The OU List

[Hyra]

Dragontamer suggested 99% as a cut off. I think his list had like 57 Pokemon being OU, including Frosslass.

I don't care which is the cut-off, but 75% allows for a larger UU metagame.

 

[Aldaron]

I concur with everything you stated, X-Act. I do however have questions regarding the 75% cut off.

I see that you already pointed out that it is an arbitrary cut off, but I am wondering why we reach that particular one.

For example, and this purely speculative and assuming a lot, let's say the usage statistics form a normal distribution, would 68% be a better cutoff? To represent the first standard deviation in terms of usages?

Granted, assuming it is a normal distribution is completely arbitrary, yet, type of distribution aside, should we try and define usages in terms of standard deviations?

Speaking completely qualitatively, the basis for this is that in normal distributions there are ways of making holistic assumptions about each specific standard deviation range; all who fall within the first standard deviation are average, within the second good and within the third elite (the positive side of the spectrum). 105 I.Q., for example is average; 120 is good and anything above 130 is considered to be "elite."

 

[X-Act]

@Hyra: Dragontamer's calculations had an error.

@Aldaron: I could have done that. However, I used a cumulative frequency, so I'm using percentiles and not standard deviations. The 75% means that the OU Pokemon are above the lower quartile of usages. If you want, we could change the percentage.

EDIT: I analysed the data and I concluded that the usage of Pokemon does not follow a normal distribution. So we cannot use standard deviations, unfortunately.

 

[X-Act]

After continuing to research on this, I finally found out that Pokemon usage does not indeed follow a normal distribution, but an exponential distribution. Lambda seems to be around 0.0259 for the weighted list.

The interesting thing is that the mean and standard deviation of an exponential distribution are equal. Also, in an exponential distribution, a whopping 86.47% of all usages would lie within one standard deviation from the mean, which is 77 Pokemon in all.

I'm not implying to use 86.47% as our cutoff point, of course, but to show that we need to be careful with our choice of the cut-off point.

 

[chaos]

Do you have a cutoff point you'd like to use? Of course, we could use our better judgement and just PICK a cutoff number everytime we update the list.

 

[monkfish]

I'd just like to point out that if UU battles are popular then some UU pokemon will see a statistical rise in usage and may get banned from UU next 'quarter'. Then they will be in the OU category and nobody would use them because they aren't powerful enough for OU battles - creating a cycle of pokemon that are allowed in UU. Not really desirable imo.

If there was a way to discriminate between OU and UU battles when getting the usage statistics then the problem would be solved.

 

[Aldaron]

Nice work discovering the actual distribution, X-Act.

Upon this discovery, I ask, why not have 77 pokemon declared as OU?

BL is NOT a metagame; it is simply an area where all pokemon who subjectively determined to be too strong for UU play are thrust into.

Setting 77 pokemon as the line for OU would not even lower the number of pokemon available for UU. Among the 77 most used in the weighted list for Shoddy, only three are currently not considered OU or BL (I am ignoring obvious upgrades, such as Tentacruel and Mamoswine to OU or BL at least). Those would be Clefable, Poliwrath and Crobat, all of whom, at least with my singular battle experience on Shoddy, more than deserve BL status (but that is beside the point).

All having 77 pokemon declared as OU would do is shrink the BL list, which would finally get people to stop trying to establish some sort of BL metagame. It would establish a much clearer line between OU and UU, as the BL list would no longer be so controversial.

And, of course, from a statistical standpoint, it makes intuitive sense to define "over used" as all those falling within one standard deviation.

I 100% support declaring the new line for OU to be the 77 most used pokemon. It would give statistical basis for the list, reduce the BL category and more clearly define the difference between OU and UU, all without decreasing the pool of available pokemon for UU.

 

[X-Act]

@chaos: I was trying to find a cut-off point that makes sense statistically, so that people will not contest it. Picking a cut-off point ourselves will surely be controversial in my opinion. Basing the cut-off point on hard statistics won't.

@m0nkfish: I don't know how the Shoddybattle ladder matches are, but I'm assuming that ladder matches are always OU. I'm guessing that if you'd want to play with a UU in Shoddy, then you either play a nonladder match (in which case its statistics are not registered), or else pit your UU team against a potential OU team in a ladder match.

@Aldaron: I think that the OU Pokemon should be a much smaller number of Pokemon. Also, from what I've been reading, saying that the data within one standard deviation from the mean is the 'normal' data is only applicable for distributions having skewness zero, i.e. having the same mean and median (like the normal distribution). I'm working on it right now.

 

[junior]

IMO, a cut off point something around 50 is a good point.

Cutting off the OU tier at 70 is a little bit too high. If this was the case, a lot of the BL "tier" pokemons would be push up to OU, thus limiting the number of BL pokemons as well as restricting the pokemons we could use for that particular tier.

X-Act, shoddy battles are generally OU teams. In some rare cases, people use BL teams too so they'll be used a lot less often. UU teams almost has no appearence.

 

[X-Act]

Thanks Junior for the information.

Anyway, this whole exponential distribution and skewness thing shows that we cannot use the standard deviation to determine the cut-off point.

After some thought, I'll suggest that 75% of the data be the cut-off point, as I said before. This is the upper quartile, which is at ln(4)/lambda (assuming it's a perfect exponential distribution). Since lambda will probably be somewhere between 0.025 and 0.027 (it is 0.0262 for October and 0.0259 for November), OU would comprise around ln(4)/0.025 to ln(4)/0.027, or around 51 to 55 Pokemon. (It would be 49 for November because the distribution is not a _perfect_ exponential distribution but is very close.) Of course, lambda will change when the statistics of October, November and December together are taken into account.

Thanks for all your input, everyone. So shall I go ahead and fix the OU tier as soon as the December stats are posted in the Shoddybattle website?

 

[Time Mage]

I'd like to make a suggestion, however. Instead of just taking the data of the last three months, maybe using the data of the last 6-12 months (not now, obviously, but when it becomes is available), and weight it accordingly would help smoothing the results, so that there aren't any sudden jumps from one tier to another from month to month.

I go in more detail in THIS post. Please, take a look at it.

 

[Dragontamer]

@Hyra: Dragontamer's calculations had an error.

Yes, they did. I am sorry >_>

As far as the actual percentage...

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

There is one last thing before we go ahead and start changing the OU tier around. There was a bit of discussion about whether the OU tier should try and predict the curve, or whether it should represent the curve.

If we always try to represent the curve, we'll be always 3-months behind as far as the metagame is concerned. However, we'll have solid statistics to back all of our choices.

If we try to predict the curve, we'll probably be more or less up-to-date (as long as we make somewhat frequent updates), at the obvious cost of losing our statistical background.

The thing is, anyone can look at Shoddy for the statistics list. And btw: looking at the curve this way means that Tentacruel is not a fad, lol. Its grown in popularity >_> So predicting the curve at this point makes tenta definitly OU...

I'm fine with either way. But I thought I'd bring up the issue.

 

[X-Act]

@Time Mage: I read your post. I would like to make it clearer (since it doesn't seem that clear) that the OU list will not change every month, but every three months. That's why I'm considering the data of the previous three months. Weighting the usages of October, November and December is an interesting idea, though, since it gives preference to what was being used recently (in December). I was thinking to weigh them 9/19, 6/19 and 4/19 respectively. They add up to 1, and this way, the October stats are 2/3 as relevant as those of November, and the November ones are 2/3 as relevant as those of December. Maybe the 2/3 can be changed to a smaller or larger number, though. 1/2, perhaps, or 3/4. Any suggestions?

@Dragontamer: I am no prophet, nor am I _THAT_ good at statistics, so I cannot predict what will be used in January, say, from the October-December data. That means that the OU list will always be representing what is currently being used, not what will be used.

@all: So I'll have to ask your opinion again on three things, lol:

1) Should we go for the three months having the same preference, or should we give preference to the last month? (My suggestion: give preference to the last month.)
2) If we give preference to the last month, how will the preference be? (My suggestion: October is 2/3 as important as November, November is 2/3 as important as December.)
3) Should we update the OU list every three months, or maybe every month is better (using the last three months either way)? (Either way is fine with me.)

 

[Time Mage]

1) Should we go for the three months having the same preference, or should we give preference to the last month? (My suggestion: give preference to the last month.)
2) If we give preference to the last month, how will the preference be? (My suggestion: October is 2/3 as important as November, November is 2/3 as important as December.)
3) Should we update the OU list every three months, or maybe every month is better (using the last three months either way)? (Either way is fine with me.)

1) I prefer a weighting.

2) That weighting, for three months, seems fine. The only way to get a really good weighting is through experimentation, though, but we don't have enough data yet, so I think that's fine.

3) As I said, I'd prefer a middle ground. Updating every month is too often, and I don't think it's good for the community to have such a dynamic OU list. People shouldn't be constantly worried about knowing the OU list every month. However, updating the OU list every three months, and only counting three months makes the list being based in totally new data every time. That is not bad per se, but as I commented, the goal of my suggestions is to make the fluctuations as small as possible, so taking into account only the new data each time doesn't seem ideal to me. As a starting point, I think it' fine, but as more months are added, I think they should be taken into account, with lesser weights. They work would be to add stability to the list.


And speaking of stability: What about minimal fluctuations in the OU border? Has a pokémon that was at 75.1% in the previous list the right to be labeled "OU" if it's at 74.9 three months after? Such small fluctuations seem pointless to acknowledge to me, for two reasons: One because they only add confusion. And two, because those pokémon jumping up and down will most probably be BL pokémon, who already compete in the same tiers as the OU ones, so including them in OU won't really change anything. That's why I suggested in my post a minimum % increase/decrease to grant a promotion/demotion.

 

[X-Act]

...as I commented, the goal of my suggestions is to make the fluctuations as small as possible, so taking into account only the new data each time doesn't seem ideal to me. As a starting point, I think it' fine, but as more months are added, I think they should be taken into account, with lesser weights. They work would be to add stability to the list.

I don't see why the list should be that stable in the first place if the metagame changed within three months. If, three months from now, Gallade, say, is in half of the teams, then it should be OU in three months from now, and if Zapdos, say, is used by nobody, then it should be BL. The stability of the OU list depends solely on the stability of the metagame. If the metagame is stable, the OU list will be relatively the same after three months. If there were huge changes in the metagame, then the OU list should reflect those changes and have a huge overhaul.

And speaking of stability: What about minimal fluctuations in the OU border? Has a pokémon that was at 75.1% in the previous list the right to be labeled "OU" if it's at 74.9 three months after? Such small fluctuations seem pointless to acknowledge to me, for two reasons: One because they only add confusion. And two, because those pokémon jumping up and down will most probably be BL pokémon, who already compete in the same tiers as the OU ones, so including them in OU won't really change anything. That's why I suggested in my post a minimum % increase/decrease to grant a promotion/demotion.

Again I don't see a problem with a Pokemon jumping from OU to BL to OU, if its use dropped and then increased. After all, we're talking about a three-month wait for each jump, so it's not like we're seeing huge fluctuations going on. Maybe the word "borderline" for BL would have more than one meaning here, lol. They are "borderline" for a reason. It simply wouldn't be fair for Alakazam, say, being left in the OU list if it was at 74.6% and then goes to 75.2% after three months, while Dragonite, say, goes to BL if its percentage goes from 67.9% to 75.1%. They should both go to BL. If 75% is the cut-off point we agreed upon, then we stick to it.

 

[Time Mage]

The stability of the OU list depends solely on the stability of the metagame.

You've convinced me with that single phrase. No point in forcing stability to something that should be stable enough by its own.

 

[Dragontamer]

Again I don't see a problem with a Pokemon jumping from OU to BL to OU, if its use dropped and then increased. After all, we're talking about a three-month wait for each jump, so it's not like we're seeing huge fluctuations going on. Maybe the word "borderline" for BL would have more than one meaning here, lol. They are "borderline" for a reason. It simply wouldn't be fair for Alakazam, say, being left in the OU list if it was at 74.6% and then goes to 75.2% after three months, while Dragonite, say, goes to BL if its percentage goes from 67.9% to 75.1%. They should both go to BL. If 75% is the cut-off point we agreed upon, then we stick to it.

Well, going into the "future" thing... if it was OU once in the past, there is a good chance it will be OU again in the future.

We can keep this mechanical if you want. A pokemon has to drop to say... 72.5% usage before it is BL, and a pokemon has to grow to 75% usage before it is OU.

Although, things like Rhyperior make me really wonder. That thing truely has a "BL" feel to it by now...

 

[X-Act]

We can keep this mechanical if you want. A pokemon has to drop to say... 72.5% usage before it is BL, and a pokemon has to grow to 75% usage before it is OU.

So you're saying that if something drops from 68% to 73% it's BL, and if something climbs from 80% to 74% it's OU? That doesn't make sense to me. One cut-off point is the way to go. If you think the cut-off point is too high, then we lower it.

And about Rhyperior, it would be BL anyway using Shoddy's statistics.

 

[X-Act]

Okay, so after analysing the Pokemon usages even further (as if I haven't analysed them enough already), this is what I will do come January:

1) Take the October, November and December weighted lists from Shoddybattle's statistics.
2) Weigh the months further as follows: October - multiply each weighted usage by 19.1%; November - multiply each weighted usage by 30.9%; December - multiply each weighted usage by 50%. This is a slight change.*
3) Add up these weighted usages for every Pokemon, and produce a percentage cumulative frequency.
4) The Pokemon having percentage cumulative frequency of 75% or less are put in OU.
5) Pokemon having more than 75% cumulative frequency will be put in BL if they currently are in OU, otherwise they will be left in the tier they currently are in.


* I wanted the October and November statistics taken together to be as relevant as those of December alone, so I put December as 50%. I then wanted the October stats to be x% as relevant as those of November, and the November ones to be x% as relevant as those of December. It turns out that x must be roughly 61.8% for all this to hold. Thus, the November weighting becomes 50% x 0.618 = 30.9%, and the October weighting becomes 30.9% x 0.618 = 19.1%. Checking, 30.9% + 19.1% = 50%, as I wanted. Incidentally, this approximate value of 0.618 can be written exactly as (sqrt(5)-1)/2, which is the reciprocal of the so-called Golden Ratio. The percentages can also be written exactly as (3-sqrt(5))/4 = 19.1%, (sqrt(5)-1)/4 = 30.9% and 1/2 = 50% for October, November and December respectively.

 

[obi]

Some Pokemon that are currently OU could probably be dropped down to UU, rather than BL.

 

[X-Act]

Which Pokemon are you referring to, Obi? Looking at the current OU list, I'd only consider Yanmega and maybe Donphan to be UU if they were not overused - all the rest, in my opinion, would be BL if they were not overused. I'd like a list of current OUs which would be UU and not BL if they were not frequently used.

 

[Dragontamer]

Yanmega is seriously not UU >_> Tinted Lens Specs Bug Buzz does like 40% to 50% to max/max Calm Tentacruel, 42% to 50% to max/max Noctowl and has a chance to lower Sp. Def or crit. In either of those lucky cases, its a 2-hit KO. With stealth rocks, its a good chance for a 2-hit KO for these top UU walls. (If you test it in Shoddy... I'm almost sure there is a bug in Shoddy that makes it not work in strange cases, but I'm still investigating the specifics. It happened again today... Colin said he fixed it the other day but I'm not sure if he's reset the server yet to implement the fix or maybe he fixed another bug with tinted lens or something...)

Anyway, no UU wall can handle repeated Specs Tinted Lens Bug Buzz. Tentacruel and Noctowl are among the best who aren't weak to Bug Buzz.

Every metagame needs more spinners, so I'd welcome Donphan in that respect, but 120 Atk with those huge defenses means that I don't think that UU can really handle him from a logical point of view.

So you're saying that if something drops from 68% to 73% it's BL, and if something climbs from 80% to 74% it's OU? That doesn't make sense to me. One cut-off point is the way to go. If you think the cut-off point is too high, then we lower it.

If something drops to 73% from 76% it is still OU. Drops to 72% it is then BL.
If it grows from 71% to 73%, it is BL. If it grows all the way to 76% then it is OU.

This way, possible fads that are on the OU / UU boarder (Froslass is a potential example) have a chance to prove to us that they really aren't fads.

 

[X-Act]

Honestly that is why I was asking Obi about which OU Pokemon would go to UU if they were not used enough, because I cannot see any Pokemon of the current OU list which wouldn't overcentralise the UU metagame. That's why I asked for specifics from Obi. Also, as regards to spinners, I think DP has Claydol, Cloyster, Sandslash, Torkoal and Blastoise all in UU... seriously that's enough to choose from and none of them are of the level of Donphan.

About the cut-off percentage(s), I'd rather not commit myself to your further point, Dragontamer, since it would be an unnecessary complication in my opinion. The OU list will only be changing every 3 months, so we won't see frequent fluctuations anyway. Thanks for your suggestion, though.

 

[chaos]

Sounds good to me. Thanks for your effort man. Do me a favor and update the tiers document in Blue Kirby's thread with an explaination your process for DP. It'd be cool to mention that RB/GS/RS are stable as well...

 

[Hipmonlee]

What is a concern for me is the possibility of staple UUs suddenly becoming OU.. For that reason I think the OU list should be updated as little as possible.

Have a nice day.