EPTTiebreakers pt. 2 - Page 3
| Blogs > motbob |
Heyoka
Katowice25012 Posts
| ||
|
mdb
Bulgaria4059 Posts
| ||
|
Otolia
France5805 Posts
| ||
|
DefMatrixUltra
Canada1992 Posts
On March 24 2013 05:38 motbob wrote: Game score should always be used as a tiebreaker before head-to-head. Already I can feel the angry posts being written. Let me explain why I think what I think. ![[image loading]](/staff/motbob/UKgroupA.png) In this group, ForGG and Babyknight were tied in match score at 3-2. Because of that tie, a tiebreaker had to be used. There were two pieces of information that the tournament organizers could have drawn on to make their decision on who to advance. The first piece of information is that Babyknight defeated ForGG 2-1. This implied that BabyKnight should have advanced over ForGG. The second piece of information was that ForGG's record against common opponents of ForGG/BabyKnight was 7-2. Compare that record to the 5-5 record of Babyknight. The final game score of the group, derived mostly from the record of ForGG and BabyKnight against common opponents, implied that ForGG should have advanced. So there are two conflicting pieces of information here. Depending on which piece of information is deemed more important, one of these two players should advance. I think that the game score is a more important indicator of who the better players is because it draws on a larger sample size. Variance exists in SC2. The best player does not always win the game, or the series. As with all phenomena with variance involved, it's always better to increase your sample size to see what is really happening. Tournaments choose the H2H tiebreaker as the one that takes priority because it "feels right." It feels just that a tie between two players should be decided by the games played between those two players. But if Player A has beaten Player B, it's very difficult for Player B to overcome that loss and do better than Player A in game score. ForGG had to obtain at least a +3 advantage in game score over BabyKnight against common opponents to overtake him. If ForGG had lost 0-2, he would have had to obtain a +5 advantage! (In this way, you can see that the H2H tiebreaker is actually included in the game score tiebreaker.) If a player who has lose the H2H match has overcome that loss in subsequent play, he deserves to advance. ForGG should have advanced over Babyknight, and players with better game scores should advance even if they have lost the H2H match which whom they are tied. Good post explaining the concept. The real key here - the real essence of the problem - is that given these two possibilities (H2H match or group score), you want to choose the best one. Neither method is a bad method - the issue is that one is better than the other. The fundamental problem with choosing H2H is that you are choosing to throw out so much additional pertinent information. Mathematically, it's very analogous to using an ELO-like system vs. using a "true" ladder. A true ladder will result in a new top player every time someone beats the top player, but it will give a poor indication (a single moment in time) of how close or far the top few players' performance is from each other. To look at it the opposite way, if H2H was the best tiebreaker, then round robins or any group format except elimination are inferior. If we could get more information about who the better player is from H2H, then having every group stage mechanism be a H2H elimination (loser leaves the tournament) would be the best possible system. If we take this logic even further, we might even could argue that the best thing to do to determine player performance would be to have an entire tournament of nothing but single elimination (1-0). The fundamental issue is that there is more information and less variance across more matches. A lot of people are bringing up the point that random circumstance can lead to one player being artificially favored by a group score. That's definitely 100% true. But does random circumstance have more variance in one match or over many? There are a large number of factors that can affect a single match, and singling out one of them is intellectually dishonest for the purpose of determining a system to find the best player. We have to trust in the group system because our only other option is to trust in an overall inferior system. | ||
|
ffadicted
United States3545 Posts
On March 24 2013 05:48 Plexa wrote: I actually feel the opposite, I can never understand why mapscore should be used over head to head in any instance. Starcraft is not a perfect game and there are many factors which influence the result of a game - blind countering, bad maps and so on. Players can drop a map just due to the random nature of SC2. Matches are Bo3 to eliminate this kind of variance and to encourage the 'better player' to win. Moreover, when players are eliminated they often mentally 'give up' which means that the map scores in the latter series with this player are more likely to be 2-0's - which influences the map score of player who are still in contention. This is yet another source of variation that should be taken into consideration. Choosing map score over head to head is not inline with this. Map score says that random variation should influence who advances not who was 'the better player' during the series. Since tournaments choose Bo3s over Bo1s, and choose to eliminate as much random variation as possible - they should be choosing H2H every time. I think you counter your own argument. H2H is 3 games, overall game score was potentially 10+. How is that not significantly better at eliminating the "randomness" you're talking about. | ||
| ||
