International Saimoe League 2013

[KanbeKotori]

Arena 01: 綾崎ハヤテ 葉月渚 桐ヶ谷和人 [キズナ・アスティン（ハンプニー・ハンバート）] [狡噛慎也] 黒崎一護
arena 02: 赤司征十郎 [日向秀樹] [木下秀吉] 朽木白哉 真島太一 桜満集
arena 03: 伏見猿比古 はたけカカシ 紀田正臣 ルルーシュ・ランペルージ [夏目貴志] [セバスチャン・ミカエリス]
arena 04: シエル・ファントムハイヴ 衛宮切嗣 神田空太 クラピカ [折木奉太郎] [音無伊御]
arena 05: 青峰大輝 長谷部豊 キルア・ゾルディック [真奥貞夫] 沖田総悟 [音無結弦]
arena 06: エレン・イェーガー 氷室辰也 [上条当麻] [富樫勇太] ザークシーズ＝ブレイク 横寺陽人
arena 07: [赤坂龍之介] [不破真広] ギン イスカンダル 桂木桂馬 黄瀬涼太
arena 08: 阿良々木暦 クロム 火神大我 [高坂京介] [岡部倫太郎] 夜刀神狗朗
arena 09: [ルーミア] ナズーリン 水橋パルスィ 風見幽香 紅美鈴 [博麗霊夢]
arena 10: 村紗水蜜 古明地こいし 霧雨魔理沙 鍵山雛 [蓬莱山輝夜] [チルノ]
arena 11: 八坂神奈子 [鈴仙・優曇華院・イナバ] 古明地さとり 小悪魔 十六夜咲夜 [封獣ぬえ]
arena 12: 西行寺幽々子 [レミリア・スカーレット] 小野塚小町 洩矢諏訪子 [因幡てゐ] 聖白蓮

[iamadooddood]

Originally Posted by wontaek

1. Because we can't perfectly place everyone in correct seed, some fluctuation is inevitable. Mikoto's best schedule turned out, in hind sight, to be at the first 2 period where she failed to go 7-0. She went undefeated in the remaining 3 periods, but the best schedule part was already gone. Thus, in hind sight ( which I have to emphasize ) , Mikoto had following handicaps in winning the necklaces (1) Tier 2's lower expected SDO ( Please note that even if you use VD/VF , Tier 2 would likely have lower VD/VF as well, since they would have to face one more Tier 1 contestant assuming all the Tier 1 contestants are of same strength ) (2) Her losing the first match of the regular season ( This had profound psychological impact on the voters and likely to have haunted her in ALL NECKLACE GROUP MATCH - Do note that necklace group match is heavily impacted by irrational feelings of the voters, thus the reason for not wanting it as the sole judge for the necklace ) (3) Class of 2011 characters fans' desire to keep Mikoto in check ( This is because Mikoto is a class of 2009 who beat all the class of 2011 for the 2011 Tiara ) (4) Late appearance of group of voters who wanted Nova division to win the necklace ( This was very unorganized group comprised of diverse people who just happen to share the desire to promote class of 2013 girls just because they like new characters or don't want the old characters any more. Mikoto was unlucky because she was the obvious target at the first real chance Nova division got. Mikoto still scared the heck out of them in the necklace group match, which made them even more wary of any boost to Mikoto's chance. Since Mikoto did beat Kanade during regular season, I have to say their worry was understandable ) (5) Mikoto fans' misunderstanding that only necklace group match matters along with Kanade match. Other factions, especially the pro-Nova factions, had many people saying "it is better if A wins because ..... ", thus it was easy to know who to vote for if you want BBB to have better chance at the necklace. Mikoto fans didn't bother to make such posts and often didn't show up for most of the matches. Necklace was supposed to be for total overall performance within the necklace period, not just because someone shined in 1 or 2 matches. SDO was designed to enforce this idea. Many Mikoto fans were not aware of this, and they suffered for it.

(2) Let me reiterate something I said before. In 1st hour of voting, ISML usually throws away more definitely confirmed multivotes compared to all the votes J AST receives for the entire 24 hour voting period. There are dozens of multivoting groups, all supporting different characters, that continually tries to bypass whatever preventative measure we set up against them. It is unrealistic to think we can stop them all the time, so we need a system that will minimize damage in case they breach our multivote defence. Simple VD/VF using any average system is too vulnerable to even low level of multivotes if it escapes our detection for the entire necklace period. It also is vulnerable to just one massive burst came in a way that we couldn't properly mark it as multivotes . Even though I am a fan of RPI ratings, several number simulation has shown it to be more vulnerable compared to current "partial point for the loser + SDO system + Necklace group round" format we got.

3. Yes, I care about longevity of any system I help set up. I do not want ISML to fall apart in a year or two. Best Moe Tournament showed us that an event that attracted 10,000+ voters, internationally, can quickly fall apart. In my analysis, two of the main reasons for this was (1)Negative feeling generated by impression that 2 groups with about 2000 supporters controlled everything (2) The Tournament structure and attitude not being open to diverse groups of fans = There wasn't enough hook around for the fans of losing girls to hang around

4. When I talk about diversity, my main concern is to make sure that fans of characters at bottom half of the rank feeling relevant and occasionally be rewarded for sticking around, so we would have dozens of factions vying for attention instead of 2 or 3 factions. You need diversity in faction characteristics to make it possible for surprise results and also to let new and old girls take turns to shine in the spot light. It is likely that system which will reward stronger girls by letting them punish weaker girls to be much more likely to drive away weaker factions, nor give some emerging factions sufficient time to build up its strength to challenge the status quo. Despite the dominance of class of 2011, the list of postseason participants showed consistent introduction of new girls and retiring of the old girls. Most of postseason members from 2008 and 2009 now struggles to even make the regular season. Without using hard limiting rules like the "3 year rule" in Best Moe Tournament, nor limiting eligibility to 1 year window like Best Moe Tournament and J AST, ISML has achieved continual transition towards newer girls almost every year. We want people to show their support for both old girls and new girls. This is not possible without encouraging minor weak factions to stay around

5. A large faction of 2000 voters will get to have things go their way most of the times regardless of the system in place. I need to have a system that will keep them in check to not let 20% of the coordinated voters dominate the other uncoordinated 80% too much. These 20% will get rewarded no matter what system we use. I strive to find a system that will "protect" the uncoordinated 80% and also allows for a scenario where a faction that used to control 20% might weaken to be minor faction one year and then come back into dominance in the next year . This is why we haven't embraced the idea of retiring Tiara winners while they are still strong.

6. In the example, it was mentioned that two characters whose opponents had same average win percentage will have different SDO because one averaged 10% margin while the other averaged 7% margin. Two things to be aware of is that this SDO margin will translate to less than 1% difference in SDO, since this margin will matter only in 1 of the 7 matches her opponents have. Furthermore, the necklace group match is there to account for this. Current scoring formula is SDO/3 + group round . Even if a character is 3% behind in SDO, if she is just 1.5% ahead in the necklace group match, she will win since (-3)/3 + 1.5 = (+0.5) . A consistent vote for a character, for ALL EIGHT MATCHES in a necklace period, will come out NET POSITIVE.

7. Check this out to see what would have happened if we kept VD/VF rule for 2008.

http://internationalsaimoe.com/forum...art=20#p170975

Using VD/VF rule, these would have been winners for 2013,

Aquamarine : Tachibana Kanade
Amethyst : Tachibana Kanade ( Kuroneko if we enforce 1 necklace per year rule. By the way do you want this rule? What justification will you come up for this rule? I want to hear your justification before telling my reason for supporting it )
Ruby : Tachibana Kanade ( Shiina Mashiro with 1 necklace rule )
Emerald : Tachibana Kanade ( Takanashi Rikka with 1 necklace rule )
Topaz : Kuroneko ( Misaka Mikoto with 1 necklace rule )

Note that Mikoto might have won in Topaz, but her schedule, barring the Kanade match was REALLY WEAK in Topaz. So, VD/VF system without 1 necklace rule would mean absolute dominance by Tachibana Kanade ( a class of 2011 ), and with 1 necklace rules would give the necklace to those with LUCKIEST WEAKEST schedule. Contrary to claims of many, I find many cases where VD/VF depends more on luck compared to SDO

8. I must say I find it interesting that people started complaining about SDO when Yuuki Asuna took a clear lead while it was announced that Eucliwood would face Kanade in Ruby 7. Any one can tell me any reason for this?

9. Wouldn't you worry if 5% of the voters who happened to be coordinated will determine more than half of the things because the 95% is uncoordinated? That is what will happen with VD/VF scheme. 2008 showed us that.

1. So many coincidences? Btw your last sentence means the rules needed to be reiterated.

2. So I see. However, I still stand by my stance that SDO affects more than just one character. Think of it as splash damage, if you will.

3. If you say so. Personally I feel that a long-standing Saimoe competition that has completely thrown away what moe is, has lost its meaning. I don't know about Korea's Best Moe Tournament, but Japan's AST failed for completely different reasons (said reasons being sudden blocking of IP addresses, implementing of new rules halfway throughout the tournament, lack of transparency, perceived biasedness, etc).

4. Best way to do this would be to give the losing girl fractional points based on the absolute total number of votes she got. Fraction to be decided at a later point in time. This way, we can reward losers and motivate them to stay, and at the same time we can motivate winners to vote to ensure their victory; a win-win situation, no?

5. All I can say is, you wanted the regular voters, you got them.

6. Saying that the SDO margin will be made up by the voting consistency is no reason for lowering SDO for a character that is clearly more deserving. Not only does this damage the integrity of SDO, because the girl who is clearly more deserving to win has a lower SDO than the girl less deserving to win. This is the only such anomaly; everywhere else the girl has a higher SDO the more deserving she's seen to win. Furthermore, lowering SDO like so will disadvantage a character against another who has a better win percentage, more so than it already does.

Also, why SDO/3 and not SDO/total number of matches for that period? The latter makes more sense to the layman at first glance. SDO/3 will need some explaining.

Finally, the main problem is why SDO is being incorporated into the necklace match in the first place.

7. I have no problems with the 1 necklace per year. It awards more than one girl, after all. Actually, 2 necklaces per year is fine with me too. The line should be drawn at half the necklaces available for the year, though. Or the square root of the number of necklaces available. Or something.

Or we could try something like: if a character wins more than one necklace in a single year, the year after achieving that feat she's not allowed to win the same necklaces she won last year. So suppose Kanade won those 4 necklaces (Aquamarine, Amethyst, Ruby, Emerald) this year, and next year she's eligible again, and we have 5 necklaces again. She'll only be eligible for Topaz then.

And like I said earlier, the luck factor for VD/VF can be mitigated like so:

Originally Posted by iamadooddood

7: Yeah, I know. But we're talking about crushing victories, not narrow victories. I was thinking of a hybrid SDO-VD/VF system such that the weaker your opponent is, the larger a margin you have to win to make up this difference. For example, you have to win by a margin of say, 50% against a 4-3 opponent in order to be equivalent to someone who beat a 6-1 opponent by a margin of, say, 5%. (These numbers are only estimates; actual numbers are highly likely to be different.)

8. No idea.

9. If the votes were legit, then they fully deserved to win. The only thing I'd worry about would be my own faction, because I would be asking, why can't we be as coordinated as them?

Originally Posted by wontaek

1. Mikazuki Yozora (4240) - Yui (3473) : Check the RUBY period record, which was before Emerald. Yozora was 12th in Ruby while Yui was 13th. Yozora winning was not a surprise given that Yui and SAO girls were criticized for winning the Ruby. Also note that TRE predicted Yozora to win with 51% of the votes

2. Takanashi Tōka (3310) - Mikazuki Yozora (3853) : Takanashi Touka was 16th in Ruby. Again, Yozora was doing better going into the match. Surely you can't say Yozora's Ruby performance was in anticipation for Emerald?

3. Honma Meiko (3267) - Kirishima Shōko (3814) : Meiko did do better in Ruby, but Do look at Cumulative whole season SDO right before that match ( Please understand that SDO here is used as one of the indicator of strength for comparing these 2 characters ) At the end of Emerald 2, Menma had Cumulative SDO of 237.145 while Shouko had 252.674 . This strongly suggest that Shouko was likely to be stronger than Menma and her number was suppressed by strength of opponents she beat. Kirishima Shouko also beat Yozora before this match, suggesting that she is stronger than the numbers indicates. TRE predicted Meiko to win with 52%. Since she ended up with 46%, this is a 6% swing, a bit outside random noise. Still, do also note the Shouko went 3 - 4 for the Emerald Period while Menma went 4- 3 . What gain that was achieved by Shouko's win here probably was lost in other matches.

4. Nymph (4099) - Iwasawa Masami (5018) : Going into the match, it did seem Nymph should win the match. Still Look what happened in Topaz period which is after Emerald. Nymph went 1 - 6 in Topaz while Masami went 4 - 3. There are reasons to suspect Nymph was losing steam and Masami was rising again probably due to tendencies of new incoming voters. TRE did predict 54% for Nymph at this match, but I strongly doubt TRE would give that number again in hind sight. Also note that Masami went 1 - 6 for the entire Emerald Period. Whatever the gain that was achieved by Masami's win here was probably lost in the 6 other defeats.

5. Charlotte Dunois (3708) - Sakagami Tomoyo (3111) : Again, in Ruby, Charlotte was 21st while Tomoyo was 27th. Charlotte was expected to win . TRE predicted 53% for Charlotte

6. Sakagami Tomoyo (3198) - Kuroi Mato (3866) : Kuroi Mato was 22nd, right below Charlotte. If Tomoyo lost against Charlotte, we should expect her to lose by similar amount to Mato .

7. Fujibayashi Kyō (3808) - Tōsaka Rin (3537) : Rin was better in Ruby, but she was WORSE in Topaz. Rin was on her way down in Emerald, thus this result isn't much of a surprise. Also keep in mind that Kyou historically did better in ISML compared to Rin . TRE did predict 53% for Rin here, but this was Emerald 1, which means people could not have been certain who would be Mikoto's opponent. Rin ended up getting 48%. The difference is right at 5% borderline, which means we cannot completely rule out random fluctuation being responsible for this result.

I find that you should compare to both the period right before the period each match occured in, as well as overall. Comparing it from only one point of view, and only the one favourable to your argument at that, will make your post look really biased.

Also, what is the predicted probability of the fluctuations being serious enough to cause upsets? Attributing upsets to random fluctuations without giving probabilities for them is another way to look biased.

[mistress_kisara]

RENA 01: Ayasaki Hayate Hazuki Nagisa Kirigaya Kazuto [Kizuna Astin (Hampnie Hambart)] [Kōgami Shin'ya] Kurosaki Ichigo
ARENA 02: [Akashi Seijūrō] Hinata Hideki Kinoshita Hideyoshi Kuchiki Byakuya [Mashima Taichi] Ōma Shū
ARENA 03: Fushimi Saruhiko Hatake Kakashi Kida Masaomi [Lelouch Lamperouge] [Natsume Takashi] Sebastian Michaelis
ARENA 04: Ciel Phantomhive [Emiya Kiritsugu] Kanda Sorata Kurapika [Oreki Hōtarō] Otonashi Io
ARENA 05: [Aomine Daiki] [Hasebe Yutaka] Killua Zoldyck Maō Sadao Okita Sōgo Otonashi Yuzuru
ARENA 06: [Eren Jaeger] [Himuro Tatsuya] Kamijō Tōma Togashi Yūta Xerxes Break Yokodera Yōto
ARENA 07: Akasaka Ryūnosuke [Fuwa Mahiro] Gin Iskandar Katsuragi Keima [Kise Ryōta]
ARENA 08: Araragi Koyomi Chrom [Kagami Taiga] Kōsaka Kyōsuke Okabe Rintarō [Yatogami Kurō]
ARENA 09: Rūmia Nazrin Mizuhashi Parsee Kazami Yūka Hong Meiling Hakurei Reimu
ARENA 10: Murasa Minamitsu Komeiji Koishi Kirisame Marisa Kagiyama Hina Hōraisan Kaguya Cirno
ARENA 11: Yasaka Kanako Reisen Udongein Inaba Komeiji Satori Koakuma Izayoi Sakuya Hōjū Nue
ARENA 12: Saigyōji Yuyuko Remilia Scarlet Onozuka Komachi Moriya Suwako Inaba Tewi Hijiri Byakuren

[Frailty]

ARENA 01: Ayasaki Hayate Hazuki Nagisa Kirigaya Kazuto [Kizuna Astin (Hampnie Hambart)] [Kōgami Shin'ya] Kurosaki Ichigo
ARENA 02: Akashi Seijūrō Hinata Hideki [Kinoshita Hideyoshi] [Kuchiki Byakuya] Mashima Taichi Ōma Shū
ARENA 03: Fushimi Saruhiko [Hatake Kakashi] [Kida Masaomi] Lelouch Lamperouge Natsume Takashi Sebastian Michaelis
ARENA 04: Ciel Phantomhive [Emiya Kiritsugu] Kanda Sorata [Kurapika] Oreki Hōtarō Otonashi Io
ARENA 05: Aomine Daiki Hasebe Yutaka [Killua Zoldyck] Maō Sadao [Okita Sōgo] Otonashi Yuzuru
ARENA 06: Eren Jaeger Himuro Tatsuya [Kamijō Tōma] [Togashi Yūta] Xerxes Break Yokodera Yōto
ARENA 07: [Akasaka Ryūnosuke] Fuwa Mahiro Gin [Iskandar] Katsuragi Keima Kise Ryōta
ARENA 08: Araragi Koyomi Chrom Kagami Taiga Kōsaka Kyōsuke [Okabe Rintarō] [Yatogami Kurō]
ARENA 09: Rūmia Nazrin Mizuhashi Parsee Kazami Yūka [Hong Meiling] [Hakurei Reimu]
ARENA 10: Murasa Minamitsu Komeiji Koishi [Kirisame Marisa] Kagiyama Hina Hōraisan Kaguya [Cirno]
ARENA 11: Yasaka Kanako [Reisen Udongein Inaba] Komeiji Satori Koakuma [Izayoi Sakuya] Hōjū Nue
ARENA 12: [Saigyōji Yuyuko] [Remilia Scarlet] Onozuka Komachi Moriya Suwako Inaba Tewi Hijiri Byakuren

[Soverence]

ARENA 01: Ayasaki Hayate [Hazuki Nagisa] Kirigaya Kazuto [Kizuna Astin (Hampnie Hambart)] Kōgami Shin'ya Kurosaki Ichigo
ARENA 02: Akashi Seijūrō [Hinata Hideki] [Kinoshita Hideyoshi] Kuchiki Byakuya Mashima Taichi Ōma Shū
ARENA 03: Fushimi Saruhiko Hatake Kakashi [Kida Masaomi] [Lelouch Lamperouge] Natsume Takashi Sebastian Michaelis
ARENA 04: Ciel Phantomhive Emiya Kiritsugu [Kanda Sorata] Kurapika [Oreki Hōtarō] Otonashi Io
ARENA 05: Aomine Daiki Hasebe Yutaka Killua Zoldyck [Maō Sadao] Okita Sōgo [Otonashi Yuzuru]
ARENA 06: Eren Jaeger Himuro Tatsuya Kamijō Tōma [Togashi Yūta] Xerxes Break [Yokodera Yōto]
ARENA 07: [Akasaka Ryūnosuke] Fuwa Mahiro Gin Iskandar [Katsuragi Keima] Kise Ryōta
ARENA 08: Araragi Koyomi Chrom Kagami Taiga [Kōsaka Kyōsuke] [Okabe Rintarō] Yatogami Kurō
ARENA 09: [Rūmia] Nazrin [Mizuhashi Parsee] Kazami Yūka Hong Meiling Hakurei Reimu
ARENA 10: Murasa Minamitsu Komeiji Koishi [Kirisame Marisa] Kagiyama Hina Hōraisan Kaguya [Cirno]
ARENA 11: Yasaka Kanako Reisen Udongein Inaba [Komeiji Satori] Koakuma Izayoi Sakuya [Hōjū Nue]
ARENA 12: Saigyōji Yuyuko [Remilia Scarlet] Onozuka Komachi [Moriya Suwako] Inaba Tewi Hijiri Byakuren

[Triple_R]

@wontaek - I just want to thank you for the very in-depth and comprehensive explanations. At the very least, I now see where the ISML Staff is coming from on some things that raised eyebrows for me.

I'm going to take some time to fully digest and think on what you wrote. After that, I may reply to it. But in any event, thanks for the informative reply.

[wontaek]

Originally Posted by iamadooddood

I find that you should compare to both the period right before the period each match occured in, as well as overall. Comparing it from only one point of view, and only the one favourable to your argument at that, will make your post look really biased.

Also, what is the predicted probability of the fluctuations being serious enough to cause upsets? Attributing upsets to random fluctuations without giving probabilities for them is another way to look biased.

Indeed, I looked at the period before and after. Now, if a character's strength was changing in the Emerald period, which period should we refer to, before or after the change? If in any case, the character's strength was dipped ( or peaked ) only in Emerald, that would be VERY STRONG indication that some manipulation was happening. If the character's strength was changing and that change extended to periods before or after Emerald, then there is good room to consider that all the upsets may only felt such because the character was in midst of that change. Frankly, I expected at least one case where the winner of the problem match was weaker in BOTH the period before AND after Emerald. The fact that I failed to find such a case makes me think that at least the half of the reason for these upsets was indeed due to character's strength transitioning to different level.

Some manipulation was attempted, since we saw many people posting " Vote for BBB so CCC can win the necklace " type of writings. My big point is that only case these manipulation succeeded was in case the character's strength was changing for other reasons. When you have a character on the rise being cheered on by people with "ulterior motives", or when a character on a decline happens to meet another character that is important to certain faction's goal, these rise or decline in strength becomes much more accentuated. So, it is very likely that these matches were ripe for an upset to start with, and some faction's push for manipulation helped a little.

No matter which system is used, you will have cases where certain faction will attempt to manipulate the standing. That happens in real world sports as well. Any sizable faction will have some success at the manipulation no matter what kind of system you use. My big point is that currently, ISML has diverse enough groups of voters and factions, that most of these attempts fail unless it "luckily" involves characters already in decline or ascent . Most of the surprise upsets, in hind sight, and considering what transpired in period before and after the period in question, has almost always turned out to be something that was likely to happen. When you look at all the data, from 2008 to now, you will see that SDO manipulation attempts to have hard time overcoming random noise level of 3~5%, and often, even when it does, the random makes them lose whatever gain they got in the next few matches. Do consider that most of the listed as making the upset had very hard time winning other matches. If this was 2009 Rozen faction or 2010 Kugimiya factions pulling the strings, these characters should have went 5 - 2 and gave big boost to the tier 1( or 2) character of interest. You have to wonder if such manipulation was the reason when that the one who pulled the upset went 1 - 6 in the period, which isn't helping much.

Now as for probability of upset , this is where TRE and I might differ. I do NOT know what is the level of uncertainty TRE thinks his model have. The gold standard will be Gallup level which is 95% likely that the result will be in 3.4% of the predicted results. If we are talking 7 out of 7 * 36 matches, then 1 - (1/36) = 97.2% thus even Gallup people will quote more than 3.4% of level of uncertainty when you cherry picked 1/36 of all the matches. My experience with modeling suggests upset probability that can be simulated by assuming standard deviation value of 3% and then finding the Z-score, thus associated probability for it, in most cases related to ISML. Using STDEV value of 3%, chance that the prediction will be off by 6% or more is P ( abs (Z) > abs(+-6/3) = 2 ) which would be 2*0.0228 = 0.0456 = 4.56% . So such thing is expected to happen in one out of every 22 matches.

------------------------

Ok, My statements in the last paragraph has bugged me, myself, and I think I know a very good way to DISPROVE it. I like to ask you to compile the deviation between actual and TRE's predictions. Please keep the sign so if he predicted 51% but the actual was 48%, that should be -3% while if the actual was 54%, then it should be recorded as +3% . Please compile these values for at least all the rounds 2013 except for the Emerald. I would really prefer it if you would also compile values from the 2013 prelim, postseason and all the 2012 as well, since more data point likely would mean more accurate result. Then please tell us the (1) mean (2) STDEV (3) Median (4) 10th percentile value, and (5) 90th percentile value. Once we get these numbers, I can tell you much better number for the probability that TRE's prediction was off by 6 or more percent. If STDEV and/or 10th/90th percentile value strongly suggest deviation of 6% to be something extremely unlikely, then at least for that match, we have valid ground to suspect that some manipulation is played major role in causing this deviation.

[Rajura]

*DISCLAIMER: All males votes are made on the awesomeness factor; female votes hold to the baseline standard

ARENA 01: Ayasaki Hayate Hazuki Nagisa Kirigaya Kazuto [Kizuna Astin (Hampnie Hambart)] [Kōgami Shin'ya] Kurosaki Ichigo
ARENA 02: Akashi Seijūrō Hinata Hideki Kinoshita Hideyoshi [Kuchiki Byakuya] Mashima Taichi Ōma Shū
ARENA 03: Fushimi Saruhiko [Hatake Kakashi] Kida Masaomi Lelouch Lamperouge Natsume Takashi Sebastian Michaelis
ARENA 04: Ciel Phantomhive [Emiya Kiritsugu] Kanda Sorata Kurapika Oreki Hōtarō Otonashi Io
ARENA 05: [Aomine Daiki] Hasebe Yutaka Killua Zoldyck Maō Sadao Okita Sōgo Otonashi Yuzuru
ARENA 06: [Eren Jaeger] [Himuro Tatsuya] Kamijō Tōma Togashi Yūta Xerxes Break Yokodera Yōto
ARENA 07: Akasaka Ryūnosuke [Fuwa Mahiro] Gin [Iskandar] Katsuragi Keima Kise Ryōta
ARENA 08: Araragi Koyomi Chrom [Kagami Taiga] Kōsaka Kyōsuke Okabe Rintarō [Yatogami Kurō]
ARENA 09: Rūmia Nazrin Mizuhashi Parsee Kazami Yūka [Hong Meiling] [Hakurei Reimu]
ARENA 10: [Murasa Minamitsu] Komeiji Koishi Kirisame Marisa Kagiyama Hina [Hōraisan Kaguya] Cirno
ARENA 11: Yasaka Kanako [Reisen Udongein Inaba] Komeiji Satori [Koakuma] Izayoi Sakuya Hōjū Nue
ARENA 12: [Saigyōji Yuyuko] Remilia Scarlet Onozuka Komachi Moriya Suwako Inaba Tewi [Hijiri Byakuren]

[Ahasuerus]

ARENA 01: Ayasaki Hayate Hazuki Nagisa [Kirigaya Kazuto] Kizuna Astin (Hampnie Hambart) [Kōgami Shin'ya] Kurosaki Ichigo
ARENA 02: Akashi Seijūrō [Hinata Hideki] Kinoshita Hideyoshi Kuchiki Byakuya [Mashima Taichi] Ōma Shū
ARENA 03: Fushimi Saruhiko Hatake Kakashi [Kida Masaomi] Lelouch Lamperouge [Natsume Takashi] Sebastian Michaelis
ARENA 04: Ciel Phantomhive [Emiya Kiritsugu] Kanda Sorata Kurapika [Oreki Hōtarō] Otonashi Io
ARENA 05: Aomine Daiki [Hasebe Yutaka] Killua Zoldyck Maō Sadao Okita Sōgo [Otonashi Yuzuru]
ARENA 06: [Eren Yeager] Himuro Tatsuya [Kamijō Tōma] Togashi Yūta Xerxes Break Yokodera Yōto
ARENA 07: Akasaka Ryūnosuke [Fuwa Mahiro] Gin [Iskandar] Katsuragi Keima Kise Ryōta
ARENA 08: [Araragi Koyomi] Chrom Kagami Taiga Kōsaka Kyōsuke [Okabe Rintarō] Yatogami Kurō
ARENA 09: [Rūmia] Nazrin Mizuhashi Parsee Kazami Yūka Hong Meiling [Hakurei Reimu]
ARENA 10: Murasa Minamitsu Komeiji Koishi [Kirisame Marisa] Kagiyama Hina [Hōraisan Kaguya] Cirno
ARENA 11: Yasaka Kanako [Reisen Udongein Inaba] Komeiji Satori Koakuma Izayoi Sakuya [Hōjū Nue]
ARENA 12: Saigyōji Yuyuko [Remilia Scarlet] Onozuka Komachi Moriya Suwako [Inaba Tewi] Hijiri Byakuren

LMOST FORGOT; THANKS ks

[HasuMasu]

That was more-or-less just as difficult as the Girls Division...

[ninryu]

It is the bishies' turn. And Touhou, cause you gotta have Touhou.

[Chaos2Frozen]

I don't know anything about Touhou... 030

ARENA 01: [Ayasaki Hayate] Hazuki Nagisa Kirigaya Kazuto [Kizuna Astin (Hampnie Hambart)] Kōgami Shin'ya Kurosaki Ichigo
ARENA 02: Akashi Seijūrō [Hinata Hideki] [Kinoshita Hideyoshi] Kuchiki Byakuya Mashima Taichi Ōma Shū
ARENA 03: Fushimi Saruhiko Hatake Kakashi [Kida Masaomi] [Lelouch Lamperouge] Natsume Takashi Sebastian Michaelis
ARENA 04: Ciel Phantomhive Emiya Kiritsugu [Kanda Sorata] Kurapika Oreki Hōtarō [Otonashi Io]
ARENA 05: Aomine Daiki [Hasebe Yutaka] Killua Zoldyck [Maō Sadao] Okita Sōgo Otonashi Yuzuru
ARENA 06: Eren Yeager Himuro Tatsuya [Kamijō Tōma] [Togashi Yūta] Xerxes Break Yokodera Yōto
ARENA 07: [Akasaka Ryūnosuke] Fuwa Mahiro Gin Iskandar [Katsuragi Keima] Kise Ryōta
ARENA 08: Araragi Koyomi [Chrom] Kagami Taiga [Kōsaka Kyōsuke] Okabe Rintarō Yatogami Kurō


...And Chrom is so out-of-place

[Team Rocket Elite]

The returnees appear to be back on top form along with some promising newcomers. Kirito looks as strong as ever and ready to defend his title. Nagisa's performance here is pretty lackluster so Free doesn't looks like it'll be close to living up to the hype. Hideyoshi looks good and Hinata does better than I expected. Bleach characters have seen better days, though. Strong performance by Lelouch who wants his crown back. Kakashi does surprisingly well here. Sorata looks better than last year but still can't compete with Hotoro. Also a good performance by Sadao but similarly can't compete with Yuzuru. A pretty disappointing performance by Yuuta. It's not just he lost to Yeager who I think is pretty strong but he's not far ahead of Touma. It seems like Touma doesn't have much luck this time around. Being strong enough to get that close to Yuuta might have let him advance in a more normal group but he drew a bad group and doesn't advance. Similar to Sorata, Ryuunosuke has a good performance but also similarly can't compete with Keima who is another returner who looks to be in good shape. Poor Chrom. I'm happy to see he got in but this is not a contest he'll do well in since he doesn't have an anime. Saruhiko may have done terribly but Kurou is a different story. He edges past Kyouma and Kyousuke to advance. You can't help but wonder if Kyousuke would have advance here if he picked the "right" girl...

The Touhou is pretty far out there from what I know about the Touhou fanbase. Although my experience is mostly North American centric with a bit of Japanese knowledge. Reimu stomping everyone is expected. Meiling getting killed by Parsee is pretty out there. Rumia is hard to place. She seems a bit high but I wouldn't see it's too crazy for her to be where she is. Marisa and Cirno being neck isn't too far off expectations. I think Marisa is normally a more clearly the second most popular behind Reimu but she did lose to Alice that one time in ISML. Sakuya followed by Reisen also feels about right. Nue did surprisingly well, though. She got pretty close to beating Reisen. Remelia and Yuyuko and first and second once again feels about right. Byakuren did a lot better than I thought she would. I guess I should have put more faith in the Touhou 12 characters even if they weren't quite strong enough to advance.

[iamadooddood]

Originally Posted by wontaek

Indeed, I looked at the period before and after. Now, if a character's strength was changing in the Emerald period, which period should we refer to, before or after the change? If in any case, the character's strength was dipped ( or peaked ) only in Emerald, that would be VERY STRONG indication that some manipulation was happening. If the character's strength was changing and that change extended to periods before or after Emerald, then there is good room to consider that all the upsets may only felt such because the character was in midst of that change. Frankly, I expected at least one case where the winner of the problem match was weaker in BOTH the period before AND after Emerald. The fact that I failed to find such a case makes me think that at least the half of the reason for these upsets was indeed due to character's strength transitioning to different level.

Some manipulation was attempted, since we saw many people posting " Vote for BBB so CCC can win the necklace " type of writings. My big point is that only case these manipulation succeeded was in case the character's strength was changing for other reasons. When you have a character on the rise being cheered on by people with "ulterior motives", or when a character on a decline happens to meet another character that is important to certain faction's goal, these rise or decline in strength becomes much more accentuated. So, it is very likely that these matches were ripe for an upset to start with, and some faction's push for manipulation helped a little.

If a character was weaker in both before and after, that does not necessarily mean manipulation as well. If, for example, she was growing in strength period by period up to Emerald, and became steadily weaker after the same way, it likely means she reached her peak strength during Emerald.

So, looking at both before and after, what were the full details for each individual match?

Originally Posted by wontaek

No matter which system is used, you will have cases where certain faction will attempt to manipulate the standing. That happens in real world sports as well. Any sizable faction will have some success at the manipulation no matter what kind of system you use. My big point is that currently, ISML has diverse enough groups of voters and factions, that most of these attempts fail unless it "luckily" involves characters already in decline or ascent . Most of the surprise upsets, in hind sight, and considering what transpired in period before and after the period in question, has almost always turned out to be something that was likely to happen. When you look at all the data, from 2008 to now, you will see that SDO manipulation attempts to have hard time overcoming random noise level of 3~5%, and often, even when it does, the random makes them lose whatever gain they got in the next few matches. Do consider that most of the listed as making the upset had very hard time winning other matches. If this was 2009 Rozen faction or 2010 Kugimiya factions pulling the strings, these characters should have went 5 - 2 and gave big boost to the tier 1( or 2) character of interest. You have to wonder if such manipulation was the reason when that the one who pulled the upset went 1 - 6 in the period, which isn't helping much.

In most real world sports, manipulation is generally overlooked as long as none of the rules are broken (eg. resting players is allowed, match-fixing is not). Not like in here.

Originally Posted by wontaek

Ok, My statements in the last paragraph has bugged me, myself, and I think I know a very good way to DISPROVE it. I like to ask you to compile the deviation between actual and TRE's predictions. Please keep the sign so if he predicted 51% but the actual was 48%, that should be -3% while if the actual was 54%, then it should be recorded as +3% . Please compile these values for at least all the rounds 2013 except for the Emerald. I would really prefer it if you would also compile values from the 2013 prelim, postseason and all the 2012 as well, since more data point likely would mean more accurate result. Then please tell us the (1) mean (2) STDEV (3) Median (4) 10th percentile value, and (5) 90th percentile value. Once we get these numbers, I can tell you much better number for the probability that TRE's prediction was off by 6 or more percent. If STDEV and/or 10th/90th percentile value strongly suggest deviation of 6% to be something extremely unlikely, then at least for that match, we have valid ground to suspect that some manipulation is played major role in causing this deviation.

Huh, I'd have gone for (3/51)*100% = 5.88% in that case.

Unless they calculate it differently from the impression I have?

[Archaeon]

ISML Male tournament gain more votes than J-Saimoe?

How many of those voters in ISML for both female and male tournament are from Japan?

[mistress_kisara]

Kirito is still strong I thought his popularity would be waning by now..and I guess even in the male category female-oriented shows will have no chance..

[HasuMasu]

I didn't expect Sorata to win.

[wontaek]

Originally Posted by iamadooddood

If a character was weaker in both before and after, that does not necessarily mean manipulation as well. If, for example, she was growing in strength period by period up to Emerald, and became steadily weaker after the same way, it likely means she reached her peak strength during Emerald.

So, looking at both before and after, what were the full details for each individual match?



In most real world sports, manipulation is generally overlooked as long as none of the rules are broken (eg. resting players is allowed, match-fixing is not). Not like in here.



Huh, I'd have gone for (3/51)*100% = 5.88% in that case.

Unless they calculate it differently from the impression I have?

1. Yes. If there was only one instance where we had the character peak or dip for a period, then that might be mere luck or accident. If you have several of them do that, then we really need to worry. For SDO, the performance for entire period is much more important compared to merely one match for the opponents. If a character usually go 1 - 6, and the character does 1 - 6 again for the period, then it is possible to consider that sole victory as something expected from that character.

2. In ISML, it is rare to have a character's strength level change by the week. It usually takes at least a month for any change to fully manifest itself. So, if a character was weak in period 3 but strong in period 5, then at period 4, that character would be in some sort of Schrodinger's Cat like situation where she would be in weak/strong overlap case, where the opponent and the situation may make her look very weak in one match while rather strong in another. To push Quantum Mechanics/Logistic Regression talk a bit further, you can say a character's expectation value for her strength is slowly changing, but it is hard to know that while the change is occurring, because the wave-function collapse/expression of result is bit polar, either definitely weak or strong. ( Note: You can consider every match to be case for wave function to collapse/results to be expressed, thus you might have series of alternating cases of strong and weak while things are in transition ) You have to look at the whole thing in long term to understand that the expected value was in transition.

3. In case of TRE's prediction, what I want is the residues = actual percentage - predicted percentage. For example, if TRE predicted 54% for character A, but that character got 48%, then the residue from that match would be 48 - 54 = -6% . For Regular season matches in 2013, TRE posted 19 predictions in each round for the purpose of Fantasy games. That means, excluding Emerald period, since Emerald period is what we really want to investigate, our control sample is the 4 other period * 7 rounds * 19 predictions = 532 predictions which can be compared to the actual results. This means we can have 532 residue values calculated as Actual - Predicted. For those 532, I want to know the (1) mean, (2) STDEV (3) Median ( 4 ) The 10th percentile ( or 25th percentile if you want more text book stem-leaf plot ) , and (5) the 90th percentile ( or 75th ) value. Once we have these 5 values, we can then assign probability for each of the 7 events in the question that also have TRE's posted predicted value. If (1) - (5) shows that whole thing can be generalized as normal distribution and (2) STDEV value indeed is near 3% as per my guess, then we can definitely claim the deviation of 6% is something you expect to see once or twice every match day, thus it is more likely that Mikoto was unlucky compared to any large group of people deliberately working to sink Mikoto's chances. This statement can be disproved if (2) STDEV value is less than 2% ( Because that will make P ( |Z| > | observed value | ) too small ) or while the whole distribution can't be generalized as normal distribution, deviation greater than 6% still was something very unlikely in those 532 data points.

4. Faction campaigning to have people vote in a way they like isn't breaking any rules. Campaigning is encouraged. If they use computers to break the 1 vote 1 person principle, then they would be breaking rules. We hope that for all the votes, there exists an unique individual that can say " Yes, I knew I would be voting for AAA,......., ZZZ when my vote was submitted. "

-------------------------

After typing all this, I feel like playing the role of Bohr, Heisenberg, Schrodinger, and Born against Einstein in insisting there is no Hidden Variable in QM and that "god does play dice with ISML world" . Heck, I can stretch quantum entanglement analogy to some ISML situation, though they aren't that interesting here.

[exec]

ARENA 01: Gilgamesh Hijirikawa Masato [Ledo] Makishima Shōgo Matsuoka Rin [Yagami Light]
ARENA 02: [Akabane Raishin] Alibaba Saluja Orihara Izaya [Sakurada Jun] Tachibana Makoto Uchiha Itachi
ARENA 03: [Alis Color] [Hijikata Toshirō] Kyon Nanase Haruka Nase Hiroomi Togami Byakuya
ARENA 04: [Edward Elric] Itsuka Shidō Mikoshiba Seijūrō Naegi Makoto Natsume Kyōsuke [Okumura Rin]
ARENA 05: Ichinose Tokiya [L-elf Karlstein] Naoe Riki Okazaki Tomoya Sakamaki Izayoi [Sakata Gintoki]
ARENA 06: Doraemon [Hikigaya Hachiman] L Lawliet [Satō Jun] Takasu Ryūji Takigawa Yoshino
ARENA 07: Aikawa Ayumu Levi Roy Mustang Sakishima Hikari [Tsukiyomi Ikuto] [Usui Takumi]
ARENA 08: [Accelerator] [Heiwajima Shizuo] Komaeda Nagito Kuroko Tetsuya Ryūgazaki Rei Sakamaki Ayato
ARENA 09: Toramaru Shō [Shameimaru Aya] Reiuji Utsuho Mystia Lorelei Hinanawi Tenshi [Alice Margatroid]
ARENA 10: [Tatara Kogasa] Shiki Eiki Yamaxanadu Patchouli Knowledge Kamishirasawa Keine Flandre Scarlet [Chen]
ARENA 11: Yakumo Ran Kochiya Sanae Kaenbyō Rin [Ibuki Suika] Ibaraki Kasen [Fujiwara no Mokō]
ARENA 12: Yakumo Yukari Yagokoro Eirin Nagae Iku [Konpaku Yōmu] Kawashiro Nitori [Inubashiri Momiji]

[Kotohono]

ARENA 01: Gilgamesh Hijirikawa Masato [Ledo] Makishima Shōgo Matsuoka Rin [Yagami Light]
ARENA 02: Akabane Raishin Alibaba Saluja Orihara Izaya Sakurada Jun Tachibana Makoto Uchiha Itachi
ARENA 03: Alis Color Hijikata Toshirō [Kyon] Nanase Haruka Nase Hiroomi Togami Byakuya
ARENA 04: Edward Elric Itsuka Shidō Mikoshiba Seijūrō Naegi Makoto Natsume Kyōsuke Okumura Rin
ARENA 05: Ichinose Tokiya L-elf Karlstein [Naoe Riki] [Okazaki Tomoya] Sakamaki Izayoi Sakata Gintoki
ARENA 06: Doraemon [Hikigaya Hachiman] L Lawliet Satō Jun [Takasu Ryūji] Takigawa Yoshino
ARENA 07: Aikawa Ayumu Levi Roy Mustang Sakishima Hikari Tsukiyomi Ikuto Usui Takumi
ARENA 08: Accelerator Heiwajima Shizuo Komaeda Nagito Kuroko Tetsuya Ryūgazaki Rei Sakamaki Ayato
ARENA 09: Toramaru Shō Shameimaru Aya Reiuji Utsuho [Mystia Lorelei] Hinanawi Tenshi [Alice Margatroid]
ARENA 10: Tatara Kogasa [Shiki Eiki Yamaxanadu] [Patchouli Knowledge] Kamishirasawa Keine Flandre Scarlet Chen
ARENA 11: Yakumo Ran [Kochiya Sanae] Kaenbyō Rin Ibuki Suika [Ibaraki Kasen] Fujiwara no Mokō
ARENA 12: [Yakumo Yukari] [Yagokoro Eirin] Nagae Iku Konpaku Yōmu Kawashiro Nitori Inubashiri Momiji

Thought about voting for Kyōsuke, but Refain turned me away from voting for any of the LB males besides Rikki.