UEFA Seeding & Qualification based on ClubElo Values - Updates & Feedback

including formats, draws, seedings, etc.
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

You may have noticed that I have added average and median Elo values for each pot -- this helpful when comparing pot strengths within and across the tournaments.

ImageImageImage
fabiomh
Senior Member
Posts: 1466
Joined: Thu Sep 24, 2020 20:00
Location: Milan, Italy

Post by fabiomh »

Stadion wrote: Mon Feb 05, 2024 08:55 You may have noticed that I have added average and median Elo values for each pot -- this helpful when comparing pot strengths within and across the tournaments.

ImageImageImage
Very interesting picture.
I have a question:
will 2024-25 CL Pot1 be like now, including the Champions of the first 6-8 Countries?
Or will it based on Teams coefficient only like Pot2, Pot3 and Pot4?
Hope for more partecipants in the next Prediction Game
amirbachar
Senior Member
Posts: 1747
Joined: Thu Oct 18, 2007 02:22

Post by amirbachar »

There seems to be a bug, since Zalgiris are somehow exoected to qualify in Q3 and Olympiacos eliminated.
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

amirbachar wrote: Fri Feb 09, 2024 14:40 There seems to be a bug, since Zalgiris are somehow exoected to qualify in Q3 and Olympiacos eliminated.
Might have been temporary, because currently Olympiacos are projected to go to the ECL groups while Zalgiris are eliminated in Q2.
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

Stadion wrote: Mon Jan 29, 2024 04:26 NCQ2 -> ELQ3
CHQ3 -> ELPO

ELQ1 -> ECLQ2
ELQ2 -> ECLQ3
ELQ3 -> ECLPO
Based on the revised logic above, I want to check if the spreadsheet is currently displaying the correct information:

- Molde starts at ELQ3 which they don't progress from, due to their Elo being lower than the majority of clubs in ELQ3
- Being the highest ranked unseeded club, my logic has them drawn against the lowest seeded club: Panathinaikos, with a coefficient of 5.585
- Since they move down to ECL as a loser, they take on the lower coefficient of the pair
- In ECLQ4 they are now the lowest ranked club, and unseeded
- Though thanks to their Elo being higher than the majority of seeded clubs, they are still forecasted to progress to the ECL group stages

Is this correct?
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

In other news, I am including average and median Elo of seeded and unseeded clubs in each qualifying round -- currently only for UCL. I can expand this to UEL and UECL if it's useful.
Squid
Posts: 48
Joined: Tue Aug 18, 2020 15:00

Post by Squid »

Stadion wrote: Wed Feb 14, 2024 03:33
Stadion wrote: Mon Jan 29, 2024 04:26 NCQ2 -> ELQ3
CHQ3 -> ELPO

ELQ1 -> ECLQ2
ELQ2 -> ECLQ3
ELQ3 -> ECLPO
Based on the revised logic above, I want to check if the spreadsheet is currently displaying the correct information:

- Molde starts at ELQ3 which they don't progress from, due to their Elo being lower than the majority of clubs in ELQ3
- Being the highest ranked unseeded club, my logic has them drawn against the lowest seeded club: Panathinaikos, with a coefficient of 5.585
- Since they move down to ECL as a loser, they take on the lower coefficient of the pair
- In ECLQ4 they are now the lowest ranked club, and unseeded
- Though thanks to their Elo being higher than the majority of seeded clubs, they are still forecasted to progress to the ECL group stages

Is this correct?
This is how I interpret the logic - I think you are right.
User avatar
emortal
Posts: 949
Joined: Fri Oct 19, 2007 22:21
Location: Belgium

Post by emortal »

@eye thank you for the spreadsheet, very handy (https://docs.google.com/spreadsheets/d/ ... =632987334).

For some reason I cannot select the language, it only shows up in Dutch, but that's not an issue.

The CL-W rebalancing table could be improved by including all high-ranked teams by associations where the champion doesn't qualify automatically, along with their respective position in the league.

A similar result could be achieved if there was a country filter on the club ranking.

I.e. currently PAOK is showing up top, but Zvezda has a higher ranking (40.000 vs 35.000) and is in second place in the Serbian league only trailing Partizan by one.
elkjiaer is back
Posts: 271
Joined: Fri Jul 28, 2017 09:08

Post by elkjiaer is back »

Stadion wrote: Wed Jan 24, 2024 08:22 LINK TO SPREADSHEET: https://docs.google.com/spreadsheets/d/ ... XcSfrTFvk/


Hi everyone,

I've decided to revive this spreadsheet for one purpose: simulating the qualification rounds using ClubElo values of all clubs.

- Access list and assumptions around what clubs start where are the same as @eye uses for his excellent spreadsheet

- Draw logic is the same as suggested by @amirbachar up-thread. Generally speaking, if a club on either side of the seeding line has a higher Elo than a majority of the clubs on the other side of the seeding line, it will progress to the next round. Or, if a seeded club has a lower Elo than the majority of the unseeded clubs, it will be replaced by (drawn against and lose) the club with the highest Elo on the unseeded side.

- To elaborate more on the previous point on draw logic with a couple of examples:
- In CLQ2, all seeded clubs are expected to progress except Vikingur, because it has a higher Elo value than only 3 clubs on the unseeded side. HJK also has a 'low' Elo value (in fact lower than 5 clubs on the unseeded side), but its Elo is still higher than the majority of the unseeds. Hence statistically it's more likely to be drawn against a team with a lower Elo. Hajduk on the unseeded side has a higher Elo than 7 of the seeded clubs, hence will take Vikingur's spot ("be drawn against them and win") and progress to CLQ3.
- In CLQ3, all seeded clubs except Slovan have a higher Elo than all unseeded clubs, and hence will progress. However ALL unseeded clubs except HJK have a higher Elo than Slovan. Hajduk has the highest Elo which is used as a tiebreaker, and Hajduk progresses (by being 'drawn' against Slovan and winning).

- Clubs in bold are projected to progress to the next qualifying round. I'm showing Elo values instead of coefficients throughout the qualifying rounds, but coefficients in the league phase. Though clubs are always sorted by coefficient unless otherwise stated (e.g. in rounds where there is no seeding)

- The spreadsheet does NOT update automatically -- it takes me a couple of seconds to update it manually, and I plan on doing so frequently. In the meantime I'll try to work out of way to automate it.

- The "ClubElo Sim GS Forecast" tab shows how many guaranteed and forecasted (based on the spreadsheet methodology) league phase spots each league has.

Keen to hear any feedback, and thanks to @eye for doing a lot of the legwork here.


LINK TO SPREADSHEET: https://docs.google.com/spreadsheets/d/ ... XcSfrTFvk/
Hi, first of all let me congratulate you for this excellent spreadsheet!
Secondly I have 2 suggestions for improvement, a very small one and a bigger one:
1) I might have overlooked it, but it would be helpful to add a cell showing when was the spreadsheet last updated
2) I love Clubelo ratings and I have been using them in my simulations for several years now. I have also spent lot of time comparing Clubelo with Opta ranking and EuroClubIndex in terms of predictive performance. But..Clubelo only covers 27/55 of domestic European leagues. When I had to simulate the qualifiers for the current season, I had to retrieve Elo ratings for all 196 teams (super easy with the API) but only 92 had a reliable Elo rating (less than 50%). A long story short: Clubelo is surely the best when it comes to teams from stronger countries, but it becomes much less useful otherwise as it is the case for qualifiers. I would therefore recommend you to test either Euroclubindex or Opta Power Rankings instead since they cover for all teams and for all competitions. But retrieving ratings from those website is far from easy.
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

Thank you for the kinds words, and feedback!

1. The spreadsheets updates automatically (league standings, club coefficients, clubelo values) on a daily basis so I don't know how meaningful it would be to show when it was last updated.

2. First off, is your issue with the Elo values for lower ranked leagues that:
- Elo values are missing entirely for some qualified clubs, or that
- The existing values (e.g. for the 3-4 clubs that there's data for) are inaccurate?

In case of the former, I am 'solving' this by assigning a value to clubs that are missing an Elo value. This is done by taking the average of 1. the median elo of all clubs (from that league) that we have ClubElo values for, and 2. the Elo value of the lowest ranked (in Elo terms) club of that league.

E.g. for VPS Vaasa (the club with an assigned Elo value that progresses the 'furthest' - to ECL'Q3), its Elo of 1,230 is derived from the average of the median Elo of all Finish clubs (1,252) and the Elo of the lowest ranked Finish club (Haka = 1,208).

Any club with an Elo value in Orange colour has been assigned its value based on the logic above.


If your argument is the latter - that clubs from lower leagues have Elo values that are inaccurate, then I agree that perhaps ECI or Opta are better sources!
elkjiaer is back
Posts: 271
Joined: Fri Jul 28, 2017 09:08

Post by elkjiaer is back »

Stadion wrote: Thu Mar 07, 2024 05:12 Thank you for the kinds words, and feedback!

1. The spreadsheets updates automatically (league standings, club coefficients, clubelo values) on a daily basis so I don't know how meaningful it would be to show when it was last updated.

2. First off, is your issue with the Elo values for lower ranked leagues that:
- Elo values are missing entirely for some qualified clubs, or that
- The existing values (e.g. for the 3-4 clubs that there's data for) are inaccurate?

In case of the former, I am 'solving' this by assigning a value to clubs that are missing an Elo value. This is done by taking the average of 1. the median elo of all clubs (from that league) that we have ClubElo values for, and 2. the Elo value of the lowest ranked (in Elo terms) club of that league.

E.g. for VPS Vaasa (the club with an assigned Elo value that progresses the 'furthest' - to ECL'Q3), its Elo of 1,230 is derived from the average of the median Elo of all Finish clubs (1,252) and the Elo of the lowest ranked Finish club (Haka = 1,208).

Any club with an Elo value in Orange colour has been assigned its value based on the logic above.


If your argument is the latter - that clubs from lower leagues have Elo values that are inaccurate, then I agree that perhaps ECI or Opta are better sources!
Well both issues you mentioned are strictly related. Clubelo is basically doing the same as you do, trying to assign an Elo rating to those teams which are not in the database already, by looking at the Elo ratings of other teams in the same country. This is quite an approximation, but le´s say we can live with that.

The other issue may be more relevant. Teams from lower leagues only play a very few european matches and therefore their Elo is changing super slowly over time. Moreover non considering domestic competition makes it very difficult to access the relative strength of teams from the same league. This becomes very clear when you compare Clubelo with OPTA for example. Essentially it is the same as trying to rank european teams only by using Uefa Coefficients. And the fact you are using Clubelo ratings instead of Uefa coeff. to predict teams to progress to next round means you clearly know why Elo is better :-)
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

So in terms of data availability, Opta definitely wins out over ClubElo since they seem to provide a rating for pretty much any club out there.

Are Opta's rankings more accurate and reflective of actual club strength though? Would be interesting to do a comparison for clubs that we have a decent 'subjective' track record for (i.e. clubs that participate in qualifying rounds year after year and we know how they typically do) and compare that with ClubElo and Opta ratings respectively.
elkjiaer is back
Posts: 271
Joined: Fri Jul 28, 2017 09:08

Post by elkjiaer is back »

Stadion wrote: Fri Mar 08, 2024 03:51 So in terms of data availability, Opta definitely wins out over ClubElo since they seem to provide a rating for pretty much any club out there.

Are Opta's rankings more accurate and reflective of actual club strength though? Would be interesting to do a comparison for clubs that we have a decent 'subjective' track record for (i.e. clubs that participate in qualifying rounds year after year and we know how they typically do) and compare that with ClubElo and Opta ratings respectively.
Some time ago I compared the ratings of Clubelo, Opta and Euroclubindex for all 96 teams partecipating in european group stage phase .
You can find it here: https://x.com/lorenzo_carli83/status/17 ... 95194?s=20
Please note that Opta does not provide directly their Elo ratings, but they transform and convert them into a 0-100 scale . So the strongest team has always a rating of 100 , and all other teams are scaled in comparison to the strongest and weakest.
To compare the 3 ratings I had to apply some complex conversion into Elo scale, a long story short:
For about 1/3 of teams the agreement is really good (Elo range<=20), for 2/3 is lower than 40 elo points, then we have 10 teams whose ratings vary a lot. Biggest difference is Ballkani (Kos): its Elo rating is about 1200 for both Clubelo and ECI, and almost 1400 for Opta.
A team having an outstanding season domestically will inflate its Elo rating (see Klaksvik or Zrinjski last season ), so inter-league adjustment must be performed wisely, unless we really believe,as Opta does, that only Cukaricki and Breidablik were weaker than Aberdeen.

if you need to run simulations then only option is to use Clubelo , if you just need to compare two list of teams and decide which ones progress to the next round , then I would definitely use Euro Club Index.
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

[Update]

https://docs.google.com/spreadsheets/d/ ... XcSfrTFvk/

The spreadsheet has now been updated as below:

- All access list and rebalancing changes based on UEFA's latest regulations (and based on @eye's excellent spreadsheet.

- Losers relegated to the tournament below take the lower coefficient of the pair: now also for CHQ1->ECLQ2cp, CHQ2->ELQ3cp, CHQ3->ELPO, ELQ3cp->ECLPOcp.

- Seeding based on coefficient (either actual or inherited) is now used across all rounds and tournaments

From a group stage participation perspective it yield the following forecast:
Image
Stadion
Posts: 389
Joined: Sat Mar 31, 2018 16:13

Post by Stadion »

Question regarding how coefficients are inherited across multiple qualification rounds:

- Partizan Belgrade (coeff 25.5) starts in NCQ2 where they lose to Servette and inherits their coefficient of 9.0 into ELQ3. They are now unseeded in ELQ3, based on Servette's coefficient.

- Nordsjelland (coeff 7.5) starts in ELQ2 where they win against Cluj and inherits their coefficient of 26.5 into ELQ3. They are now seeded in ELQ3, based on Cluj's coefficient.

- Nordsjaelland draw Partizan in ELQ3 and wins. What coefficient to they have in EL Play Off? Partizan's of 25.5 (even though Partizan came with a lower coefficient from a previous round)? Their own of 7.5? Surely not Servette's or Cluj's?
Post Reply