New Porto meta after arbs got 1g cheaper, organ guns 3g?
I give up trying to explain to you thhat more data doesn’t matter here.
For me having with a 99% a result that is less than 1% of from the true winrste is good enough. Maybe you require 99.9% chance to be less then 0.1% off. Maybe we have enough data for that in 2-3 years.
I agree on that…
332 games 1v1 (1250-1650 elo) = 43.40% winrate
142 games 1v1 (1650+ elo) = 42.96% winrate
426 games Team (1250-1650 elo) = 45.56% winrate
572 games Team (1650+ elo) = 47.00% winrate
More data would be good? Sure… But this is enough to take some conclusions.
Thanks. I was about to bring similar data (but i have to go to bed…)
Aoestats provide the total number of games, so everyone can calculate the portuguese numbers by using pickrates.
The thing is, those numbers applies for all maps. We need data for arabia, arena, and water maps.
Arabia and arena are played a lot, but water maps not so maybe data from water wont be enough big (i beg someones to check this 
)
not very large sample sizes.
water map sample sizes are miniscule.
I recently watch the statistics. And that is not enough. Does not make sense yet.
But I would say it is relativ good. So my opinion they still need a buff but Portuguese are not bad civ.
The data are still very unstable, for instance it turns out that vietmamese are a bad civ, which does not follow the common sense.
I would say that Portuguese are now in better spot that Italians and Turks (which are the two weakest civs imo). However Portuguese may be given something more. Not sure however if this has to take place in the early game. Probably a solid imp UT fix/buff would be enough
1650+ 1v1 elo would be the best data to analyze, those are players that know how to take the civ advantages, if they can’t make a civ work then it probably means there is a problem. Sadly it takes a lot of time to gather sufficient data for this bracket.
1250-1650 data is not that bad, but I think 1250 to 1400-ish players barely know how to really use every characteristics of the civs. Overall is a good data to check winrates because it has a good sample size.
Bellow 1250 elo the winrates may be weird… players are still learning the game. But at least you get A LOT of data lol.
For Team games I think only 1650+ data is relevant.
Agree. They may be at koreans level…
These data seem to be too noisy. Koreans have a higher winrate on Arabia than Britons.
Tbh, we are trying to say that the “winrate number” is representative of the probability of victory of a civ.
We all agree that the probability of victory should the reference for the balance, but it does not seem to me that we can model the probability of victory with the winrate.
This approximation (probability of victory = winrate) implies results which are clearly wrong like in Arcipelago Bulgarians are much stronger than Persians, or Vietnamese are the worst Arabia civ. If this approximation brings these results I think there are 2 possibilities:
In teamgames the winrate is a poor metric since you always have allies. And then pick rate dictates which allies you got mostly. It is all very concoluted and biased towards the true win of the most picked civs.
This pulls everything towards some number around 50.
I tried proving this mathematically but well, I am a bit out of shape and failed.
In 1v1 we still have the pick rate problem, for example if goths pick rate rises, mayans winrate is expted to go down.
This map pool has fewer water maps, which results in lower winrate for Italians for example.
only civ matchup and map specific winrrates could adjust these effects.
the amount of data for that isn’t available.
But the winrate is a good metric for when you don’t don’t know the map nor the opponent prehand (so mappool and enemy civ winrste affect this) .
Once they are chosen then the conditional propabilites are needed. (basically I am talking about the beyers theorem)
This is definitely correct, the probability of victory clearly depends on the map and on the matchup. However it makes sense to me to fix a reference scenario (1v1 arabia ofc), and to think of the probability of victory over there (i.e., along all the possible matchups).
In this respect, the picking rate at high level (or better in tournaments) is a more reasonable metric of the probability of victory. In fact we can comfortably assume that every player picks a civ in order to maximize the probability of victory (this is quite sure in tournaments where you win money).
Note that this is way less noisy. It never happens that in a 1v1 arabia tournaments Britons have a lower picking rate than Koreans, while it may happen than in some days the winning rate of Koreans is larger.
The main point, here, I guess is civ pick vs random. The higher the pick rate the higher the probability for each game that the respective civ was picked instead of being played by going random. Maybe for 2k+ that doesn’t matter too much in terms of win rate but for most levels it does. People that pick civs usually are better prepared to play with them. So when Mongols have 5x the pick rate of portoguese, in most of their games they will have been picked while not so for portos.
First of all, you have to determine which samples are relevant. Since we are talking about balance, here, imo we should exclusively look at 1650+. Sure we can talk about the others and make assumptions about their meaning but when we seriously want to infer conclusions for possible balance changes beginner and intermediate levels are highly problematic to incorporate. This leaves us with 150 games for portuguese and there is really no argument for that being a sufficient sample size.
It is not, but it can indicate a trend.
