Statistics against GSL style group format

On December 21 2012 11:28 jmbthirteen wrote:
I really dont care if the player is 2-3 and against the other guy and advances. They aren't playing a bo5. Its two 2 bo3s. I think the GSL system is by far the best out there. I'm glad more and more tournaments are using it.

On December 21 2012 11:31 courtpanda wrote:
1) the player who wins is better

On December 21 2012 11:20 chuky500 wrote:
A thread in April discussed whether the GSL group format was flawed or not and at the time 82% people judged that the format was fine. Since then many popular tournaments have adopted it as it is shorter, easier to run and provides more drama. As more groups were played this way we've seen more games leaving that awkward feeling when there's a rematch and the player qualified isn't the one with more wins, such as the WCS finals when Hero had qualified over Stephano despite having a record of 2-3 against him. It happens because when 2 players meet again, the result of the first bo3 is discarded. This means to qualify one player has to beat his opponent in 2 bo3 while the other one qualifies with a 1 bo3 draw.

Since the GSL format feels fine otherwise I've decided to count the games where a rematch happened. I didn't count groups in bo1 such as older Code S, neither groups where a player forfeited. Everything was counted by hand. 2012 GSTL season 2 and 3 were included.

Here's a list of the tournaments included :
+ Show Spoiler +

Here are the results compiled :

169 groups were counted.
85 groups had 2 players meet twice.

After the rematch :
41 times the qualified player won both bo3.
44 times the loser of the 1st bo3 has qualified by winning the 2nd bo3.



Regarding games where a player qualifed by losing the first bo3 and winning the second one :
14 times the qualified player had more wins. (3-2)
20 times both players had the same number of wins. (2-0 each or 2-1 each)
10 times the qualified player had less wins. (2-3)



First observation : when you rematch players, the loser of the first bo3 has 52% chances of qualifying. It's close to even.

Second observation : when nobody won both bo3, 68% of time the GSL format is wrong about qualifying the concerned player. Either because both players were tied or because the qualified player had less wins.

Which gives us, in case of a rematch :
41+14 = 55 players qualified for good reasons (winning both bo3 or having more wins)
20+10 = 30 players qualified for bad reasons (having the same number of wins of less)



The conclusion is in the case of a rematch the GSL format gives wrong results 35% of times.

If you want explanations about why it isn't fair to discard the first bo3 you can look at the April discussion.

First, contrary to popular belief, a GSL group isn't an exact double elimation. In a double elimination the loser bracket is reversed so 2 players can't meet in the next 2 games, while it can happen in a GSL group.

Then, here's the "unfair" finished GSL style group :
A 2-0
B 2-1
C 1-2
D 0-2

C won the first bo3 against B, but because of this B had an easier group than C.

Basically you give an advantage to the loser B by matching him against D, the group loser, while the winner C of the 1st bo3 gets to play A, the group winner. But it isn't actually fair either for A who had to play C and D the last 2 players of the group while C had to play A and B the first 2 of the group.

If you compare a regular round robin group with a GSL style group, in a regular round robin C wins the tie breaker against B in case of a tie, while in a GSL group format you give the loser B several advantages because he gets an easier 2nd opponent and gets the chance to redeem himself. When there's a rematch, it can be seen as a tie breaker, except one player qualifies by getting a draw 1 bo3 to 1 bo3, while the other player has to win 2 bo3.

How to fix this :
First, the GSL group format is perfectly fine until two players have to meet again. This is the only case where the format should be adjusted.

-I believe the best way to fix the format is instead of rematching you play the missing games. B and C don't rematch but instead A plays B and C plays D. You end the GSL group like a regular round robin. The advantage is you count the actual results and don't make assumptions about who is better than whom. The downside are it takes more time to organize, it doesn't prevent from 3 way ties and since D is already eliminated he may not be playing at his full potential and he can influence whether his opponent is qualified or not.

-Very soft extended series : you rematch B and C in a bo5, where you give the winner of the 1st bo3 1 point to start.

-MLG style soft extended series : you rematch B and C in a bo7 while starting from the result of the first bo3.

-Double elimination style hard extended series : instead of thinking of games on their own you think of a game being an indivisible bo3. You rematch B and C in a bo3. If C wins he's qualified (he won 2 bo3). If B wins (1 bo3 to 1 bo3) you make a new bo3 where the winner is qualified.

On December 21 2012 11:35 Canucklehead wrote:
So what you're arguing is for gsl to adopt extended series to fix this?