Forum

xml	No replies possible in the archive

Author: bert.kassies
Date: 02-06-2007, 12:23

Last year some observations were made on the Seeding groups for UEFA R1. Executor published a method how UEFA deals with two teams from the same country in one group, and Auke published some observations with respect to the spread of CL-QR3 losers over the groups. I tried to use these methods to predict the UC-R1 groups in 2005, but it seems much harder than in 2006.

The question is: Who can improve the "executor-Auke" method" such that the method can explain the UC-R1 groups of 2006, 2005, and 2004, and can be used to predict the UC-R1 groups of 2007?

Use the rankings of 2006, 2005, 2004, and the groups of 2006, 2005, 2004.

Facts:

1) Initial distribution of clubs (1-80 based on ranking) over the 8 groups:

pos  G1  G2  G3  G4  G5  G6  G7  G8
  1   1   2   3   4   5   6   7   8
  2  16  15  14  13  12  11  10   9
  3  17  18  19  20  21  22  23  24
  4  32  31  30  29  28  27  26  25
  5  33  34  35  36  37  38  39  40
  6  48  47  46  45  44  43  42  41
  7  49  50  51  52  53  54  55  56
  8  64  63  62  61  60  59  58  57
  9  65  66  67  68  66  70  71  72
 10  80  79  78  77  76  75  74  73

2) Changes are applied only at the same row (or the same position in the groups)

3) The title-holder is not treated as top-seed but ranked by coefficient

4) No country can have two or more teams in the same group

Observations:

1) Besides the country constraint (fact 4) UEFA applies more changes. These changes seem to be related to a more or less evenly spread of CL-QR3 teams over the groups. See the comments by Auke.

2) When changes are required for one or two groups they are applied by swapping teams at the same row. For a single group the swap is with a team on the same row with a lower coefficient (if possible). See the method of executor. But this method can e.g. not explain the multiple changes on row 8 in 2005.

3) A comparable initial distribution (fact 1) and only changes at the same row (fact 2) also seem to be used for the groups in UC-QR2. But the changes here are hard to explain (sometimes the changes put three clubs of the same country in the same group while not required, etc.)

Author: Francisco
Date: 04-06-2007, 15:15
Edited by: Francisco
at: 04-06-2007, 15:16

Are you sure there's a logic behind the scheme?

I followed your instructions, and then tried a few methods to find the final groups for 2006 and 2005. Let's say that 2005 is impossible

.2006, I just didn't understood a few changes (especially, the Ajax/Feyenoord switch) but I'll try to justify them as follows...

So what I tried and didn't work:

1 - from top to bottom, i realized that changes must start at the bottom and climb their way to the top. The main reason for this, is that after switching 3 or 4 times in the seeded zone, practically all groups were perfect, but then I found out that the unseeded teams although acceptable at this stage were very different from the UEFA groups.

2* - from bottom to top, i realized that most switches use this technic, but there are still a lot of unresolved and apparently unreasonable switches.

(3 - I also tried the group by group approach, but this was really bad)

Using (Method 2)* I applied the ranking number to each team, and tried to justify some switches by the need of balancing group strength. At the beginning all groups add up to 405 (1+16+17+etc=405, 2+15+18+etc=405) and one of my theories was that even after having the first switches satisfying all rules (especially country protection), UEFA looked up to check if there were any groups too unbalanced (for example most groups were in the 403-407 range), but the truth is that one important switch happened! In their original positions (Ajax=404 and Feyenoord=411) Feyenoord's group is real easy (sort of

) while Ajax group is balanced. Thanks to the switch Ajax = 409 while Feyenoord 406. My opinion though is that instead of exchanging with Ajax(that originally had the balanced group) UEFA should have changed Feyenoord with other team (not Ajax, since 409 is also pretty easy), but I think due to fear of having to start the regrouping again, UEFA decided to give the easy group to Ajax instead of looking for another possibility.

So my suggestion to add a new logic to the grouping is to give to the 80 teams their rank (1,2, ... 80) and that rank is the number of points used to test group strength. At the beginning 405 is the sum for all groups but due to the logical switches (as identified by executor) UEFA's tries to balance the groups as best as it can.

Also in my opinion I don't think the CLQ3 teams influence the choices of groups. I think it was a coincidence because there's nothing in 2005 or 2006 that indicates protection for these teams...

Author: alexdcro
Date: 04-06-2007, 16:55
Edited by: alexdcro
at: 04-06-2007, 16:57

I looked over the 2004 initial distribution and i observed the following:
1) 39 (48.75%) teams were correctly placed using the proposed method
and 41 (51.25%) teams were not.
2) 24 (60%) seeded teams were correctly placed and 16 (40%) were not
3) 15 (37.5%) unseeded teams were correctly placed and 25 (62.5%) were not
4) there were 2 groups of seeded teams that were 100% guessed
5) there were 0 groups of unseeded teams that were 100% guessed
6) there were 0 groups of seeded teams with 0% teams guessed
7) there was 1 group of unseeded teams with 0% teams guessed
8) all the seeded teams on the 2nd row were 100% guessed
9) no row had 0% team guessed
10) no top-down nor down-top strategy is likely since neither row 1 nor row 10 is 100% guessed (their strategy must allow to come back to a row that was finished and than advance) yet row 2 was 100% guessed
11) there is a Middlesbrough - Millwall anomaly: Middlesbrough should has been in group 4 and so was Newcastle. Both teams were not in the group yet Millwall was in Middlesbrough spot. Why Middlesbrough did not keep it's place and was replaced by another English team?

Re: Contest to improve the "executor-Auke" method
Author: bert.kassies Date: 05-07-2007, 15:21	What hardware is used by UEFA? A table, and 80 pieces of paper each with club, country, and coefficient? Which are then layed on the table according to the initial distribution. Applying changes by swapping pieces of paper?

Re: Contest to improve the "executor-Auke" method
Author: abzpablo Date: 23-07-2007, 17:14	Let's get back to just sticking the names in a hat and seeing what comes out.