Champions League 24/25, Matchday 4 (5-6 November)

[Dniprovec]

@Lyonnais Thx for posting this after every MD, always interesting! One question, what are your expected points for Bayern?

I asked @Lyonnais after MD 2 the same question and he wrote:

Let me try explain how to works

Bayern are playing next at Barcelona

Current ELO points
Barça 1899
Bayern 1912
This gives us an 1X2 probability of 43-27-30
My calculation then assumes that Bayern has 27% to catch 1 point and 30% points to get the 3 points, and thus will get 27%*1+30%*3= 1.17 point (1.56 for Barça)

I make the additions of all these probabilities
1.98 point vs Benfica (home)
1.66 point vs PSG (home)
1.97 point at Shakhtar (away)*
1.59 point at Feyenoord (away)
2.84 points vs Slovan B. (home)

* the "away game" concept can be challenged on this one I agree

This gives us a total of 14.21 points (already 3 + 11.21 projected)

So, if we change now the predicted points vs Barca and Benfica (together 3.15) into the effective 3 it makes - assumed that the ELO points didn't change that much in the meantime - 8.06 point more and 14.06 in total. Definitely too less for top 8. However, I count 9 points from the last 3 matches (i.e. 2.6 more than predicted) and at least 1 point from the PSG match (0,66 less than predicted). So it makes a total of 16 points that should be enough for top 8.

16 points is about 60% chance for top-8 for now.

[ralfinho]

16 points is about 60% chance for top-8 for now.

Are you sure? I remember another thread here in which the nerds came to the result that 15 points maybe will be enough, 16 points quite sure and 17 points definitely.

[Dniprovec]

16 points is about 60% chance for top-8 for now.

Are you sure? I remember another thread here in which the nerds came to the result that 15 points maybe will be enough, 16 points quite sure and 17 points definitely.

Well, it was before the start of the competition. With how the actual results are now the bar is higher. I'd guess it has to do with somewhat an unusually low draw percentage so far - 14%.

[Dniprovec]

Post MD4 chances:

[babaluj1]

16 points is about 60% chance for top-8 for now.

Are you sure? I remember another thread here in which the nerds came to the result that 15 points maybe will be enough, 16 points quite sure and 17 points definitely.

Well, it was before the start of the competition. With how the actual results are now the bar is higher. I'd guess it has to do with somewhat an unusually low draw percentage so far - 14%.

Exactly, very few draws in CL and CO leagues, only 2 draws on average per round, unlike them in EL there are far more draws, 19 in total compared to 8 draws in CL. This happens because of the big difference in the class of the best and weakest clubs in CL and CO, while the quality of clubs in EL is much more equal. In EL, 15 points should be enough for the Top 8, while in CL they will probably need more.

[SimonB]

Are you sure? I remember another thread here in which the nerds came to the result that 15 points maybe will be enough, 16 points quite sure and 17 points definitely.

Well, it was before the start of the competition. With how the actual results are now the bar is higher. I'd guess it has to do with somewhat an unusually low draw percentage so far - 14%.

Exactly, very few draws in CL and CO leagues, only 2 draws on average per round, unlike them in EL there are far more draws, 19 in total compared to 8 draws in CL. This happens because of the big difference in the class of the best and weakest clubs in CL and CO, while the quality of clubs in EL is much more equal. In EL, 15 points should be enough for the Top 8, while in CL they will probably need more.

I don't disagree with this, but I would also suggest another reason on top of this which might also be contributing to fewer draws.

In the old system when you got a draw it was against a direct opponent so it was of some value to you in the group. Now if you get a point it isn't useless but overall you are probably losing ground against the field as a whole and so therefore you are probably going to be less satisfied than before with just a single point, so the new system perhaps encourages more positive play. I don't know if UEFA thought about this or whether it is just a happy coincidence. Also, there are games between teams from the same pot so you would expect more draws when these teams meet in theory at least and it appears not to be happening that much.

We are only 4 matches into the leagues (only 3 in the CO), so perhaps we ought to look at the numbers again at the end and see how things have worked out at that stage, whether they are still following the same pattern, or if anything has changed.

[ralfinho]

Thank you all for your contributions.
I might suggest another issue. The ELO system often delivers point distributions between 1.5 / 1.5 and 2 / 1. It seems that actually the higher ranked teams win more often than predicted. That leads to a higher spread between the upper and the lower teams. In mathematical terms we expected the distribution of points per team as a (Gaussian) normal distribution with a mean of - let's say - 12 and a standard deviation of 3. Actually the deviation will be higher, maybe about 4 or even more.

[Dniprovec]

Thank you all for your contributions.
I might suggest another issue. The ELO system often delivers point distributions between 1.5 / 1.5 and 2 / 1. It seems that actually the higher ranked teams win more often than predicted. That leads to a higher spread between the upper and the lower teams. In mathematical terms we expected the distribution of points per team as a (Gaussian) normal distribution with a mean of - let's say - 12 and a standard deviation of 3. Actually the deviation will be higher, maybe about 4 or even more.

What makes you think so?

First of all, almost half of all matches so far are between the teams with distributions between 2 / 0,8 and 2,8 / 0,1.

Secondly, when comparing top-18 teams by expected points vs bottom-18, the first group has 130 points out of 128 expected, while the second group 76 out of 72 expected.

Total number of actual points is +3% (206 vs 200) higher than expected. I know you can't really approximate it like that, but 16 points expected to be enough for top-8 become 16,48 with 3% more points out there up for grabs.

[ralfinho]

First of all, almost half of all matches so far are between the teams with distributions between 2 / 0,8 and 2,8 / 0,1.

Secondly, when comparing top-18 teams by expected points vs bottom-18, the first group has 130 points out of 128 expected, while the second group 76 out of 72 expected.

I did not check that totally. That was my impression whenI looked to Lyonnais' predictions of the Bayern matches. They were 1.17 point (Barça), 1.98 (Benfica), 1.66 (PSG), 1.97 (Shakhtar), 1.59 (Feyenoord) and 2.84 (Slovan B). So only one in the range you mentioned.
Anyway, it's clear that the spread is wider than expected. Maybe due to the low number of draws, maybe due to results than are more onesided than expected.

[MartinW]

Based on current standings, the knockout rounds could be:

Code: Select all

Liverpool(1)/Sporting(2) v  Celtic(15)/Dinamo Z(16)  or  Bayern(17)/Real M(18)
Dortmund(7)/A Villa(8)   v  Atalanta(9)/Man City(10) or  Atletico(23)/PSV(24)
---
Monaco(3)/Brest(4)       v  Leverkusen(13)/Lille(14) or  Benfica(19)/Milan(20)
Inter(5)/Barcelona(6)    v  Juventus(11)/Arsenal(12) or  Feyenoord(21)/Brugge(22)

[babaluj1]

Post MD4 chances:

How did you come up with these predictions? I find it a little strange that Dinamo, which has 7 points and plays three more home games, has a 53% chance of being eliminated. I think that his chances of going to the knockout stage are certainly higher than being eliminated. One win is enough for Dinamo to be safely in the knockout phase, and in some combinations even 1 point.

[fabiomh]

ELO system works with 2 pts for win, not 3. (sum of points is constant)
So, in my opinion, comparing expected vs actual pts calculated with 3 pts per win can be tricky.

ELO system don't take into account the risk approach of the LS, well explained by SimonB; so probably ELO system overestimates draw probability in the LS, causing some bias in the 3-pts-based calculation.

In order to check if top18 has more or less points than expected we should:
1) define ex-ante which teams are top18 (the teams with higher ELO before start of LS)
2) compare expected vs actual with 2-pts-based calculation

[fabiomh]

If your prediction were to hold exactly true, then we would have a 50/50 chance of a Bayern/Real Madrid play off match if I am not mistaken about the new draw method!

Yes, if Bayern finish 14th and Real finish 19th, there will be 50/50 probability to have Bayern-Real M. at KO (1st leg in Madrid)

Of course now it's still open, probabilty are really spread, and currently the probability of such a match are probably about 3-5%.

[Dniprovec]

Post MD4 chances:

How did you come up with these predictions? I find it a little strange that Dinamo, which has 7 points and plays three more home games, has a 53% chance of being eliminated. I think that his chances of going to the knockout stage are certainly higher than being eliminated. One win is enough for Dinamo to be safely in the knockout phase, and in some combinations even 1 point.

Dinamo (1577 elo) plays Borussia (1829), Celtic (1700), Milan (1834) at home and Arsenal (1943). Now, depending on where you have you home-field advantage, since Dinamo plays three games at home, it can make a big difference. In my model it is at at +65 elo. I run the simulation 10 000 times.

But without going into too much detail, let's try to explain that change. Those elo differences are roughly translated into the win-draw-loss %: 20-20-60 vs Borussia (21-24-55 per bookies now), 25-35-40 vs Celtic, 20-20-60 vs Milan and 5-10-85 vs Arsenal. Now, the model says that with 7 and 8 points it is virtually no chance to advance given the results so far and goal differences (especially in the Dinamo's case, with it almost always having the worst goal difference compared to the clubs with the same amount of points). Dinamo has 34% chance of getting 0 or 1 point in the next 4 matches given those wdl%. Model also says it is only 33% to qualify with 9 points for the clubs that do, but it will be lower for Dinamo, since worse goal difference (2 draws and 2 losses to be at 9 points would only worsen goal difference). Dinamo has about 10% of getting 2 points given those wdl%, but probably would qualify only in 10-20% with 9 points. So, it gives us roughly 45% chance of Dinamo being eliminated if it gets 0-2 points. I wouldn't calculate the chances of getting 10 and 11 points, but there are some cases (especially for Dinamo given goal difference) where even those won't be enough. So this is how 53% chance of elimination can be explained.

Maybe if taken bookies chances it would be few % higher here and there, but it still will be close to a coin toss at this point.

[Dniprovec]

ELO system works with 2 pts for win, not 3. (sum of points is constant)
So, in my opinion, comparing expected vs actual pts calculated with 3 pts per win can be tricky.

ELO system don't take into account the risk approach of the LS, well explained by SimonB; so probably ELO system overestimates draw probability in the LS, causing some bias in the 3-pts-based calculation.

In order to check if top18 has more or less points than expected we should:
1) define ex-ante which teams are top18 (the teams with higher ELO before start of LS)
2) compare expected vs actual with 2-pts-based calculation

You can convert elo difference to wdl%, similar to how clubelo does: http://clubelo.com/UCL/Games, so that your expected scores would reflect it accordingly. For example, clubs with equal elo will have something like 35-30-35, which would translate to 1,35 expected points for each. While the difference of, for example, 200 gives you 65-20-15, which is 2,15 for one club and 0,65 for another.

1) it is done, the only difference is teams with higher sum expected points given their pre-match elos.
2) expected points as explained above are compared to 3-pts-based calculation