Jasong123 wouldn't have had Kane in his team for much of the season if he already knew how many points each player would have got at this point of the season. You said yourself, it's just hindsight. Maths doesn't have much to do with it.
I was trying to point out that when i wrote "over 2m coins for every point" i wasn't necessarily refering to the best value players.
If he wouldn't be injured he'd stil have the essential status or close to it for most on here. Not to mention that very likely it would be without armband 'til DGW37. By the sound of it that would include Jasong123 aswell.
This is how I make all my decisions in fantasy, good work
One area of interest for me is interpreting bookies odds, which I put at the forefront of my decision making, however often the odds are skewed for certain players that, for example, the bookies expect people will bet on to score even if they shorten the odds. So this always needs to be taken into account
Yes I agree bookie odds are not perfect.
It might therefore require you to use your own discretion when the decsisuon is very close I.e. if the expected value is between 3.7 and 4.3 then just use your best judgment (arguably 4.6 is within this range).
He has been lately better, so my number was something between 4.7-4.8. Best gk option anyway and worth the hit.
This is basicly the correct way to calculate expected points for any player. And no need for further or more complicated (or precise) calculations in this case. I use same math to decide who to bench etc.
Yes I agree my estimates were conservative given his average number of saves and bonus points are higher in recent games since the arrival of the new manager.
All the more reason to take the hit!
And the most likely outcome is -1 pt. But it is the price to pay to have that nice chance to get cs points and baps. And in this case it is worth the risk.
Yes exactly but that is why I emphasised the importance of looking at the mean rather than the median above.
I agree that it will lose you points more often than it will gain you points, but when you when you lose the loss will be less than what the gain will be when you win.
Good point about using it for benching decisions - that it the other area I've mainly used this sort of analysis throughout the season.
I've done it.
Think he's got a good chance of a clean sheet against an awful Everton attack.
With his potential save points and bps I think he could easily pay the hit back.
If he doesn't it's hardly the end of the world.
Twas a long thread & my brain was starting to hurt, at the point of scrolling past a few comments I saw this & thought, yes! -4, Butland in & armband!
Pretty fair analysis - seems worth a go to me - it's not as if it's a season-ender if you end up losing a couple of points!
Ooo my first hot topic - my whole life has been building to this moment.
Thanks guys!
Nice job mate!
Sound Logic & Statistical Analysis
(Very important that this is for 1 GW Only tho! Cause otherwise, that -4 can be paid back in following weeks as well)
Keep ‘em coming JasonG123!
Love your stats.
First off, respect for actually using decision theory and expected utility to make your fpl decisions. I've often planned to do this, but never found the energy.
Now, one thing that you need to be very careful of is to take into account any correlations between the different factors you are using to calculate the expected utility. Here, you've used both clean sheets and bonus points as independent factors. However, it is easy to see that they are highly correlated. 4 of the 5 occasions on which Butland got bps were also occasions on which he kept clean sheets. This means that his expected bps will actually be significantly lower than you've calculated here. Similarly, his save points are likely to be strongly correlated with clean sheets and conceding more than 2 (probably gets more save points when they concede a lot), so again the expected value for save points that you have here is probably too high. Given that the margin is pretty slim, this would be enough to suggest that getting him in for a hit has about a 50/50 chance of paying off in gw31.
*to clarify, i mean that save points are negatively correlated with clean sheets and positively correlated with conceding multiple goals.
Nah actually the correlation does not affect the expected value calculation at all. I've seen a couple people make this mistake above and it is a basic statistics concept that correlation does not affect expected return.
However, correlation does affect the variance of his scores (it will make the variance higher). This is what you have touched on. The variance refers to the fact that the correlation between saves, clean sheets and bonus points will mean he will either do really badly (1-2 points) or really well (8-9 points). But importantly on average it is still 4.6 points (using my rough calculation).
As I've said above we must focus on the expected value and not the variance provided we only care about overall rank. Conversely, if we cared about protecting a lead in our mini-league or catching up in a mini-league then we would care about variance.
Hope that clarifies it!
sorry man, i gotta disagree. the correlation is important. To simplify, imagine there existed a goalkeeper, call him Pavel, who only ever gets either 3 or 0 bonus points. Suppose that overall, he gets 0 bps 75% of the time and 3 bps 25% of the time. So on average he gets 0.75 bps per game. Now suppose that Pavel's chances of getting 3bps when he keeps a clean sheet are 50% and his chances of getting 3bps when he concedes are 5%, and that the chances of him keeping a clean sheet in the next match are only 30%. According to you, we'd calculate his expected value as 2 + 0.75 + (0.3)(6) = 4.55.
However, we can directly calculate the probability of Pavel getting bps in the next game as follows
P(3bps) = P(3bp|clean sheet)P(clean sheet) + P(3bp|no clean sheet)P(no clean sheet) = (0.5)(0.3) + (0.05)(0.7) = 0.185.
So the expected value contribution from bps would be 3 times 0.185 = 0.555, not 0.75 as you would use in the calculation. This shows that by ignoring the correlation, you will make incorrect estimates of the relevant expected values.
note that one important thing that's going on here is that pavel's chance of keeping a clean sheet in the next game is different from the general chance of pavel keeping a clean sheet in any given game (it's lower). You can think of it as the conditional probability of him getting a clean sheet given that he's facing a particular opponent or something.
The betting odds already take into account the fact that he's facing a particular team. While the save and bonus points do not, as I mentioned above these are actually conservative estimates given Everton at home is a relatively easier fixture and he's been in great form in the last 6 weeks.
right, this is kind of my point. the betting odds take into account that he's playing against everton, but when you calculate the expected value for baps, you don't take that into account (and just use a season-long average). My point is just that using the season long average for baps is wrong in a similar way to how using a season long average for clean sheets is wrong.
What you have done is create two different probabilities/numbers for the probability of him getting three baps (of 0.25 and 0.185).
You have said that the P(3 baps) = 0.25 but then you've said that the P(3 bps | clean sheet) = 0.5 and P(3 bps | no clean sheet) = 0.05 but these numbers don't match up because you've just created them from this air. I'm not going to work it out now but if the P(3 baps) = 0.25 then this would imply different probabilities for P(3 bps | clean sheet) and P(3 bps | no clean sheet).
In other words, if the P(3 baps) = 0.25 then the P(3 bps | clean sheet) cannot equal 0.5! The numbers have to match up with each other. You have defined the P(3 bps | clean sheet) as a fixed parameter when it is really a variable.
As an analogy - what you have done is say:
6 + p = 9
6 + p = 10
Therefore, p is both 3 and 4.
Summing up in layman's terms: it was already implicit in my figure that the expected value of Butland's BAPS was 0.5 what the probability of him getting BAPS given a clean sheet and the probability of him getting BAPS given no clean sheet.
So here's a concrete probability distribution that demonstrates that all of the above is consistent. I didn't fill in all the necessary details above.
Let CS = Keeps a clean sheet, BPS = gets 3 bps, E = opponent is Everton
In general, probability of keeping a clean sheet is 4/9, i.e. P(CS) = 4/9
The probability of getting 3bps is given by P(BPS|CS) = 0.5, P(BPS| no CS) = 0.05,
By the law of total probability, P(BPS) = P(BPS|CS)P(CS) + P(BPS|no CS)P(no CS) = (0.5)(4/9) + (0.05)(5/9) = 0.25
Now suppose that the next match is against Everton, and the probability of keeping a clean sheet against Everton is given by P(CS|E) = 0.3. Suppose also, that when playing against everton, the probability of getting bps given that you keep a clean sheet is the same as usual, i.e. P(BPS|CS and E) = P(BPS|CS) = 0.5, and P(BPS|no CS and E) = P(BPS|no CS) = 0.05.
Then the probability of getting bps against Everton is
P(BPS|E) = P(BPS|E and CS)P(CS|E) + P(BPS|E and no CS)P(no CS|E) = (0.5)(0.3) + (0.05)(0.7) = 0.185.
Whereas the general probability of BPS (playing against an unspecified opponent) is 0.25, as I showed above.
This is all a completely coherent probability distribution, and I'm really just making a very simple point. BPS are obviously correlated with clean sheets. If you want to work out expected value from BPS, you need to condition on any variables that can affect the probability of keeping a clean sheet.
Anyway, sorry for sucking all the fun out of the 🙂
*condition on any variables whose value you already know
** out of the thread 🙂
I can now see what you're missing: in my opening post the probability of Butland getting a clean sheet was 35% - this was based on the betting odds from the match against Everton itself. This figure was not the number of clean sheets for Butland divided by the total number of games. Therefore, this correlation that you're talking about is already captured within the 35% figure.
Also the numbers weren't consistent: you just changed the clean sheet odds for them to arbitrarily be consistent. Look up whether the correlation affects expected return, you'll find many papers stating what I've said (here's one example from a statistics forum: https://www.bionicturtle.com/forum/threads/expected-return-covariance-correlation.6077/).
sorry man, but i still don't get what you're saying.
I know that you used the betting odds for the probability of getting a clean sheet. That's obviously the right thing to do. But it doesn't account for the relevant correlation (between bonus points and clean sheets) because you're still just using a long term average to calculate expected bonus points. You look at the probability of stoke keeping a clean sheet conditional on them facing Everton, but you look at the unconditional probability of bonus points. You should use the probability of keeping bonus points conditional on facing Everton.
P.S. The probability distribution I gave in the previous post is definitely consistent. I even put it in mathematica to double check.
Ah I see where the confusion arises: you are correct in saying that I used a long-term average to calculate expected bonus points and expected save points whereas I used the betting odds for this particular match to calculate the expected clean sheet points. As I indicated above, this was just a rough measure since there are no betting odds available for save points and bonus points. As I mentioned, by taking a season average I actually took a conservative approach since Everton at home is a relatively good fixture and Butland has scored much better since the arrival of the new manager, so it is indeed likely that his expected points is greater than 4.6.
It's always important to note whether any methodology has bias and even more important to note the direction of the bias. Here, the bias is negative, meaning I'm actually underpredicting Butland's score, which means it makes even more sense to take the hit!
P.S. it's only consistent because you 'invented' the clean sheet odds of 5/9 which don't make any sense and weren't in your first post. You treated the clean sheet odds as a variable when it should have been a fixed parameter and you treated the probability of getting three baps given a clean sheet as a parameter when it should have been a variable. Anyway all of this is actually irrelevant since it was just a misunderstanding about my methodology, but thanks for the interesting discussion!
Cool! Thanks to you too mate. I think we are mainly on the same wavelength now, and I definitely approve of your overall methodology. I still think the original calculations are a little off because the save points are strongly negatively correlated with clean sheets, and they make quite a big contribution to the overall expected value. But of course, any such calculation is always going to have to make some idealizations, and you're definitely in the right ball park.
Yes, because I used to different methodologies the calculations will be a little off. The correlation matters in this instance because of the different methodologies but theoretically it wouldn't matter if I were using the same methodology - that is where the confusion arose.
Note that for Butland he averages 3.4 saves in clean sheets (only 5 so small sample size) where he averages 4.1 saves in games where he concedes. It is a noticeable difference but not overly huge, probably wouldn't positively bias the expected value too much and it would be offset by the negative bias created by not adequately taking into account the correlation between bonus points and clean sheets.
Clean sheet will gone with one mistake by him or any team mates..
We're trying to take a long-term view and look at his average score in the long-run rather than focusing on small fluctuations.
It is important not to be results-orientated.
Your patience in this thread is absolutely incredible!
Toffees to win 2-0. Its a no from me
That is a possible, but not a probable outcome.
FFS projections have Butland at 4,0 points for gw 31.
If we add a bit of entertainment value as well (you would be really happy if he in some way got you 10+ points, but if he only got you 2 it would probably be no big deal), then it's probably fine to take the hit.
The likely outcome is either -2 or -1 for the hit, or +2 or +3. Either way it's no big deal.
There are many different models of predicting a player's performance. The expected value, regression analysis, the FFS model etc. Choose which model you think is the most reliable and base your decision on that.
I get your point but...
The problem is that thats average points
It will be hard for him to get 4 points and inpossible to get 4,6 points
So probably he will get 2-3 or 7-9 points
So you will lose or gain some points
If you are gonna wc after this gw its all about this gw and luck
Yes but we play FPL in the long-run since we play over many seasons - so it makes sense to make decisions based on this type of analysis.
Of course in any individual week his score could range from 0-15.
You are misunderstanding how expected value works.
Love this whole thread, captures beautifully just how much time and effort we're willing to put into the most marginal of decisions in this silly game 🙂
It is a marginal decision in a silly game. But it is not just this marginal decision in FPL, but it can apply to all decisions in FPL!
Indeed even more importantly, it can apply to any decision in life!
To be clear - not meaning to deride your efforts in the slightest, genuinely applauding them!
I got myself a maths degree the thick end of 20 years ago so I always enjoy this kind of thing 🙂
I know you weren't deriding my efforts - and by the same token by post was positive towards yours in a joking kind of way.
Thanks for the response - since my background is only in economics and statistics and not in maths, it would be cool if you had any insights to weigh in on this discussion as well. 🙂
I'm afraid 20 years of not using the maths/stats tools I learnt have dulled their keenness. I like to think I still have an aptitude but there's nothing I can bring to this table - it's already groaning under the weight of well argued numbers 🙂
+0.6
Nope. It is actually much closer to +1 😉
This whole thread makes me feel thicker that I already thought I was. Don't measure my thickness tho please. It's well documented 🙂
People who aren't happy with their ranks need to realise this post/thread might contain the most important advice they'll ever need to be more successful in the game. This kind of thinking also applies, as you say, everywhere in life!
Errrmm, all of this probabilistic analysis is great. But isn't this only true if you're willing to keep Butland after BGW31 too?
Assuming you get him out after BGW (given Stoke's fixtures aren't great), then that's an extra transfer and the required yield is 8 points before it's worth it. (Even if it's a free there's the opportunity cost of 4 points)
This whole debate is about whether the hit is worth it for one player and for one GW. In other words, it's assuming a WC in GW32, which negates any context after this GW.
You could tweak it if you plan to WC in GW33. But if you plan to WC in neither week, then you don't have the luxury of simplifying the decision in this way, and you would have to take more factors into account, as you are saying.
I like the cut of your jib. I've tried talking about means in a right-skewed distribution recently to no effect.
Have you considered the idea that a hit is just a cheap lottery ticket, so you may as well buy one to increase your chances of a jackpot? 😉
https://www.fantasyfootballscout.co.uk/2018/03/12/line-up-lessons-gameweek-30-part-one/?hc_page=5#hc_comment_17798652
I wish this site was a bit more forum-y, so there where spaces to see all the good work people share about specific topis. Good posts could rise to the top, and fuzzy thinking could sink to the bottom. It's frustrating to see useful posts so rarely, and for discussions to be so fleeting, forgotten so quickly. I wonder whatever happened to the new comment filter/search function we were expecting this year.... Anyway, I'm off to have a cry in a corner. Keep up the good work. 😀
I’m bringing in Baines and Butland for -4. Think that will pay off!
Good analysis. That’s how I make decisions over the season too.
However I would just like to point out that you are evaluating a single Gameweek. In such cases, expected value analysis tend to be less useful over a short period of time. I prefer to use scenario analysis -> what’s the worst case and upside case. Putting in this context makes a hit for GK with slight positive expected value gain less attractive, as compared to a hit for an attacker with similar expected value gain => better worst case of 2 pts (assuming no YC) and more upside potential
In theory, even if it's a single GW, we should use probability and expected value ("EV") to make our decision. The point is, if you make most of your decisions by EV throughout the whole season, you will enjoy the benefits in long-run, in which this Butland trade is just one of the many decisions you have made based on the same EV analysis.
There is just too much randomness in a single gw for me. The core assumptions (eg cs odds of 35% used in the calc), are more useful on average over many gameweeks.
Using EV for short-term decisions may lead to false precision and comfort. The -4 hit is certain but points potential is uncertain.
I think using scenario analysis/decision tree is a more rational approach for short-term decisions esp. when EV gain is insignificant. It simply re-frames the EV analysis by thinking about downside and upside potential.
A well-constructed scenario analysis/decision tree will actually turn to EV.
You talk about taking hit for an attacker is more attractive, because you assume a striker has higher ceiling than a GK, and is less likely to score less than 2 (GK will score you 1 pt if conceding 2 goals). Yup, I agree with these statements.
However, you need to think about how likely such attacker will score, and how many he will score. In theory, every attacker can score 5 or 6 goals in a single match, but you know the chance is higher for Aguero/Kane, and alomsot zero chance for Mounie/Crouch/Deeney (but still it's not 0%). So when you talk about ceiling, it's not precise at all.
You still need to put a probability of each attacker to have 1/2/3/4/5 goals and 1/2/3/4/5 assist. If you don't do so, you are more on a gut feeling of who will score/assist next GW. But actually, such gut feeling should be and could be quantified. Being a member of this site and you can take a look of the EV of all attackers playing next GW, adding/subtracting those numbers by your own findings, then you know who are worth to take a -4 hit.
Last but not least, based on what I analyse so far, bar Liverpool, Shaqiri and Stanislas, the chance that Butland gets a CS is higher than most of other players getting a goal or an assist. Butland also has higher chance to accumulate save points. While Stoke is not a high scoring team, if Stoke keeps a CS, Butland has higher chance to get 2 or 3 bonus points, in which it is the opposite for the likes of Karius/Begovic. That said, Mounie/Wilson/King may score 5 goals, but at the same time they are a lot more likely to end up with 2 points.
It's not just about the ceiling and the worst case, but how likely they will gonna be happened.
Hey Kenneth - thanks for taking the time to write a thoughtful response. Appreciate the useful exchange of ideas.
Yes, I agree using scenario analysis is similar to EV, hence I suggested it simply re-frames the EV analysis. I am merely suggesting that the distribution outcomes matter more for a 1-gameweek analysis than making decisions purely on EV gain.
Eg if an attacker has an EV of 4.6 same as a GK, the expected gain is the same. However when we recognise that EV is only true on average over time, it may be more helpful to reframe using distribution outcomes. Simplified example below:
Attacker:
- worst case 2 ppg (no cards) with probability 80%, upside case 15 ppg (2 goals) with probability 20% => EV 4.6
GK:
- worst case 0 ppg (no cards; concede 4) with probability 10%, avg case 2 ppg (concede 1) with 30% prob and upside case 10 ppg (cs + max bonus + save) with probability 40% => EV 4.6
Over many games, I am indifferent between both options. But over a single gameweek, recognising the impact of randomness on short-term outcomes, I would prefer taking a hit for an attacker with lower downside and more upside optionality.. From my example above, I agree the prob of a brace for an attacker (20%) is lower than GK keeping cs (40%). But the downside is also more attractive.
Finally, I also fully agree with your EV analysis of the current options. Personally I have a EV of 4.4 for Stanislas and 4.3 for Butland. I recognise Butland is a good option (btw I am bringing him into my team), just sharing that between Stanislas and Butland, I would take a hit for Stan but probably not for Butland by reframing my EV analysis.
I really like your explanation and approach, you are a great FPL player and am confident you will understand my perspective in time.
Sorry, I realize the GK example should be updated below for math to work haha:
GK:
- worst case 0 ppg (no cards; concede 4) with probability 30%, avg case 2 ppg (concede 1) with 30% prob and upside case 10 ppg (cs + max bonus + save) with probability 40% => EV 4.6
Hi friend, just read your replies, and thx for them. It's a great discussion. Quite interesting to see the results end up quite close to the probabilities. Stan > Mounie/Wilson/King. Butland and Shaqiri are outliers this time that should not put into comparison, due to the unexpected early RC by Crazy Adam. That said, Butland still score 3 points 😛
2 things:
1 thank you for this excellent analysis, I’m now bringing butland in!
2 It’s more fun to have a player who is playing than someone that isn’t, not that that forms part of your analysis!
Absolutely admire your patience in this thread! Surprisingly few understand what Expected Value (EV) means. I guess it has something to do with the fact that I've played poker for living quite a long period so it is automatically part of my decision process and those who don't work around math/numbers don't really think like that. Been saying the whole week Butland is worth the hit. I didn't even have to break it down to numbers as you did 🙂 Just looked at his save points and figured it would be a +EV decision to gamble for cs (+bonus/pensave). Your math backs it.
And like someone said in this thread, you didn't even include the value of owning player ;)That is especially valuable for casuals like me so I don't have any hesitation to take couple gambles this week when risk is quite minimal and there is clear upside. This part is besides the point but it shouldn't be completely excluded either haha.
Keep up posting man, quality stuff.
also from a poker background and agree wholeheartedly.
The particular circumstances of this GW and the GK position make it ideal decision to analyse in this way - which JasonG123 has done brilliantly. Such math analysis is beyond me - but like Figu reassuring what seemed like a +EV call is backed up by the numbers.
Is he worth -8?
😆
You have basically created a simplified version of the RMT members function.
All your thinking and understanding of the expected points approach to the game game is sound and your patience explaining to others borders on saintly.
Cheers, never actually used the RMT feature (was a member the last two seasons, but not this season).
I'm sure it does a good job of this as well.
pickford
ake taa
walcott salah
wilson king firmino
0 FT
GTG?
you're gonna get overrun in midfield
Mane for -4?
Shame that Ake and TAA are blocking Stanislas, VVD and Mane. I'd probably consider a hit for Zaha or Shaqiri.
Hey Jasong 123 - you might be able to solve this for me. Is there anything wrong with getting in a defender and attacker who are facing each other if you think they represent the best value that week? Should this be part of your decision-making, or is it irrelevant?
Ps Wondered why you asked a million questions on here about every little decision you make - obviously you've made a determined logical decision to mine the football experts (lol) for their info.
What you think for the below GW31 FH starting 11. Unsure about GK and Mili/Stanislas pick?
Lossl/Pickford
Coleman Zouma Ake
Salah(C) Mane Shaqiri Mili/Stanislas
Mounie Wilson Firmino
know it doesnt matter but just out of interest, how much itb?
Team is fine
8.8M
Any news about kane? and aguero?Make sense have a team with: Lukaku, Firmino and CLV instead aguero or kane?
Make sense keep 3 players of Man united after exit of CL? in my mind i have: De Gea, Smalling and Lukaku. or matic instead de gea to have make money for other player.
Thanks.
Pickford
Mariappa, Robertson
Walcott,Salah,Mili,Shaqiri
Firmino,Wilson
Is this enough or should I get Tosun in or just save ft which would let me have 2 transfers allowing GW33 WC rather than 32
PVA or Baines for a -4 punt?
Currently have Schindler so any attacking returns PVA does get would impact his points.
Baines
PVA!...
Zaha or Shaq for a hit?
Zaha
Wilson is probably not going to start this weekend with Defoe back in the picture right?
defoe was back in the picture last week too
100% correct, sell for a hit ? 😉
Just trying to gather opinions. 😉
Wilson had an ok game vs Spurs, would be surprised if he didn't start this week.
Best transfer guys ?
a) kane > mounie as FT (wud mean 10 gw 31 players)
b) alonso > PVA & stephens > stanislas for a -4 (wud mean 11 gw 31 players)
Tnks
B
B
Really is some dross getting touted for hits. I think 9/10 times it is going to be worth going without
How about Choupo > Stanislas, Kane > Mounie and Dunk > Schindler for a hit?
It’s only -2 though 😉
How can you say that it’s onky -2 when the likes of Kun Kane Ayew didn’t even muster a total of 2 points between them all this week? No points are garanteed
In fact 30million worth of player would have gotten you 1 point.
If I’m miss reading sarcasm then do excuse me 😀
Irony fella - look at the wink 😉
Only works on forwards and mid if they play atleast 60 min and no yellows.
🙂
Agreed Tinman
Always hype before where suddenly everyone begins to become an option then reality sets in once the game begins.
I say this as I was contemplating Rondon for a hit
Sshhh.. let them take the hits. Good for us.
Which one better:
a) Sterling to Stanislas
b) Elliot to Hennesey and Sterling to Mkhi next week
B,
Neither really. Think you'll want both Stanislas this week and will want an Arsenal attacker next week.
why?
A
Good meowing one and all.
Zaha and Stanislas are my targets with 2 FT.
I have 10 outfield players for GW31 and considering a GK for a hit.
See above my post on Butland, I think he's worth the hit.
Butland might be worth it. I've got Hennessey otherwise I probably would have. Butland seems almost guaranteed 3pts these days
Please advise - Kane or Aguero for Firmino?
Thanks a lot!
Kane
Worth a -4 ?
Ogbonna > PVA ?
Would mean going with 11 GW 31 players.
Tnks guys,
Yes, I like it.
Can’t wait to wildcard. The following looks so juicy playing around with the squad:
DDG
Alonso Davies Mee
Salah Son Mahrez Willian Silva
Lukaku Aubameyang
4.5 4.5 4.5 4.3
Looks a good team.
Who will be the 3rd forward?
Not sure mate, will have to look at DGW’s etc
Do you plan to play BB in GW34 with those "4,5" players?
Dropping Firmino early with those fixtures ?
Will the FA Cup semi finalists rotate players in the second game week of 34 on the basis that it is the only trophy they can win ?
Dont hide your bench please 🙂
Can anyone please explain 1 thing to me on the Free hit please?
If i make changes today with my team- Can i continue to make further changes to my team through the week? Or do i have to confirm my free hit team at 1 time?
Unlimited changes like WC
You can confirm your free hit after transfers and they will all be wiped clean, you can then change your team a gazillion times if you want ...
How would you improve this team?
Courtois Elliott
Otamendi Dawson VVD Duffy Kenny
Salah Shaqiri Walcott Milivojevic Van La Parra
Kane Firmino Quaner
1 FT 9.0 ITB. Plan to WC Gw32 or 33 and FH Gw 35
Thinking Courtois > Butland or Lossl. Any other suggestions?
courtouis to any gw31 is no brainer, i would do it...you all set on gw31 player i tihnk
is there any hot topic to discuss wc gw 32 v gw 33? still cant decide yet..
Depends on how happy you are with the fixtures of your players in GW32.
Ayew to who out of Mounie/Rondon/Wilson/Benteke?
- -
VVD, Bauer, Coleman - -
Walcott, Salah, Shaqiri, VLP -
(ayew), King, Firmino
wilson/mounie but idk if they guarantee to start
Wilson - go for glory in case Bournemouth explode 🙂
Loving the Eliteserien discussions, it is all about the community helping each other and providing information, rather than people just asking others to pick their team for them 🙂
Who will score more points this GW ?
a) stanislas
b) mounie
Tnks guys.
Stan
Stan the man
15 points each
Mane or Firmino as 3rd Pool asset?
Firm
This is basically my quandary.. really fancy Mane but fear Firmino ownership.
Probably going with fear and get Firmino
Mane if you're chasing
Chasing but only 10 pts behind so nit a big difference
How do you reckon SPurs will line-up in PL without Kane?
Son as a 9? with Moura on the right, Alli on the left and Eriksen in the hole?
Llorente will start
Son
Lamela Alli Eriksen
Two to buy;
A
Firmino and Walcott
Or
B
Mounie/King/Wilson and Mane
Already got VVD and Salah
Thanks
How many DGWs are we likely to have before the end of the season?
Two, three or four ... most likely two though
2
WC next week to set it all up, then BB for one, Free Hit chip for the other?
Yes very sensible - means you can concentrate on the early doubles in GW34 (or when ever they are), and have a completely different team of doublers in 37
GW31 Hall of Shame:
Bauer, VVD, Kenny*
Shaq, Salah, Walcott
Firmino, DCL
Fab, Speroni
Dunk, Alonso
Sterling, Arnie
Kane
2FTs + 0.9m
Any thoughts?
Hi Managers,
6 players so far:
VVD
Stanislas Shaq Salah
Firmino Wilson
Had to do Siggy to Stanislas with the FT so...
Any players worth a hit so that I have 7,8 or 9 players this GW?
A) Alonso to Baines (-4)
B) + Kane to Mounie for (-8)
C) + Obiang to Zaha/ Wallcott. (-12)
With Kane now out, lineups will be highly Salah (c). This could leave leagues in dreadlock.
Current week 31 players
Lossl
Coleman/ Robertson
Salah/Shaqiri/
Firmino/Crouch
Thinking about bringing in Stanislaus and Zaha for -8
But tempted to get PVA as well for - 12.
Worth it or too much of a gamble?
Thanks
- 8 probably
- 12 too much for me
This will be my team for the weekend with 1 hit taken.
Vvd dawson
Salah walcott shaq Stanislas
King firmino
Is it worth moving to - 8 and bringing in mounie too?
Thanks
It depends on whether you are chasing in your ML. If you are I think it's worth a punt. If not then I'd stick with what you've got
I'm not chasing
I have a slender lead
