Power matching sucks: a theoretical offering

Stained Diviner · Post by **Stained Diviner** » Mon Dec 21, 2009 7:02 pm

Andrew, as I said, advanced stats are not as good as teams playing each other. Our system had teams playing each other instead of advanced stats.

Harry, my point was of course exaggerated. To put it without sarcasm: In power matching, seeding matters less than it does with pools. As I alluded to upthread, if everybody on the 9 seed team quit the day before PACE, and the coach rounded up a couple of freshmen to play the tournament who had never held a buzzer before, then the winner of the match between the 24 and 25 seeds would make the top bracket while the losers of the 10/23, 11/22, 12/21, 13/20, 14/19, 15/18, 16/17 matches would not make the top bracket even though the winner of the 24/25 match had done nothing at all to hint at the possibility that they were better than seeds 17-23. In fact, judging by the seeding, the winner of the 24/25 match is very likely to be worse than at least some and probably most of those teams. In other words, Matt's statement about robustness at the beginning of this thread is not only unproven, it is incorrect. (I have only proven it incorrect in this one case, and Dwight has only proven it incorrect in his one case--at this point I can say that I am confident that Matt is incorrect even though the proof Dwight alluded to does not exist at this point, though there are enough programmers who read these boards that we could have the proof if somebody laid down the parameters. If PACE used power matching, the bad 9 seed would lose its first match to the 56 seed, which would lose its next match to the 24 seed, which would lose its fifth match to the 13 seed, after which point the seeds would pretty well approximate what they should.) And, as Matt said, the point of these systems is about fairness--an unfair system is a bad system. It is true that tournaments are not all about winning, but they should still strive to use a system that is as fair as possible, which is what power matching does and what the point of this whole discussion is.

Yesterday, the teams we seeded 1, 2, 6, 8, 10, and 14 made the playoffs, which was bad seeding on our part. The worse seeds were able to make the playoffs because this system gives them more of a chance to prove themselves, and gives the better seeds more of a challenge to defend themselves, than pools. That's why Andrew's joke about picking the top eight teams pre-tournament applies more to pools than to power matching.

Stained Diviner · Post by **Stained Diviner** » Mon Dec 21, 2009 7:37 pm

Let me add one thing that would make Dwight's proposed experiment difficult to carry out--the natural number of rounds and advancing teams for pools and power matching is different. With 64 teams in pools, it makes the most sense to play seven rounds and then have either 16 or 24 teams advance. With 64 teams in power matching, it makes the most sense to play six rounds and then have either 7 or 22 teams advance. Discrepancies like this, and the fact that there are certain numbers of teams not handled well by power matching, make pools the better option in certain circumstances despite everything I've said upthread. In our case, and in many other cases, power matching is the better option.

Matt Weiner · Post by **Matt Weiner** » Mon Dec 21, 2009 8:09 pm

I think there's some sort of disconnect in how our minds are working in this thread. I don't care about "what happens if all the players on the 9th seed at PACE quit before the tournament" or "what happens if a TD puts the top 8 teams in the tournament into one bracket." I'm dealing with empirical real-world scenarios that can happen, did happen at New Trier, and do happen every time power matching is used. The fact that you think you can "prove" extreme cases in the abstract by setting several assumptions at odds with reality as axiomatic just goes to what I said before, which is that people who love power matching are more concerned about demonstrating their ability to perform mathematical sleight-of-hand than about tournament fairness in real-world conditions.

Another thing I don't like in this thread is the common bad-quizbowl-defending practice of pretending that no other tournaments exist. Instead of looking at the many tournaments which have run reasonable formats for 32 to 48 teams and seeing how they ended at a reasonable time, entered stats efficiently, and used a ranking system that wasn't a total screwjob, people come up with these wild scenarios about what will happen based on their own speculation in a vacuum and then use them as the basis to declare that such goals are impossible to achieve. This is, incidentally, another manifestation of "set the premises to things that range from controversial to outright untrue so that I can derive conclusions about questions no one is asking and then defend them due to their own internal logic" reasoning (henceforth to be referred to as "mathturbation") over the use of evidence.

Stained Diviner · Post by **Stained Diviner** » Mon Dec 21, 2009 8:56 pm

The reason I started using power matching is that I went to plenty of tournaments with 32 to 48 teams that used pools and didn't like them. They are poor in practice and in theory. Teams show up missing players, or teams show up with all their good players for the first time on a regular basis, leading to unfair pools. Or hosts distribute teams in pools poorly. The list goes on.

As to your great injustice at New Trier, you continue to be wrong about that. Let's take this one round at a time, using the team I believe you think got screwed the most, Stevenson, and the team you think got gifted the most, OPRF:

In Round 1, Stevenson had the 5 seed, so they played the 28 seed Maine East and won. OPRF had the 8 seed, so they played the 25 seed Niles West and won. So far, nothing confusing or controversial.

In Round 2, Stevenson, still having the 5 seed, played Naperville North. Naperville had started as the 21 seed, but they advanced to play Stevenson because they upset the 12 seed in the first round. Stevenson won. OPRF lost to Barrington, the 9 seed, so they became the 9 seed. OPRF had to win its last three matches to make the top bracket.

In Round 3, Stevenson played Loyola, who had started the day as the 4 seed and maintained the seed until that point by winning their first two matches. Stevenson upset Loyola, so Stevenson at this point had to win both of its next two matches to get a one-game advantage in the afternoon or one of its next two matches to make the top bracket. OPRF beat Hoffman Estates, who had started the day as the 17 seed and kept that seed through that point.

In Round 4, Stevenson lost to Auburn, who had started the day as the number 1 seed and was still the number 1 seed because they had won their first three matches. This meant that Stevenson had to win their last match. OPRF played Wheaton North, who started the day as the 20 seed but moved up to 12 by beating Naperville North.

In Round 5, Stevenson lost to Northmont. Because Stevenson had the 4 seed, they got to play the lowest seeded 3-1 team, but they lost. Northmont had started the day as the 10 seed, and they were still the 10 seed until they defeated Stevenson. OPRF beat DCC, who started the day as the 3 seed and was still the 3 seed until they lost to OPRF.

Please explain who got screwed and how they got screwed. Type slowly so that I understand, because there isn't a single point you've made in this thread that makes any sense to me at all. While you're at it, teach me how to end a tournament at a reasonable time, using examples from last year's PACE NSC and ACF Nationals.

Captain Sinico · Post by **Captain Sinico** » Mon Dec 21, 2009 9:13 pm

Matt Weiner wrote:I think there's some sort of disconnect in how our minds are working in this thread. I don't care about "what happens if all the players on the 9th seed at PACE quit before the tournament" or "what happens if a TD puts the top 8 teams in the tournament into one bracket." I'm dealing with empirical real-world scenarios that can happen, did happen at New Trier, and do happen every time power matching is used.

What scenarios?! You keep saying "the cases you can prove that power matching is better for don't matter." First of all, that's wrong because you've claimed that power matching is always worse, and it isn't; people have proven cases in which power matching is better, whereas you have proven nothing whatsoever. Second, your discarding of those cases is baseless as it may (often) be the case that one doesn't have very good seeding data for many teams and therefore most of the seeds are or might as well be essentially random. For example, at New Trier, I'd have had little idea how to rank teams below the very best teams, and you'd have had less still.
The disconnect here (in the signal, since there's now a ton of noise because we just can't stop calling one another names) seems to me to be that between a group of people saying "Every mathematical analysis indicates that different formats optimize different fairness parameters given different field sizes and round number constraints" and you trying to say "Power matching is a priori always worse because once a single marginal-playoff team missed the playoffs and played a harder schedule and also because my spidey sense says so." I don't know how many people who understand probability and mathematical modeling better than you do have to tell you you're wrong before you understand that you're wrong, but add my voice to that chorus as I'm telling you: you're wrong.
That said, let me make a deal with you. If you will state what you consider to be a realistic scenario and win probability for teams given their relative ideal seeds and I'll solve the problem of telling you what happens in the mean to a mis-seeded team either analytically or by Monte Carlo simulation. Perhaps I can even write a tool that will let us actually find a best format in general given some assumptions, which would be about the first productive thing out of this thread.

MaS

Matt Weiner · Post by **Matt Weiner** » Mon Dec 21, 2009 9:15 pm

Shcool wrote:The reason I started using power matching is that I went to plenty of tournaments with 32 to 48 teams that used pools and didn't like them.

So, you dislike the thing you are arguing against. Noted.

They are poor in practice and in theory. Teams show up missing players, or teams show up with all their good players for the first time on a regular basis, leading to unfair pools. Or hosts distribute teams in pools poorly. The list goes on.

Zero-elimination and incommensurable scheduling are not the solution to this problem.

In Round 4, Stevenson lost to Auburn, who had started the day as the number 1 seed and was still the number 1 seed because they had won their first three matches. This meant that Stevenson had to win their last match. OPRF played Wheaton North, who started the day as the 20 seed but moved up to 12 by beating Naperville North.

Please explain who got screwed and how they got screwed. Type slowly so that I understand, because there isn't a single point you've made in this thread that makes any
sense to me at all.

That would be Stevenson, who had to play a guaranteed-loss game against the best team in the tournament, while OPRF played the 20th seed, non-playoff team Wheaton North. Your error is believing that both of these games mean the exact same thing and can be compared as simple win-or-loss events, which manages to deny the very fact that a tournament which awards wins, losses, and trophies implicitly accepts: that some teams are better than others, often greatly so.

As this one example illuminates, every team who went 3-2 while playing a significantly harder schedule than a 4-1 team got screwed out of the playoffs. Furthermore, due to the fact that five games cannot suffice to be representative of a 32-to-48-team field when teams are steered towards equal-ability matches, this is necessarily the case anytime power-matching is used over a small number of rounds.

While you're at it, teach me how to end a tournament at a reasonable time, using examples from last year's PACE NSC and ACF Nationals.

Man, it sure is awful when people are "uncivil" on message boards!

I consider 2:43 PM and 3:41 PM to be reasonable ending times. Particularly, since the PACE NSC announcement stated that Round 15 would end at 4 PM (and thus Round 16 would end no earlier than 4:30 PM), and instead round 16 ended at 2:43 PM, I am proud of the efficiency which I brought to the 2009 NSC. Perhaps you do not consider ending an hour and forty-five minutes earlier than the announced end time to be "reasonable" or efficient; I would certainly like to hear more about this novel opinion.

Similarly, perhaps the fact that ACF Nationals had other problems with direction, independent of ending in the mid-afternoon about an hour after the announced time, means that any opinion supported by the TD of ACF Nationals is incorrect (the "ad hominem" fallacy); were this the case, surely your support of various and sundry fake quizbowl practices, including but not limited to speed arithmetic tossups, makes both positions wrong and sets us back to square one.

at your pleasure · Post by **at your pleasure** » Mon Dec 21, 2009 9:31 pm

Could someone clarify why power-matching is less sensitive to seeding than pool-play(if that is the case)?

Matt Weiner · Post by **Matt Weiner** » Mon Dec 21, 2009 9:50 pm

Captain Sinico wrote:What scenarios?!

Where Team A plays a significantly harder schedule than Team B (defined by the bonus conversions, PPG, or records in the same playoff round-robin bracket of the two sets of opponents) and is compared to Team B based on record alone. Unlike the hypothetical constructs used in an attempt to show that most ICTs, every ACF Nationals, every other college tournament, and 90% of high school tournaments held each year are illegitimate or otherwise bash the bracket system, my scenario actually happened not only at New Trier, but at every other actual example of a power-matched tournament that you can show me results for.

You keep saying "the cases you can prove that power matching is better for don't matter."

Because they are proofs of no relevance to the real world. It's nearly certain that people are right when they point out that power matching is superior to a bracketed system in which the best 8 teams in a 64 team field are placed in one prelim bracket, or that brackets made based on completely erroneous data about rosters can potentially flip a deserving playoff team with an undeserving one. But the first case is a constructed example which could only exist in the real world due to active TD malice, and to my knowledge never has (if you have a counterexample, the thread remains open) and the second does not establish why power matching (being inherently seeded) is less vulnerable to the same problem, or why brackets would be so much more vulnerable to it as to outweigh the other fairness problems with power matching.

Yes, there are hypothetical cases in which power matching would be better than bracketing, and other cases in which it would be difficult or impossible to tell which system is better. I am not concerned about those cases, but about actual tournaments that have happened in the past and will happen in the future.

First of all, that's wrong because you've claimed that power matching is always worse, and it isn't; people have proven cases in which power matching is better, whereas you have proven nothing whatsoever.

Power matching "is" always worse, taking "is" as a description of the real situation. People have proven that it is mathematically possible to deliberately construct situations in which power matching may be, or is definitely, better than bracketing, not that any real tournament was ever better off for using power matching, or that any real tournament run in remotely good faith in the future would be.

Second, your discarding of those cases is baseless as it may (often) be the case that one doesn't have very good seeding data for many teams and therefore most of the seeds are or might as well be essentially random. For example, at New Trier, I'd have little idea how to rank teams below the very best teams.

Then you shouldn't use a format in which the initial ranking as a projected 5th, 6th, 7th, or 8th place finisher plays an enormous role in determining how difficult your schedule will be.

The disconnect here seems to me to be that between a group of people saying "Every mathematical analysis indicates that different formats optimize different fairness parameters given different field sizes and round number constraints"

Reinstein seems to be arguing that "fairness," in and of itself, is not a concern for formats, and may be sacrificed for any number of other goals, including:
-making sure that there are 2^int teams in contention so that your swiss-pair format works as you wish it to (more important than fairness because it was achieved by engaging in zero-elimination for the number of teams necessary to create a 2^int field)
-making sure the "top teams" play each other as much as possible in the prelims (more important than fairness because it was achieved by giving wildly different SOSes to teams who were then compared on won-loss record alone)
-making sure the number of teams who make the playoffs is exactly equal to the available number of playoff rounds plus one (more important than fairness because it necessitated both of the above things, given the other constraints of the tournament)
-not using a single-elim playoff (more important than fairness because it necessitated the above and hence what's above that, given the other constraints of the tournament)
-making sure that the least interested teams in the field don't leave the tournament early (more important than fairness because it ruled out using brackets for both the prelims and the playoffs, also not at all achieved)

Whether you agree with any of these contentions is not something I'm going to speculate on, but you legitimize them when you argue that "fairness parameters" are something that follow from the tournament format, rather than the other way around.

and you trying to say "Power matching is a priori always worse because once a single marginal-playoff team missed the playoffs

My contention is that the most important fairness function of the prelims is to correctly sort the marginal teams into and out of the playoffs; if I have proven that power matching caused this to happen, then I need only prove said contention itself in order to prove that power matching is unfair.

and played a harder schedule

Which is indisputable, right?

and also because my spidey sense says so."

Abstract argument: Because the #1 (and also the #2, etc....) team can only be in one game at a time, it is inevitable that some marginal team will have a much harder schedule in a power-matched prelim than another.

Empirical argument: I can demonstrate that this occurred in every example of a power-matched tournament for which statistics exist.

I don't know how many people who understand probability and mathematical modeling better than you do have to tell you you're wrong before you understand that you're wrong, but add my voice to that chorus as I'm telling you: you're wrong.

The best math in the world does not work when the data and premises are incorrect! And what Reinstein, and to a lesser extent Dwight, is doing, is to start with premises that are factually incorrect (e.g, "tournaments are always seeded correctly"), normative ("all playoffs should be bracketed"), or question-begging ("the teams which finished the highest in my tournament were the teams who finished the highest in my tournament"), and using them to prove conclusions ("and thus, top teams played each other as much as possible and the tournament ended on time") which are non-sequiturs with respect to the contention that the tournament was unfair.

That said, let me make a deal with you. If you will state what you consider to be a realistic scenario and win probability for teams given their relative ideal seeds and I'll solve the problem of telling you what happens in the mean to a mis-seeded team either analytically or by Monte Carlo simulation. Perhaps I can even write a tool that will let us actually find a best format in general given some assumptions, which would be about the first productive thing out of this thread.

The issue is not one of specific seed but of tier; in some tournaments, you know with reasonable certainty who teams 1 and 2 are and who teams 11 through 32 are and you're trying to sort out 3 to 10 properly; in others, perhaps the elite tier has 5 teams and the bottom has 24, or whatever.

Captain Sinico · Post by **Captain Sinico** » Mon Dec 21, 2009 9:57 pm

What I'm saying, Doug, is not that power matching is less sensitive or more, but that which format is best depends on a number of things. We can determine the sensitivity pretty directly by simulation. Suppose some win probabilities for a set of teams (we have to guess these.) Set up a tournament, seed it properly, and simulate it many times using the win probabilities and see what the mean outcome is. The variance in that outcome gives one measure of how good the format is. Then, perturb the seeding (change some seeds) and see how different the new mean outcome is. My crude analyses so far say that which format is less sensitive to seeding depends on the type and extent of seeding errors and the tournament size.
Dwight demonstrated upthread, for example, that power matching is less sensitive is general for the case that teams are seeded at random, so we can definitely say that power matching is preferable with respect to seed stability in the case of very little seeding data. My intuition is that brackets are better in general in the case of perfect seeding data, but I haven't been able to prove that to my own satisfaction yet.
At any rate, I think we'd all be foolish not to acknowledge the fact that we have a complicated, non-linear optimization problem here and even the solution to it gives us a bead on only one fairness parameter, namely stability of outcome with respect to seeding. There are other valid concerns, like equality of difficulty of path to championship and value of matches against like-skilled teams, that we should also consider. In short, I'm not confident that a general answer to the broader question "Is power matching bad vis a vis bracketed round robins?" exists.

MaS

Captain Sinico · Post by **Captain Sinico** » Mon Dec 21, 2009 10:16 pm

Matt Weiner wrote:
Captain Sinico wrote:What scenarios?!
Where Team A plays a significantly harder schedule than Team B (defined by the bonus conversions, PPG, or records in the same playoff round-robin bracket of the two sets of opponents) and is compared to Team B based on record alone. Unlike the hypothetical constructs used in an attempt to show that most ICTs, every ACF Nationals, every other college tournament, and 90% of high school tournaments held each year are illegitimate or otherwise bash the bracket system, my scenario actually happened not only at New Trier, but at every other actual example of a power-matched tournament that you can show me results for.

Similarly unfair results can and do happen at bracketed tournaments, too, as the brackets are in general of unequal difficulty. The simple fact is that short of a full round robin on ideal questions, every format allows similarly unfair results.

Matt Weiner wrote:[Mathematical proof that one format is more stable than another in some circumstances may be neglected on the grounds] that on they are proofs of no relevance to the real world...

That claim is wrong. I told you why. Even were it not, it does nothing to make your argument for you, as you have yet to show that the format you have in mind is better under any set of circumstances.

Matt Weiner wrote:[If seeding data are not well known, one] shouldn't use a format in which the initial ranking as a projected 5th, 6th, 7th, or 8th place finisher plays an enormous role in determining how difficult your schedule will be.

I agree, one should try to minimize the role that seeding data play in determining the outcome if one is uncertain of them. So, in that circumstance, one ought to use one format or another depending on which minimizes that role. The fact is that the identity of that format is an open question in general and definitely depends on the tournament size and win probabilities of various teams.

Matt Weiner wrote:...you argue that "fairness parameters" are something that follow from the tournament format...

No, I don't. What I'm telling you is: one has to define what one wants out of a format (a "fairness parameter") before one sets out to design a format to optimize that parameter. I'm further telling you that different formats will optimize different fairness parameters under different circumstances. Thus, even your own definition of fairness may lead to the conclusion that a power matching-type format is better under some (realistic, again even by your own definition) circumstances.

Matt Weiner wrote:My contention is that the most important fairness function of the prelims is to correctly sort the marginal teams into and out of the playoffs; if I have proven that power matching caused this to happen, then I need only prove said contention itself in order to prove that power matching is unfair.

That's not so. To show that power matching is less fair even by your own definition, you have to show that it fails to meet what you're saying is fair in the mean less often than the format you have in mind. Again, the fact is that all formats allow some form of unfair result, so saying "power matching sometimes allows this one unfair result" doesn't make your argument for you.

That said, let me make a deal with you. If you will state what you consider to be a realistic scenario and win probability for teams given their relative ideal seeds and I'll solve the problem of telling you what happens in the mean to a mis-seeded team either analytically or by Monte Carlo simulation. Perhaps I can even write a tool that will let us actually find a best format in general given some assumptions, which would be about the first productive thing out of this thread.

The issue is not one of specific seed but of tier; in some tournaments, you know with reasonable certainty who teams 1 and 2 are and who teams 11 through 32 are and you're trying to sort out 3 to 10 properly; in others, perhaps the elite tier has 5 teams and the bottom has 24, or whatever.[/quote]
Okay. Tiers work fine, too; that's just the same as a bunch of teams with the same ideal seed. So give me the win probabilities among tiers, tier sizes, and two tournament formats and we can see which formats are actually the best.

MaS

Matt Weiner · Post by **Matt Weiner** » Mon Dec 21, 2009 10:49 pm

Captain Sinico wrote:Similarly unfair results can and do happen at bracketed tournaments, too, as the brackets are in general of unequal difficulty. The simple fact is that short of a full round robin on ideal questions, every format allows similarly unfair results.

I agree with your second sentence. That is why I address your first sentence by supporting the disregarding of prelim won-loss record in multiple-prelim-bracket systems, and instead only allowing place within the prelim bracket (determine by round-robin record within the bracket) to determine playoff seed, with only opponent-independent statistics (BC) used to compare teams across brackets when paper tiebreakers are fair, necessary, and/or sufficient.

Papa's in the House · Mon Dec 21, 2009 11:03 pm

Matt Weiner wrote:
Captain Sinico wrote:You keep saying "the cases you can prove that power matching is better for don't matter."
Because they are proofs of no relevance to the real world. It's nearly certain that people are right when they point out that power matching is superior to a bracketed system in which the best 8 teams in a 64 team field are placed in one prelim bracket, or that brackets made based on completely erroneous data about rosters can potentially flip a deserving playoff team with an undeserving one. But the first case is a constructed example which could only exist in the real world due to active TD malice, and to my knowledge never has (if you have a counterexample, the thread remains open) and the second does not establish why power matching (being inherently seeded) is less vulnerable to the same problem, or why brackets would be so much more vulnerable to it as to outweigh the other fairness problems with power matching.

There is, actually, a more general scenario than #1 which you mentioned that would "screw" a marginal playoff team out of the playoffs. This more general scenario occurs whenever you place (# of playoff berths/# of preliminary pools)+1 of the best # of playoff berths teams in any bracket. Using a 64 team field with what I'm assuming will be 8 playoff berths and 8 preliminary pools, if you place any 2 ((8/8)+1)of the top 8 teams in a preliminary pool together you will "screw" a marginal playoff team (or perhaps, the second-best team) out of a playoff berth. In the "real world" this can be seen through the example of the 2009 ACF Fall in Chicago. There were two pools of 12 and the 2 best teams (determined by won-loss record) of each pool advanced to the playoffs. As it happened, Illinois' A-C teams happened to be in the same bracket and they finished in the top 3 spots, with Illinois C (8-2, losing only to the other Illinois teams) taking spot #3 and thus being eliminated from the playoffs. Based on conversations held with players at the tournament and looking through stats from Illinois C's preliminary rounds as compared to teams in the other bracket (basically, every statistic except for -5s; posting.php?mode=quote&f=3&p=166183), Illinois C would likely have clinched one of the top 2 spots if placed in the other bracket (I cannot say they would for sure have clinched one of those spots, as an upset could always have taken place). In this case, placing 3 ((4/2)+1) of the best 4 teams in the same preliminary bracket "screwed" that third team out of a playoff berth. This being said, I'm not arguing for nor against power matching (or bracketing, etc.). Though, it seems I have taken long enough as to make this post moot as you have already agreed that a pool system can be unfair in certain circumstances.

Matt Weiner wrote:
Captain Sinico wrote:Similarly unfair results can and do happen at bracketed tournaments, too, as the brackets are in general of unequal difficulty. The simple fact is that short of a full round robin on ideal questions, every format allows similarly unfair results.
I agree with your second sentence. That is why I address your first sentence by supporting the disregarding of prelim won-loss record in multiple-prelim-bracket systems, and instead only allowing place within the prelim bracket (determine by round-robin record within the bracket) to determine playoff seed, with only opponent-independent statistics (BC) used to compare teams across brackets when paper tiebreakers are fair, necessary, and/or sufficient.

Stained Diviner · Post by **Stained Diviner** » Mon Dec 21, 2009 11:14 pm

Let me state once again that what happened at New Trier was fair. Because Stevenson established itself as slightly better than OPRF based on opponents beaten in the first four rounds, Stevenson had an easier opponent in the fifth round, which is when both teams faced elimination. In other words, Stevenson earned an advantage, and they were given that advantage when they needed it, which is the way fair systems work. The fourth round was fair because Stevenson had a cushion while OPRF was playing an elimination match. Fairness was not sacrificed--it was strengthened by using this system. When push came to shove, Stevenson's success depended on whether or not they beat the 10 seed, and OPRF's success depended on whether or not they beat the 3 seed.

If your point is that it is impossible to pick the top six teams out of 32 in five rounds, that's a point that's arguable. To the extent that it is possible, we did it as best it can be done. If I was a believer in single elimination, we could have taken all the 3-2 teams into the playoffs and done single elimination.

My ACF and PACE references were to the first days, and, as you know, my opinion is not novel. Since you are going to teach me how to finish a tournament in a timely manner, let me offer you a free lesson: It's bad form to start a thread subtitled "a theoretical offering" and then, after you have lost every theoretical argument, to equate theory with masturbation.

Matt Weiner · Post by **Matt Weiner** » Mon Dec 21, 2009 11:27 pm

Shcool wrote:My ACF and PACE references were to the first days, and, as you know, my opinion is not novel.

Oh, the PACE NSC day that was announced to end at 7 PM and ended at 7:24 PM. Yeah, that totally makes power matching the New Trier tournament a good idea.

Since you are going to teach me how to finish a tournament in a timely manner

Actually, you were the one who brought up ending time as a criterion in my discussion of tournament fairness, and I've correctly dismissed "the tournament was fair because it ended on time" as a non-sequitur.

, let me offer you a free lesson: It's bad form to start a thread subtitled "a theoretical offering" and then, after you have lost every theoretical argument, to equate theory with masturbation.

When the "theory" is just a trick of adding x and y to get z when you get to define what x and y are and why z is important, then yeah, it's about as useful to the discussion of how to run tournaments in the real world as anything else that you've embarrassed yourself in this thread with.

I may be wrong about certain things, though I doubt my opinions about power-matching are one of them. I am certainly not an alleged adult authority figure who argued my position in this thread by namecalling, declaring "I win!" louder and more frequently each time it somehow did not end the debate, using at least five logical fallacies, and mudslinging about irrelevant and/or made-up actions of my rhetorical opponent. Someone here may be coming off as having a "vendetta against power matching," but someone else is coming off as twelve years old.

Stained Diviner · Post by **Stained Diviner** » Mon Dec 21, 2009 11:35 pm

To get this back on track, let me repeat what I said earlier:

Shcool wrote:I challenge Matt once again to point out one of his arguments that has held up. I don't see one.

Matt Weiner · Post by **Matt Weiner** » Tue Dec 22, 2009 12:26 am

My core argument is that: two teams played two schedules that were way more different in terms of strength of opponents than any good-faith attempt at bracketing would have been, but were compared solely on wins and losses against these two very different sets of opponents, which is unfair.

Despite your bizarre adolescent strategy of repeatedly and loudly asserting that I have not offered the above plain demonstration of the tournament's unfairness, I in fact have.

jonpin · Post by **jonpin** » Tue Dec 22, 2009 2:11 am

I think that a very definable and relevant case where bracket play unfairly eliminates someone early would be in the PACE 8x8 setup if the 12 seed loses their science player and is no longer really one of the top 16 teams, then either the 16 and 17 seed will be eliminated unfairly. If the 24 seed has finally pulled all of their pieces together and is legitimately 12th-best at the tournament, they will still likely be bounced by the 9 seed while somewhere a less-deserving team advances. These are opposite sides of the same coin: one team incorrectly placed results in an incorrect selection of the top tier.
I do not know how power-matching fares under such scenarios.

Matt Weiner wrote:Then you shouldn't use a format in which the initial ranking as a projected 5th, 6th, 7th, or 8th place finisher plays an enormous role in determining how difficult your schedule will be.

In a 4x8 bracket setup, the 8-seed has the 9-seed in their way of playoffs, while the 5-seed has the 12-seed. That can easily be the difference between a near-even game and a near-certain win.

In general, though, I think group play should be used in regular-season tournaments because it is more robust, especially in dealing with potentially awkward numbers of teams (power-matching deviates from its purpose in round 3 if there are 28 teams).

I propose a simulation that will attempt to determine whether the two systems are reasonably close in utility. I'm still refining it, so it'll appear in my next post.

jonpin · Post by **jonpin** » Tue Dec 22, 2009 2:56 am

First let me state my qualifications: A mathy quiz bowl enthusiast with proficiency in Excel who can honestly answer "No" to the entirely reasonable question "Don't you have something better to do with your time?"

Assuming that I can make something to do this in a reasonable time-per-run, I'll simulate 100 tournaments under each system.

Basics
Each tournament will feature 32 teams.
Teams will be given pre-rankings of 50, 49, 48, 47... 19. Using those pre-ratings, the teams will be seeded 1-32 for the purpose of the power seeding and determining groups.
For each tournament, a team's actual rating will vary via a normal distribution with SD 3.0.
Using the actual ratings, the top X teams will be determined, and considered the IDEAL PLAYOFF TEAMS.
Each game will be simulated using ELO win probability P = 1/(1+10^[A-B]/10). A team with a rating 5 higher should win 3 out of 4 times.

Power Matching: X = 6, with a 5-round system from the Charter page, and the unbeaten and one-loss teams advancing.
Group Play: X = 8, with a 4x8 system seeded in snake fashion, and the two best teams in each group advancing. Ties broken by playoffs.

For each tournament, I'll note how many of the ideal teams qualified and how bad the mistakes were (i.e. did 7th qualify instead of 6th or did 13th qualify instead of 4th) and try to perform some statistical analysis based on this. I'll probably not actually get to this until next week, so I'm open for suggestions into improving the setup of the experiment.

Mechanical Beasts · Post by **Mechanical Beasts** » Tue Dec 22, 2009 3:07 am

At some point, you should probably look at situations where a team can have a slightly variable performance per game (based on packet); just adding a random value from (-.5,.5) to the true rating for a given round and calculating the ELO then might be an okay way to do it.

Also, I'd maybe prefer to see a more realistic ranking than {1, 2, ..., 50}. Account for some TD fudge-factor by using tiers as discussed above: for example, just have 15, 13, 11, 9, 7, 5, 3, 1 for predicted strengths and let inaccuracies happen from that starting point.

Pilgrim · Post by **Pilgrim** » Tue Dec 22, 2009 3:28 am

Crazy Andy Watkins wrote:At some point, you should probably look at situations where a team can have a slightly variable performance per game (based on packet); just adding a random value from (-.5,.5) to the true rating for a given round and calculating the ELO then might be an okay way to do it.

Maybe I'm missing something, but what does this accomplish that using the ELO winning percentage doesn't already do?

Mechanical Beasts · Post by **Mechanical Beasts** » Tue Dec 22, 2009 3:54 am

Pilgrim wrote:
Crazy Andy Watkins wrote:At some point, you should probably look at situations where a team can have a slightly variable performance per game (based on packet); just adding a random value from (-.5,.5) to the true rating for a given round and calculating the ELO then might be an okay way to do it.
Maybe I'm missing something, but what does this accomplish that using the ELO winning percentage doesn't already do?

Oh, yeah, you're right. I was attributing the use of the ELO to introduce uncertainty as being all about multiple repetitions of the exact same system (i.e. team A beats team B on the following twenty questions nine times out of ten) rather than about different rounds and therefore different questions (i.e. teams B and C have the same rating; A doesn't necessarily beat C in round two just because it beat B in round one). That works, then.

Joshua Rutsky · Post by **Joshua Rutsky** » Tue Dec 22, 2009 5:23 am

I don't suppose that once we get back from the lofty heights of quizbowl theory arguments taking place here that someone might try to codify some guidelines for what WOULD be a suitable format and methodology for hosting a medium-sized tournament? I've held an annual for five years now, and I certainly would like to refine it to make it as fair as possible in structure. No tournament director, I think, really wants to see their tourney labeled as unfair or having contained a screw job. I, for one, am more than a little confused by some of the theory in this thread, but I get the general idea: there is a difference of opinion on whether power-matching or bracketing make for a fairer tournament. My question would be how a TD could take a field of, say, 32 teams from multiple states, and begin to seed or power-rank them properly? Do you turn to results and start making a spreadsheet? Do you take only "major" events into consideration, or do you take things like "county tourneys" as well? Is it preferable to bracket geographically on some level to avoid having a team travel 3-4 hours to an event just to play a team they play at every local tournament? How many rounds should teams play in a good tourney?

I know these are broad questions, and probably have simple answers in many cases, but getting all those answers into a thread on ideal tournament construction might help TDs like myself, who have played for years but are relatively limited in our out-of-state experience, to bring better-quality tourneys to our respective states.

Because I'm typing late in the night after my 4 month-old son woke me, I want to state that absolutely no sarcasm or snarkiness is intended in this post, just in case I'm too addle-brained at the moment to proofread it properly for tone. I'm quite serious about wanting to improve my own tourneys.

Stained Diviner · Post by **Stained Diviner** » Tue Dec 22, 2009 7:27 am

Jon, do not use the pairings from the Charter page. They are incorrect. Use the pairings from Dwight's spreadsheet. Also, I don't know whether your experiment will find power matching to give superior results--it may only find that power matching gives equivalent results using fewer rounds. I contend that getting equivalent results in fewer rounds is a significant advantage to anybody running a large one-day tournament.

Now that I have slept, and perhaps matured a bit, it strikes me that Matt's dislike of power pairing may be due to the fact that he has seen power pairing done improperly. I don't know how often the system is used on the East Coast for local tournaments and exactly how it is done, but the pairings on Tressler's page create the problems that Matt has described. If you follow the 32 team pairings, they are fine for the first three rounds. In Round 4, however, the strong 2-1 teams play the other strong 2-1 teams, and the weak 2-1 teams play the other weak 2-1 teams. This is unfair. (I did not figure out his fifth round yet, and I apologize to him for not responding yet to his message in the thread started by Greg Gauthier.)

I know that Matt has been at several HSNCT's, and there have been similar flaws in that tournament. I can't go into too much detail about it here for two reasons--I could not figure out exactly what NAQT did because their seeds are blind, and NAQT changes a lot of the details of their system from year to year even though it appears to be the same system on the surface. In some years, they come closer to what should happen than in others, but it's possible that 2003 was the only time they got it exactly right. (I have not studied their results closely enough to state that last fact definitively, so the word possible is used there for a reason.) I believe that 2006 was their worst system. To some extent, the pairings have to change based on the number of teams, but the changes from year to year have been more significant than that. After last year's HSNCT, R and I exchanged several emails, and I explained to him how he should have done the first 10 or so rounds (or each team's first 7-8 matches) of HSNCT meeting the difficult parameters they have to think about when running a very large tournament--they try to prevent teams from having to cross the hotel when going from one match to another, and they have some teams playing their 2nd matches while other teams are playing their 1st. Suffice it to say that there is an optimal system for each team's first 6-7 matches, and NAQT did not use that optimal system.

When power matching is done wrong, some teams get screwed. When power matching is done right, which is what we did, teams don't get screwed, which is why Stevenson faced a lower seed than OPRF when both teams faced elimination.

Josh--if you're afraid of running a good tournament and then having somebody on this board telling you that you were grossly unfair, then do whatever Matt tells you. If you want to run the fairest possible tournament for 32 teams in 10 rounds, with a potential extra round or two for a few teams to break ties, then copy what we did Saturday. If you want to use 16 rounds, then I can explain how our tournament would have been better with 16 rounds.

AlphaQuizBowler · Post by **AlphaQuizBowler** » Tue Dec 22, 2009 11:20 am

Shcool wrote: If you want to run the fairest possible tournament for 32 teams in 10 rounds, with a potential extra round or two for a few teams to break ties

Couldn't you do 4 brackets of 8 (7 rounds), then tiebreak on a half-packet (7.5) and take the top 8 in two brackets of 4 (10.5), tiebreak (11), and then do a crossover game (12)? You had twelve packets, right?

Whiter Hydra · Post by **Whiter Hydra** » Tue Dec 22, 2009 11:27 am

Shcool wrote:Josh--if you're afraid of running a good tournament and then having somebody on this board telling you that you were grossly unfair, then do whatever Matt tells you. If you want to run the fairest possible tournament for 32 teams in 10 rounds, with a potential extra round or two for a few teams to break ties, then copy what we did Saturday. If you want to use 16 rounds, then I can explain how our tournament would have been better with 16 rounds.

No, you had 48 teams.

Stained Diviner · Post by **Stained Diviner** » Tue Dec 22, 2009 12:53 pm

Let me point out again how unfair pools are. When you have pools, teams that play the same strength schedule and get the same record are not treated the same way. For example, at last year's NSC, Southside A and Chattahoochee each finished 5-2 and were allowed to playoff a match to determine which one would advance. Seven other teams finished 5-2 and were eliminated because they finished 5-2. (I wasn't there. Perhaps Southside A just advanced on the basis of some statistics, while the statistics of seven other 5-2 teams did not matter because of results of matches they did not play in.)

Why was Southside given an opportunity that seven other teams with the same record were not given? Because of results of matches that they did not play in. In other words, their fate was decided by the most unfair factor possible. Seven teams were eliminated by losing twice and having results in matches they were not in not go their way, while two teams were not eliminated by losing two matches because results in other matches did go their way.

In 2006, Gonzaga and Richard Montgomery B were forced to play an extra game after finishing the preliminaries with a 6-3 record. Maggie Walker B and Brindlee Mountain A, on the other hand, got to go directly to the playoffs after finishing 5-3. That scenario is a little odd because the size of the field forced byes that year, but the same scenario could very well happen this year--some 6-1 teams could be forced to play off to earn a spot in the top brackets while a 5-2 team could go directly in. (Edit: This paragraph had some incorrect numbers originally.)

William, your system claims to require 2 half-packets but actually requires 4 half-packets depending on circles of death, and that's a lot of half-packets for a fair system. It also makes the last game single-elimination, so a team that wins its first ten games and loses its eleventh does not win the tournament even if there are no undefeated teams in the tournament. It also requires rebracketing after round 7, which is not lunch time. Since I assume you want to figure out a fair placement for all 32 teams using the results of all 7 rounds, this would result in a long pause between rounds 7 and 8. It also runs into the same problems I just described above.

Harry, you are correct. I devised a system that would allow us to do a 48 team power matching in five rounds, but I was voted down by the people I was working with because it gave too great an advantage to the #1 seed. Of the 120 matches, 119 were between teams with identical records, always matching a team from the top half of the seeds with that record with a team from the bottom half of the seeds with that record, and one match was between a 4-0 team and an 0-4 team. Because of that one match, I lost that debate. If you want to have a debate over which power matching system we should have used, that's fine. We very well might have used the wrong one.

jonah · Post by **jonah** » Tue Dec 22, 2009 6:11 pm

Shcool wrote:Harry, you are correct. I devised a system that would allow us to do a 48 team power matching in five rounds, but I was voted down by the people I was working with because it gave too great an advantage to the #1 seed. Of the 120 matches, 119 were between teams with identical records, always matching a team from the top half of the seeds with that record with a team from the bottom half of the seeds with that record, and one match was between a 4-0 team and an 0-4 team. Because of that one match, I lost that debate.

That's not why I decided not to use your proposal, but we can have that discussion by email if you want to have it again.

Whiter Hydra · Post by **Whiter Hydra** » Tue Dec 22, 2009 10:35 pm

Shcool wrote:Harry, you are correct. I devised a system that would allow us to do a 48 team power matching in five rounds, but I was voted down by the people I was working with because it gave too great an advantage to the #1 seed. Of the 120 matches, 119 were between teams with identical records, always matching a team from the top half of the seeds with that record with a team from the bottom half of the seeds with that record, and one match was between a 4-0 team and an 0-4 team. Because of that one match, I lost that debate. If you want to have a debate over which power matching system we should have used, that's fine. We very well might have used the wrong one.

Or perhaps it wouldn't have been a good idea to use power matching in that case. You were trying to sort out more than 2^5 teams in 5 rounds, which you can't fairly do in power matching, regardless of people's opinion on whether one can do it with 32 teams. You have the same problems you've always had with power matching, but you also are essentially creating divisions, since it would be physically impossible for any given team to play a third of the field at all (so you have all the issues you've raised about divisions as well).

Post by **theMoMA** » Tue Dec 22, 2009 10:56 pm

Reinstein, the records of bracket play are compared only to the records of teams who've played exactly the same schedule. So a 5-2 team in one bracket is compared only to other teams who played the exact same teams. Now obviously, wild imbalances in brackets strength would mean that all teams do not have a fair chance to make the playoffs. But power matching produces what amounts to a different prelim bracket for everyone, then purports to say that one 4-1 record equals another, and that both are indicative of a stronger team than any 3-2 record. It has been shown in this thread that neither of those statements rings true, and I think you'll find that power matching inevitably results in conundrums like this.

Power matching is still a fair prelim option if you recognize its shortcomings and create a very inclusive playoff structure that keeps all teams that deserve to be in contention in contention. In certain cases, this might be a good option (like if you have to determine the playoff teams on a limited number of packets, or with a gigantic field like HSNCT). But it seems to me that a quick browsing of the New Trier stats indicates that teams that deserved to be in contention were excluded by the comparison of imbalanced records. I won't go as far as Matt in saying that this indicates that power matching is always unfair, but it seems that if you do it the way that you did, it lends itself to unfairness.

the return of AHAN · Post by **the return of AHAN** » Tue Dec 22, 2009 11:45 pm

Earthquake wrote:... but you also are essentially creating divisions, since it would be physically impossible for any given team to play a third of the field at all (so you have all the issues you've raised about divisions as well).

As a coach of two teams in this tournament, I had absolutely no issue with our B team having no chance to get run out of the room by an Auburn A, DCC, Loyola A or St. Ig, etc. Really, those teams are awesome. I get it, and they don't need to crush my B team 800-0 to prove it. I think, for my players, it was just as worthwhile to get to play other B teams and have a chance to be competitive. Now, at Kickoffs, our B team played two A teams in their bracket and actually took Homewood Flossmoor to the final question before losing, and trailed Naperville Central by a slim margin at half. That was a good experience, too. But I wish people would stop acting like the B teams at NTV were robbed. Really, my players were happy with the experience.

And didn't the two lower-tier teams that advanced up to A level consolation brackets go a combined 1-9 in the afternoon? I simply don't understand why people are caterwauling over the division of A and B teams, especially when a program that wanted their B team to play up could've done so.

Stained Diviner · Post by **Stained Diviner** » Wed Dec 23, 2009 12:16 am

(NOTE: This was written in response to Andrew before Jeff's post. I'm not writing a long response to Harry, but I will say that he makes a reasonable point and that he's talking about a situation that does not present one solution, pool or power, that is much better than the other. Andrew also made a reasonable point--thus this post.)

To make the playoffs at NTV, OPRF needed OPRF to beat DCC, the 3 seed. To miss the playoffs at NTV, Stevenson needed Stevenson to lose to Northmont, the 10 seed.

To make the playoffs at NSC, Southside needed either Chattahoochee or Loyola A to beat State College B, and it also needed Maggie Walker A to beat Chattahoochee.

Neither system is perfect, but which is more fair?

I agree that our system was imperfect, an imperfection necessitated by the length of the tournament. However, all of the teams that people claim should have been in the top bracket lost head-to-head to all of the teams that people claim should not have been in the top bracket. Nobody has suggested a way of fixing that other than making the afternoon single elimination, which would allow more marginal teams to make the top bracket but which would make the afternoon itself less fair, using pools in the morning (so that we wouldn't have to worry about good teams playing each other) followed by a single elimination afternoon, or making the tournament longer than it could have been.

There are two ways we could have gotten Stevenson into the top bracket. One would have been to schedule the tournament so that Stevenson did not play any very good teams in the morning, allowing one or two somewhat good opponents, and then go on to single elimination in the afternoon. Their first opponent in that single elimination afternoon would have been roughly the same quality as Northmont, so I'm not sure what that would have accomplished. (I'm selling Northmont a little short here, and I hope they don't take it the wrong way, because they are a solid team and complete class act.) The other would have been to keep what we did and allow the top half of our field to go on to single elimination in the afternoon, a solution that could easily have prevented Auburn from not only winning the Championship but finishing in the Final Four. In other words, we could have taken other people's suggestions, and we would have had a worse tournament if we had done so.

If somebody wants to state that power matching is imperfect or that there are situations where tournaments should use pools, my response is Yes. If somebody wants to state that pools are always better than power matching, my response is No. If somebody wants to state that pools would have been better than power matching in our situation, my response is No. What we did was more fair than what anybody has suggested under the tournament length constraints we were under. If we used pools and Auburn got eliminated in a single elimination quarterfinal, would everybody have praised our fairness?

Tegan · Post by **Tegan** » Wed Dec 23, 2009 4:26 pm

So let me get this straight .... I have 48 teams in 8 pools of six .... with six teams advancing to the championship pool in the afternoon .... thus winners of the pools advance.

Team A is an elite superior team. The only team remotely close to them is Team B, though Team B is clearly inferior. Somehow, Team B defeats Team A, and given that the rest of the pool was properly seeded and all of the other teams are vastly inferior, A and B win out, and B advances to the championship pool.

In power matching, let's say that the preliminary seeding was done fairly well, and A and B meet in the third of five rounds. both are currently undefeated, and A is still superior to B. B still finds a way to win. A has one loss, but dropping down to face another one loss teams should win out and make it to the afternoon. Providing that B got "lucky", they should lose one or both of their last two matches ... possibly also advancing, and possibly not.

How is the pool more fair than power matching?

I think the one thing that I will say about power matching that might be in disagreement with some of its proponents is that, like pool play, it does require careful seeding of teams ... I think there are some in the community who think power matching virtually means "seeding is unimportant" .... but far from it. I think power matching will smooth over "lucky wins" by inferior teams (as noted), but in the absence of proper seeding, Team A may lose, and drop down to play Team C which is nearly equally matched with Team A (maybe the same thing happened to them), and Team A takes a second loss, while the other one loss teams (call them D and E) are both considered vastly inferior to A, assuring one of them survives to the championship pool.

The issue is not with power matching .... I think it is with seeding. Having been at NTV, I think I can explain the complications that arose there.

1. You had some teams that just never come to tournaments on a regular basis (OPRF, Niles West), and are thus difficult to properly seed. I suspect that if you were to interview coaches, players, and officials with knowledge of good quizbowl and of the local scene and ask them if they were given a proper seed prior to the competition (and a low one at that), virtually everyone would have said "yes". It turns out that these teams played better than most expected.

2. You have teams like DCC, which from what I heard, were not at full strength. I read for them twice in the afternoon, and saw them get many questions on the giveaway, and a few before that .... they put up points, but I could easily see them getting beat by some teams that were at full strength in comparison. DCC was likely seeded based on their being at full strength. So, when they lost, and everyone saw this as a major upset, it likely was not as big an upset as people should have thought, given the game day circumstances. I am not sure about Stevenson this year, but I noted Zach was not there ... could this explain why they did not advance as far as their seed indicated they should? I am not sure, but it might.

Thus, I don't see the system being more unfair than pools ... if anything quite the opposite. If anything negative is to be said, it would revolve around preliminary seeding, and how those teams were destined to meet later in the tournament (which is essentially a part of the seeding process in power matched tournaments). However, given the unknown variables (teams with little track record, teams playing shorthanded), I think that makes the seeding process more difficult under any circumstance (power matching or pool play). I agree with David Reinstein's last comment: what if this had been a pooled tournament, and Stevenson or Loyola or DCC beat the living daylights out of everyone and then lose to one team (like an OPRF) while they watch one or two other pool winners move on that they could have easily beat. Then you are back to arguing the inherent unfairness of pool play. There may have been other issues with the tournament, but I would not call the format one of them.

Two weeks ago I ran a 12 team tournament with some fairly competitive teams. I took the top two from each pool of six to the championship pool in the afternoon .... I figured all I had to do is make sure that the top four teams are separated, and the next four, etc. I consulted 3 learned coaches/officials, and thought I had a decent set up, given there was general agreement on how I seeded the pools. Turns out, one of the teams picked for the top four was not playing really well, and one of the teams I had picked as one of the middle tiered teams played great (based upon an analysis of PPB and other factors). Unfortunately, they weren't in the same pool, so one team that played very well got screwed out of advancement. Should the pool system be condemned? Of course not .... the system was fine, it was the seeding that was in error.

I think that one way around this in the future, and hardly a panacea, is to commit coaches to report if they are going to be shorthanded ... it is a pain in the butt to have to reseed based on that, but at least it would prevent giving a team too easy a path, when one may not necessarily be deserving based on the team they show up with. As far as dealing with teams who rarely or never show up to tournaments ... I am not sure there is a way to deal with that.

From what I can see, the big issues revolve around OPRF (unknown variable due to lack of tournament experience), DCC (shorthanded), Stevenson (shorthanded?), and Loyola (several eyewitness accounts say they weren't playing very well, even when they won) ... if this is all true, then it seems hardly appropriate to attack the seeding or the format based on some "upsets" that may not have been true upsets, given the players actually present, and the level of play they were playing with that day.

Mechanical Beasts · Post by **Mechanical Beasts** » Wed Dec 23, 2009 4:35 pm

You assume that pool play would break the tie in record between A and B by head to head, which is (I sure hope) not true. You also assume that assertions about how Stevenson was shorthanded and Loyola was not playing well supercede the statistics that suggest that they were, through their prelim rounds, demonstrating more knowledge than some playoff bracket teams.

jonah · Post by **jonah** » Wed Dec 23, 2009 5:01 pm

Quote statistics all you want, but based on people who actually saw Loyola and Stevenson play, they were not playing well (Stevenson in large part because they were missing Zach). At least one of Loyola's players has admitted to me that they played poorly in the morning rounds, and while I haven't interviewed Stevenson, I feel very comfortable assuming that any of their players would agree with the statement "we would have done better if Zach had been playing with us".

I did the best job of seeding that I could, and am aware that the result wasn't perfect. I knew Zach from Stevenson was going to be gone, and I knew Jeremiah from Carbondale would be missing; I tried to take both of those situations into account. I didn't know OPRF was as good as they were, but having seen them several times during the day, they really are as good as their performance indicated. I did not know DCC would be shorthanded, and I underestimated Northmont a bit. The bottom teams were very hard to seed and I probably assigned their seeds poorly, but that's not what is at issue here.

For the record, we did ask coaches to tell us if they would be shorthanded or not. Carbondale told us they would be missing Jeremiah; Stevenson didn't say they would be missing Zach, but I had already been so told, so it didn't matter. DCC didn't mention that they would be missing anyone. I don't think any other teams who are relevant to this discussion neglected to tell us about anything unusual.

the return of AHAN · Post by **the return of AHAN** » Wed Dec 23, 2009 5:06 pm

Crazy Andy Watkins wrote:You assume that pool play would break the tie in record between A and B by head to head, which is (I sure hope) not true.

TEgan wrote: .... I have 48 teams in 8 pools of six .... with six teams advancing to the championship pool in the afternoon .... thus winners of the pools advance.

Team A is an elite superior team. The only team remotely close to them is Team B, though Team B is clearly inferior. Somehow, Team B defeats Team A, and given that the rest of the pool was properly seeded and all of the other teams are vastly inferior, A and B win out, and B advances to the championship pool.

I think you misread what he said; Team B goes 5-0, Team A goes 4-1; hence no tied records.

Crazy Andy Watkins wrote:You also assume that assertions about how Stevenson was shorthanded...

Stevenson was definitely minus Zach Blumenfeld, whose specialties complement Kevin's. If Zach's there, the Northmont loss is less likely (with all due respect to Northmont). At the very least, I guarantee the 70 point margin closes.
(EDIT: Jonah beat me to part 2)

Mechanical Beasts · Post by **Mechanical Beasts** » Wed Dec 23, 2009 5:14 pm

jonah wrote:Quote statistics all you want, but based on people who actually saw Loyola and Stevenson play, they were not playing well (Stevenson in large part because they were missing Zach). At least one of Loyola's players has admitted to me that they played poorly in the morning rounds, and while I haven't interviewed Stevenson, I feel very comfortable assuming that any of their players would agree with the statement "we would have done better if Zach had been playing with us".

I'm saying the fact that they were shorthanded/ mysteriously playing poorly is irrelevant, since they were both better than teams in the playoff bracket, The fact that they would have been even more deserving of a playoff bracket in other realistic circumstances is not our concern here. If you think that there's a problem when two of the top teams in PPB are left out of the top bracket because they played harder schedules, then you think that whether or not those playoff-deserving teams were shorthanded or not.

AlphaQuizBowler · Post by **AlphaQuizBowler** » Wed Dec 23, 2009 6:17 pm

Tegan wrote:So let me get this straight .... I have 48 teams in 8 pools of six .... with six teams advancing to the championship pool in the afternoon .... thus winners of the pools advance.

Team A is an elite superior team. The only team remotely close to them is Team B, though Team B is clearly inferior. Somehow, Team B defeats Team A, and given that the rest of the pool was properly seeded and all of the other teams are vastly inferior, A and B win out, and B advances to the championship pool.

In power matching, let's say that the preliminary seeding was done fairly well, and A and B meet in the third of five rounds. both are currently undefeated, and A is still superior to B. B still finds a way to win. A has one loss, but dropping down to face another one loss teams should win out and make it to the afternoon. Providing that B got "lucky", they should lose one or both of their last two matches ... possibly also advancing, and possibly not.

How is the pool more fair than power matching?

This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.

Mechanical Beasts · Post by **Mechanical Beasts** » Wed Dec 23, 2009 6:57 pm

AlphaQuizBowler wrote:This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.

How is only taking one team per pool worse than only taking two? You can lose only twice and be eliminated from championship contention!

AlphaQuizBowler · Post by **AlphaQuizBowler** » Wed Dec 23, 2009 7:09 pm

Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.
How is only taking one team per pool worse than only taking two? You can lose only twice and be eliminated from championship contention!

Are you being sarcastic? Either way, if someone has put a good-faith effort into creating the pools, it seems unlikely that the 3rd-best team from a pool is a playoff contender. The amount of teams you take from each pool really depends on the size and caliber of the field, but I find it hard to see a situation in which only taking one makes sense.

Mechanical Beasts · Post by **Mechanical Beasts** » Wed Dec 23, 2009 7:12 pm

AlphaQuizBowler wrote:
Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.
How is only taking one team per pool worse than only taking two? You can lose only twice and be eliminated from championship contention!
Are you being sarcastic? Either way, if someone has put a good-faith effort into creating the pools, it seems unlikely that the 3rd-best team from a pool is a playoff contender. The amount of teams you take from each pool really depends on the size and caliber of the field, but I find it hard to see a situation in which only taking one makes sense.

Ah, I've come up with one! When you have n brackets and between 2-n teams is a championship contender.

Don't worry about it, I'm patting myself on the back for that one.

jdeliverer · Post by **jdeliverer** » Wed Dec 23, 2009 7:35 pm

Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:
Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.
How is only taking one team per pool worse than only taking two? You can lose only twice and be eliminated from championship contention!
Are you being sarcastic? Either way, if someone has put a good-faith effort into creating the pools, it seems unlikely that the 3rd-best team from a pool is a playoff contender. The amount of teams you take from each pool really depends on the size and caliber of the field, but I find it hard to see a situation in which only taking one makes sense.
Ah, I've come up with one! When you have n brackets and between 2-n teams is a championship contender.

Don't worry about it, I'm patting myself on the back for that one.

And think of how exciting the morning matches would be if we put all the contenders in one bracket, producing dozens of great matchups all before lunch!

jonpin · Post by **jonpin** » Wed Dec 23, 2009 7:45 pm

Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:
Crazy Andy Watkins wrote:
AlphaQuizBowler wrote:This is a straw man argument. Of course a pool system that only takes one team from each pool is bad, that's why nobody doing pools should set them up that way.
How is only taking one team per pool worse than only taking two? You can lose only twice and be eliminated from championship contention!
Are you being sarcastic? Either way, if someone has put a good-faith effort into creating the pools, it seems unlikely that the 3rd-best team from a pool is a playoff contender. The amount of teams you take from each pool really depends on the size and caliber of the field, but I find it hard to see a situation in which only taking one makes sense.
Ah, I've come up with one! When you have n brackets and between 2-n teams is a championship contender.

Don't worry about it, I'm patting myself on the back for that one.

And you're absolutely certain that you haven't missed a competitive team and that there will be no upsets.

The theory of the Second Loss (that it should take two losses to knock you out of championship contention) is somewhat well-established nowadays. In my opinion, it should be refined to the second distinct team to beat you, to avoid a situation like could've happened at PACE '09 had Maggie Walker beaten GDS in the third-place game, where both MW and SC would've beaten everyone they played except for Charter and never had a chance at each other, but one was dubbed second place because they played Charter later in the tournament.

Single-elim preliminary groups is what you're seeming to suggest are reasonable, and I disagree because a team that could actually be the second-best at the tournament is eliminated way too early.

jdeliverer · Post by **jdeliverer** » Wed Dec 23, 2009 8:23 pm

Maybe I misinterpreted, but I think his last couple posts were sarcastic.

jonpin · Post by **jonpin** » Wed Dec 23, 2009 8:29 pm

jdeliverer wrote:Maybe I misinterpreted, but I think his last couple posts were sarcastic.

It certainly seems like it, but I don't grasp the serious point he's trying to advance.

Mechanical Beasts · Post by **Mechanical Beasts** » Wed Dec 23, 2009 8:39 pm

jonpin wrote:
Crazy Andy Watkins wrote:Ah, I've come up with one! When you have n brackets and between 2-n teams is a championship contender.

Don't worry about it, I'm patting myself on the back for that one.
And you're absolutely certain that you haven't missed a competitive team and that there will be no upsets.

Yes, that is in fact the circumstance under which I will do that thing, as I stated.

jonpin wrote:The theory of the Second Loss (that it should take two losses to knock you out of championship contention) is somewhat well-established nowadays. In my opinion, it should be refined to the second distinct team to beat you, to avoid a situation like could've happened at PACE '09 had Maggie Walker beaten GDS in the third-place game, where both MW and SC would've beaten everyone they played except for Charter and never had a chance at each other, but one was dubbed second place because they played Charter later in the tournament.

jonpin wrote:Single-elim preliminary groups is what you're seeming to suggest are reasonable, and I disagree because a team that could actually be the second-best at the tournament is eliminated way too early.

Yes; if you have n brackets and mix up seeds 1 and 2n-1 or 2 and 2n, then that will happen (disregarding upsets, for the moment); that said, if you do that you're a doofus. Since in the powermatching in question, it's every bit as hard to lose only once as it is in a bracketed system to lose zero times, I'm not convinced the possibility of greater upset protection is much consolation. I'd like to see more mathematical analysis.