On Tournament Design
A moderate amount of debate has existed for some time in our community about tournament design. A number of criticisms are leveled at the current set of procedures for conducting a tournament in existence. My goal in writing this is to aggregate these voices and put forth my proposed alternative to current procedures.
Defining the ideal tournament design
Before beginning, I feel that it is probably a good idea to look at just what an ideal tournament should be, in order to compare current and potentially future practices against this mold.
As such, I feel that an ideal tournament should:
- Be fair — the ideal tournament should provide to competitors a level playing field to the furthest extent possible.
- Be reliable — the ideal tournament should return the absolute best competitors in a given pool of fellow competitors as the top placers. The ideal tournament design should also establish a way for these top competitors to prove their position by their play. As such, each and every game in the ideal tournament should be meaningful and contributory to the aforementioned goal.
- Be entertaining — the ideal tournament must be conducted in a way that is meaningful and entertaining for the spectating community, while still maintaining the integrity of the previous two points.
- Be efficient — the ideal tournament should be easily explained and understood. It is also important that the ideal tournament be reproducible and able to be used by more than one event series or organization. It follows that once an ideal tournament design is achieved, all events seek to employ or at least emulate its design. To achieve this, the ideal tournament design should avoid excessive complexity.
A survey of current tournament practices
The single-elimination tournament
The most basic tournament design is the single-elimination tournament. In a single-elimination tournament, the competitors are paired against each other in a bracket. Winners of each individual match in a bracket move on, while the losers of each individual match are immediately knocked out of tournament contention.
The simplicity of the single-elimination tournament is hedged against by a number of problems. First, single-elimination tournament design can run against the first and second tenets of an ideal tournament: fairness and reliability.
The position of any competitor within a given elimination tournament bracket at its outset is defined as a seed in that tournament. The determination of how seeds are distributed amongst a given competitor pool is the classic problem of any elimination tournament, and is most acute in the single-elimination tournament, because of the unforgiving nature of the winner-take-all approach inherent in its design. As such, tournament organizers must make a conscious decision: should they attempt to determine a fair and accurate seeding of competitors into the bracket, or should they establish seeds randomly?
Attempting to seed competitors based on past results can run into a number of problems. With the ubiquity of tournaments in the modern competitive field, how are prior results of a given player to be accurately compared with the prior results of another player? Tournaments and events in the current structure are so varied, and the establishment of seeds based on these results inevitably results in some biasing. The only really viable way for a tournament organizer to claim impartiality in decisions of seeding is if only prior results in tournaments conducted by the tournament organizer are used in the calculation.
Random seeding has its own set of problems. In seeking to be fair, random assignment of seeds into a single-elimination tournament creates unbalanced brackets and severely limits the ability of any tournament conducted in this way to produce the best of the competitor pool as the ultimate winner. Random seeding introduces the perverse possibility of the top competitors meeting in the early stages of the tournament and being knocked out, while relatively weaker players may have an easier road to the finals.
The harsh nature of single-elimination also significantly reduces the number of potentially broadcasted games an eventual loser plays in a tournament. In the field of competitive gaming, this means a significant lack of face time and return on investment present in the system. A player can be 'one and done' and out of sight very early in a tournament.
The double-elimination tournament
The structure of the double-elimination tournament seeks to address some of the issues inherent with the single-elimination tournament and discussed above. A double-elimination tournament bracket is divided into two distinct parts: an upper (winner's) bracket, and a lower (loser's) bracket. All competitors are initially included in the winner's bracket, and match play proceeds similarly to a single-elimination tournament with one significant difference. In a double-elimination tournament, competitors are not immediately knocked out of tournament contention after a single match loss. Competitors who lose a match in the winner's bracket are placed into the loser's bracket, where they are given a second chance in the tournament in what is essentially a single-elimination procedure where they meet other competitors who have been knocked out of the winner's bracket. The distinction between the winner's and loser's bracket ends at the grand final, where the last competitor left in the winner's bracket meets the last competitor left in the losers bracket, determining the ultimate winner of the tournament.
There are a number of advantages to double-elimination tournament procedure. First, the forgiving nature of the tournament design hedges against the potential unfairness present in standard single-elimination design, especially if bracket seeds are inaccurately determined or are determined at random. Second, the existence of the loser's bracket builds the third place match directly into the design of the tournament, where it is an extra, seemingly 'meaningless' match in regular single-elimination procedure. Third, the problem isolated above of potentially reduced return on investment is much mitigated as players have a second chance in the tournament after their initial match loss in the winner's bracket.
However, double-elimination procedure also has a number of problems inherent in its design. The greatest problem confronting double-elimination procedure is the sheer number of games required to conduct it. This problem becomes larger as the number of tournament entrants increases, severely reducing the scalability of any double-elimination tournament. This large number of games required also works against fairness as well, as the competitor coming out of the lower bracket will almost certainly have played a substantially larger number of matches against their winner bracket counterpart. This difference in number of games played makes the conclusion of a double-elimination tournament, especially ones conducted 'offline,' tests of endurance and strength, not necessarily in-game skill.
This endurance problem is supercharged when fairness is articulated as justifying artificial intervention by tournament organizers. The most visible example of this is the 'extended series' rule employed by Major League Gaming. The 'extended series' rule establishes that if two competitors are matched up against each other for a second time within a given tournament, the match is extended into a larger best-of-X series with the match score from the previous game being taken into account. As such, if two competitors met previously in a tournament, with the score being 3 to 2, upon their second meeting, the results from this best-of-5 translate into the new match beginning at 3 to 2 in a best-of-11. The merits of 'extended series' have been debated to death, but it is clear that its existence remains controversial in our community, and its existence is made possible because of double-elimination procedure.
It is important, I feel, to make a distinction between 'extended series' and the normal conduction of double-elimination here. In the grand finals of a double-elimination tournament, the fact that the competitor coming out of the loser's bracket has two win two best-of-X matches in order to win is often confused by many in our community as being part of the 'extended series' rule. This is incorrect. The fact is the winner of the loser's bracket, having lost a match for the first time in the grand finals, is thus knocked into the same loser's bracket within which the competitor coming from it was given a second chance in the tournament. It would be fundamentally unfair to the winner's bracket competitor to not be given that same second chance that was afforded to the loser's bracket competitor in the first place. To deny the winner's bracket competitor this second chance is to impeach the integrity and suitability of double-elimination procedure in the first place. This is where this aside becomes more pertinent to the subject of this writing. I do agree that this presents a significant challenge to the use of double-elimination on the part of tournament organizers. It is fashionable in tournament design to increase the numbers of games required to win an individual match at certain stages during conduction of an elimination tournament. Grand finals of a tournament are oftentimes long series under best-of-5 or greater conditions. This can make the grand finals of a tournament inordinately long and potentially significantly disadvantageous for the competitor coming out of the loser's bracket. Take, for example, Team Liquid's performance at the IPL Team Arena Challenge 3 Main Event, where that team had to win two best-of-9s, an endurance feat conceded even by team management as an improbable event.
If timecoding does not work, skip to 3m 40s in the video
Finally, there is the elephant in the room regarding the appeal of broadcasting loser's bracket games in the first place. The large number of loser's bracket games inherent in the double-elimination system ensure that a good deal of loser's bracket games will be shown on any tournament broadcast. It seems to be hard, however, to make loser's bracket games compelling enough for the viewership unless the players in the loser's bracket are popular, or the loser's bracket is finally concluding and the grand final approaches. As such, double-elimination procedure places stress on the broadcasters to make the long process of concluding a double-elimination tournament compelling for the viewership through the construction of story lines.
Groups into elimination
The establishment of a group stage in tournament design, on its face, seeks to ameliorate the inherent problem of determining fair and accurate seeding for competitors in an elimination tournament.
In typical group stage procedure, a given competitor pool is divided up into one or more groups of competitors. Competitors in these groups play each other in round-robin procedure, where each competitor in a group plays all of the other competitors in a group once. At the conclusion of the round-robin, competitors within groups are ranked, and from these rankings players may be seeded into or eliminated from an elimination bracket to follow.
Group stage into elimination is rapidly becoming the procedure of choice for tournament organizers in our community. Yet, its current articulation is imperfect for a number of reasons. First, the procedure for which the division of competitors into groups is conducted is oftentimes unknown to the public and, not to mention, the competitors themselves. This division can be done in three ways:
- Group selection — in group selection, the competitors themselves decide the group composition, usually with certain restrictions or through a draw of some kind. There are two chief advantages to this procedure: first, the process is transparent to the players, and second, group selection offers the tournament organizer the opportunity to develop compelling pre-show content for their event
- Tournament choice — in tournament choice, the process is rendered completely opaque, and the composition of groups is determined, perhaps arbitrarily, by the tournament organizers. The most debilitating disadvantage for this procedure is the existence of a potential for corruption and matchfixing, as the opaque nature of this procedure allows decisions to be made which can influence tournament results in a way in which a limited number of people have knowledge that is unavailable to most prior to the start of an event.
- Random assignment — group compositions, when using this procedure, are determined randomly. This should be weighed against group selection as the only two legitimate ways of determining group composition within the options outlined here.
Ironically, it is the existence of multiple groups themselves in current tournament fashion that undermines current group stage into elimination tournament procedure. Let me explain.
Group subdivisions are an imperfect sampling of what is already a limited competitor pool. The existence of a 'group of death,' in which the best competitors in a tournament are seemingly placed within a single group and will thus knock each other out is a persistent problem. Where a competitor ends up being placed into groups also matters substantially, because the racial composition of a group in StarCraft II competition drastically changes play style and competitive level. For example, if a competitor whose weakest matchup is the mirror matchup, and is placed into a group completely consisting of said competitor's mirror races, the chances for this competitor to make it out of groups are predictably slim indeed. Contrast this example with a different situation, where a competitor is placed into a group consisting of other competitors whose played race is advantageous to said competitor; the chances of said competitor making it out of groups is predictably high indeed. As a final point, the incongruity of play and strength inherent in multiple groups make it hard to accurately determine of results within one group accurately represent the strength of the competitors within specific group compositions.
The best way to eliminate this problem with the group stage is to only have a group of one, with the entire competitor pool in a single group. The most limiting factor preventing the implementation of this is scalability: as the number of tournament entrants increases, the number of games required to conduct a round-robin group procedure increases as well. At a certain point, the number of games required would make the group stage massively inefficient and undesirable from the standpoint of a tournament organizer.
Another way to mitigate this issue is to have a very large number of entrants into a tournament, such that groups can be made up of a suitably large number of competitors whose composition is much more balanced than what traditional groups of four can offer to a tournament. Time constraints and number of games required to be played are dealt with articulating and presenting this group play as an actual 'season' as part of a league. What the North American Star League does, in this regard, is the most successful implementation of such a procedure in the current mold of tournaments and events in our community. This procedure is what makes the North American Star League an actual league, and an overwhelming majority of their actual content is the broadcast of each and every single game in this group play, culminating in their offline finals at the very end.
There is also a problem which, I feel, is similar to the loser's bracket problem with double-elimination tournaments. Group stages can often times feel extraneous and unimportant relative to the elimination bracket stage of a tournament. Organizations that draw out group stages into an actual league, such as the North American Star League, largely eliminate this problem, in my view, by making the group stage a compelling reason as to why you are tuning into their broadcast into the first place. I've also heard of some tournaments held with competitive gaming titles that are not StarCraft II skipping the broadcast of the group stage entirely, showing only the elimination bracket in their broadcasts. This might be an option that is worth exploring perhaps in our community as well. Ultimately, I feel, this problem, as well as the one mentioned previously about double-elimination, can largely be handled by the skill of the people a tournament organizer has working for them in making compelling story lines and content for the viewership to consume.
'Bracketception' is a word I've just made up. It refers, however, to tournament design practices that go from elimination bracket into elimination bracket into elimination bracket. Take, for example, the 2012 MLG Pro Circuit Summer Championship in Raleigh held last month. In this tournament, an extremely large double-elimination open bracket led into a 'group play' that was really just another double-elimination bracket of invites plus 'seeds' from the open bracket play and then, after that, concluded by leading into yet another double-elimination championship bracket to decide the winner at the very end.
It is hard to see the advantage this tournament structure has over the more traditional tournament structures available to the tournament organizer, unless they consider making Liquipedia's job in covering their tournament brackets that much harder an advantage. At any rate, fairness is not something necessarily demonstrated or introduced into a system through added complexity.
The Swiss tournament is not a common sight in our community, but I will mention it here because it influences the design of my chosen alternative to the tournament structures listed so far in this writing.
In a Swiss tournament, competitors are paired off to face against each other over several rounds of competition. After every round, winners are paired against other winners of the round previous, and the same goes for the losers of each round as well. No two competitors are allowed to meet more than once in a given tournament. At the end of the rounds, the top ranked competitor based on wins and perhaps other tiebreakers is declared the winner, with other placements in the tournament going to the respective competitors and their ranks determined from in-round play.
The main advantage of this procedure is that a clear winner is established by the tournament, as well as a clear ranking of the competitor pool and thus a clear loser too. The number of matches required to establish this ranking is generally equal to or only slightly more than a traditional elimination tournament.
The main disadvantage to this procedure is the sheer lack of hype and excitement inherent in it. Elimination tournaments, by their very nature, clearly show the competitor pool being whittled down to a chosen few; the elimination part of elimination tournaments gives these tournaments an added excitement outside of the games being played and the competitors playing them. Swiss tournaments, on the other hand, have no clear winner until the very end, generally. It is also hard to construct compelling story lines for a competitor's run in a Swiss tournament.
An alternative: introducing Tabulation procedure
For the remainder of this writing I will describe a tournament procedure I have had much experience with over the last couple of years and of which I feel does a very good job harnessing the advantages of the above systems (except 'bracketception,' of course) while mitigating the attendant negatives of the use of said systems independently.
'Tabulation' is a word borrowed from the world of the academic debate activity in the United States. In essence, the tournament design I describe here is a modified Swiss tournament where rounds are cut off after a certain point, determining the composition of and seeds within an elimination bracket. 'Tabulation' is the process conducted by 'tab' rooms at debate tournaments, where the results of each round are tabulated in order to determine pairings for the next round in the preliminaries.
The goal of the preliminary rounds (prelims) is to determine who gets to make it into the elimination bracket after their conclusion. Not all competitors who enter into the tournament will make it to the elimination stage (elims); depending on how many rounds a tournament organizer wants to conduct, only the top 32 to 16 competitors make it (or 'break' out) into the elimination brackets.
The prelims are essentially a group stage, with all of the competitors placed into a single group. The two rounds of pairings are randomly assigned (and can be released to the competitors and the public before the start of the tournament), with following rounds conducted as so:
- Competitors play the same or similarly ranked competitors only — this essentially means that certain bracket classes will be established through group play. Thus, a competitor with a record of 2-0 is part of the 2-0 bracket class and will end up being paired against another competitor of the same rank. If there is a non-even number of competitors in a certain bracket class, the top ranked competitor from the bracket class immediately below the one in question will be 'pulled up' to the higher bracket class in order to fill the space.
- High-high power-matching procedure — what this means is that top rank in a certain bracket class will meet second rank in a bracket class, and so forth. The point of the preliminary stage is to prove whether or not competitors deserve their seed at the end.
- Tiebreakers — in the following order: (1) head-to-head matches won, (2) head-to-head games won, (3) total matches won, (4) total games won
These rules ensure that the preliminary stage is made up of matches that accurately determine the skill of a given competitor relative to their rivals at a given tournament.
The elimination bracket procedure for this system is a fairly-straightforward single-elimination bracket seeded based on the tabulation rankings established in preliminary play. Seeds will be power-protected: thus 1st seed hits 64th, 2nd hits 63rd, etc. This gives an incentive to players in the preliminary stage to ensure they get the highest seed possible.
Why single-elimination versus double-elimination? I'm of the opinion that once solid seeds have been established by a tournament via preliminary play, there exists no need to add the complexity and second chances to competitors given in double-elimination procedure. Simply put, if you manage to be top seed in this tournament, but you manage to lose to the 64th seed in the first round of elims, you deserved the loss because based on preliminary play, the 64th seed's victory against you in elims was a huge upset.
To avoid the issues of having to determine rankings throughout the elimination bracket, awards and prizes should be given out (past first and second place) to competitors who are semifinalists, quarterfinalists, and so forth equally based on their final respective elimination stage reached. If this is not an option, consolation matches should be played outside of the broadcast, in general.
Advantages and disadvantages of Tabulation
The principal advantage of this procedure, in my opinion, is the ability to demonstrate, through tournament play, a fair and accurate ranking of the skill of all tournament entrants at a given time. This advantage does not just benefit the tournament organizer in developing fair seeds into an elimination bracket, because it would fundamentally alter the way we can determine overall player skill throughout a long period of time. If multiple major and premier tournaments adopted this procedure, the ability to analyze aggregate and relative player skill will become much more reliable.
It also simplifies tournament structure, and above all gets rid of the perceived need to have things such as double-elimination second chances and the 'extended series' rule based on the inadequacies of a given tournament structure.
There are also a number of disadvantages to this procedure, however. Chief among these is the actual tabulation of the preliminary matches. A computer program or spreadsheet would probably be the most efficient and accurate means by which to calculate the tabulations. Hand tabulation of tournament results will end up bogging down the system. Along the same vein, all matches in a certain round must conclude until tabulation of results and the creation of pairings for the next round can be released to the competitors and the public. Stalemated games, connection drops, and their like in one individual match might just end up influencing the entire tournament, bringing it to a potential standstill.
The system of tabulation might also be too complicated for easy explanation to the viewership of a tournament. However, I feel that this disadvantage is at least already present in current groups into elimination tournaments. With apt explanation on the part of the public faces of the tournament, I feel like this should not actually be much of a problem.
In terms of entertainment, I know for a fact that the limitation of knowledge as to who is hitting whom in the next preliminary stage is a moment of excitement (and sometimes dread) for many in the community from which I am transplanting this tournament procedure. The pairings themselves, and their eventual release, can become part of the tournament story line, bringing tension to the event and making it more compelling for the viewership. The ability to determine relative skill of competitors at any given time during the preliminary proceedings also enables tournament organizers to strategically deploy their broadcasters onto games that will be certain to be good.
This limited information also acts as a significant hedge against players attempting to game the system and not play their best based on knowledge of who they are going to hit in the next rounds (and especially in the elimination rounds). This situation has many a time mired a tournament in controversy, and is best avoided whenever possible through the structure of the tournament itself.
Tabulation also provides a guaranteed number of rounds for each entrant: they will at least be playing all rounds of the prelims, meaning there is no symptom of 'one and done' or 'two and done' that is possible in the elimination tournament structures. This guaranteed number of rounds (translated into a typical tournament, this would be about two days of still being in actual contention) assures a certain level of return on investment for competitors and their teams, making it more palatable for their sponsors.
Finally, Tabulation procedure gives tournament organizers another significant advantage in the broadcast of preliminary group stage rounds compared to traditional groups to elimination tournament structures. When the last few rounds come along, broadcasters can focus on the middle of the pack, those on the cusp of being able to break out into elimination bracket contention, rather than focusing only on those who will for sure make it into the elimination rounds.
Whew, this has been quite a long writing, and the first of its kind from me here on Team Liquid at least. Let me know what you think about my analysis and Tabulation procedure itself in the comments below. I'd appreciate it if you read the entire post before replying, and please try not to get hung up on debating the merits of extended series.
Previous blog: Taeja Wallpapers
Next Blog: Inflation and Exchange Rates