So you may have heard of a stats site called aoestats, which gives data on how often people win and play as each nation. The highest win rate is Franks, aoestats says that the Franks are the most played nation and have a 54% win rate against any nation on Arabia.

So you might think that means if a player is playing Franks and their opponent is playing random, they would have a roughly 54% chance of winning. However, this is not true.

Let’s say you only play Franks, every time you win, your ELO raises, and you go against more challenging opponents. Eventually, you will reach an ELO where you win 50% of your games, no matter how good Franks are. Even if you play a selection of nations, you skew the results if you don’t play random.

The result is that the person playing Franks a lot is getting a win rate closer to 50%. The person who plays random or a large selection receives a much higher win rate when they roll Franks, which bumps them up to 54%. Franks are picked at least nine times more often than Franks are rolled randomly, so the stats will show their win much closer to 50% than it should be.

Because the Frankish win rate is consistently high, people pick them more, leading to their win rate getting even closer to 50%, without them being balanced. It is impossible to tell what Franks real advantage is from the stats given. If you could filter the results by only random picks against random picks, it would give you a much better picture of the balance, though. Although the real win rate is likely somewhere in between 54% and 90%

Franks are top pick for arabia but they aren’t OP for 1v1, once you know how to play vs thme, Franks are less than a issue.
However, in teamgames, especially for 4v4 arabia pocket position, they are just broken, with basically none of the paladin civs being able to compete with them, is that stupid Chivalry tech to start the paladin spam.

note that Franks retain their high win rate. I suspect that would drop off if we could see 2000+ ratings though.
It is important to note that a broader range of nations are picked at the higher levels, this is likely due to higher level players going random nation more often. This would make these stats more accurate (still not very accurate) to the actual win rates.
All stats are for 1v1. Team games are omitted.

Cool now write the same post with all the reasons why their win rate could be lower than the stat shows, such as the civ being easy to play, TAs being hard to counter for many civs etc.

It would be interesting to see win rates only from games where both players used random civs. I’m not sure if that is data that can be obtained.

I don’t personally find Franks to be very strong. I have a poor win rate when playing as them, and a >50% win rate when playing against them. It’s just so predictable, you’re going to see Cavalry so have spears or camels ready. There will be many cheap castles, likely defended by axemen so don’t bother with rams. Archers + Pikes + Trebs usually do the trick, sometimes mix in more siege like Scorpions or Bombard Cannons.

There is a lot of uncertainty in all of this. This is a thread about misleading statistics rather than game balance. I don’t know the game well enough to attribute the higher win rates of some nations to ease of play, better bonuses or some psychological advantage. All I know is that the statistics from this site undersell the benefit of some nations over others.

The maths is also not entirely clear cut. ELO mismatches also mess with the results. If high-level players more commonly pick franks than low-level players, that would also contribute to a higher Frank win rate, but likely not by 4%.

I do want to say, this problem cannot be imagined, it is statically impossible for this not to have any effect on the win rate unless either all the civs are perfectly balanced with equal chances to win in any matchup or everyone plays Franks exactly 1 in 11 games

Don’t get me wrong. I am okay with other people discussing game balance here. But I do not feel comfortable discussing it myself (beyond the statistical side). Suppose anyone wants to challenge my points mathematically. In that case, I am open to discussing it, but asking me to comment on if the Franks are new player friendly is something I don’t feel qualified to do, especially as I started playing this game 17 years ago.

Actually picking civs is something that breaks the stats, only worth values should be winning rate from random civ games, as for the pick preferences as an indicator of how much broken are the civs.

Franks are not balanced, they have eco/military bonus right now from the start to the end of the game, solid army composition, faster creation, etc. The civ can be played aggressively, defensive, booming, doing everything at the same time, their only weakness are water maps, but on land they are not balanced.

Gotta be honest and say it made less of a difference than I was expecting. For reference this is one of many cohorts that I’ve put together, the full set can be found here:

Additionally I also tried to put a simulation together to demonstrate the magnitude of the effect you are describing under different scenarios (it’s on my to-do list to try and expand the simulation out a bit more)

Finally! Finally somebody told this simple thing!
Win rate around 50% – is a nature of MMR system. MMR system is designed to provide 50% win rate for every player. And hence can’t be used to assess state of civilization balance.

This is only true if everyone played 100% with the same civ. Because they don’t we still do get useful information about the state of civ balance but that information is distorted and biased by the pick rates and needs careful interpretation. From my own testing the bias induced generally wasn’t as much as I was expecting it to be

Interesting, I am surprised that you concluded that the difference is less than you expected because I would say the difference is more than I expected. By the looks of your results, you omitted 25% of all Frank games, and the win rate went up from 54.5 to 56.25. This is interesting because it implies single civ players only have a win rate of 49.5 ((54.5-56.25x0.75)/0.25). Anyway, the suggestion here is that not enough matches are excluded since the statistics react precisely how we would expect them to.

To be clear I’m not saying it wasn’t a significant jump, just that in my head I thought the order of magnitude might be more like 5-6% increase (but maybe that was just naive of me).

I don’t think the dissection is as clear cut as that as the matches removed will also include random pick frank matches vs single pick other civs. I was wondering about lowering the cutoff freehold to 40% actually, open to suggestions on how to better handle it.

I was also surprised the impact was smaller than expected.

But then I noticed the play rates…

You are excluding 50% civ pickers . Yet the Frank’s play rate is still phenomenally higher than port and Malay. Meaning the results are still influenced by pick rate .

I’m wondering if you could exclud someone that only picks meta civs.? How much it would impact the top performers? aka if I was a WAAC player on the ladder I would totally just cycle between Huns, Berbers and franks because I can play them similarly but are all powerful so I get a bit of a break between civs without losing power.

I think the interaction between the pick rate and win rate is just a hard problem to “properly” solve.

My current understanding (please someone correct me if I’m wrong) is that a players observed civ-win-rate is related to the average of the true-win-rates of the civs they play.

As an example, lets say we have 5 civs A-E where civ A & B have above average WR C has a neutral WR and D & E have below average win-rate. If a player just picks any 1 civ they will then have a 50% wr for that civ. If a player picks 2 civs equally then the observed wr will depend on relative wr between them. i.e. if they picked civs A and E the average is neutral so the observed win-rates will probs reflect the true-win-rate where as if they just picked civs A & B they would still tend towards 50% for each as the average of the 2 is above average.

Sorry I think I’ve explained that poorly but I hope it makes some sense. The tl;dr is I don’t think there is any clean way of selecting “non-civ pickers” because the affect ultimately depends on the underlying (unknown) true win-rates.

There is another effect, that of different maps. Suppose players who pick Franks a lot tend to also be worse at playing maps other than Arabia. This would counter the force pulling their Arabia win rate towards 50%, as their ELO would fall when they play another map, allowing their win rate to rise above 50% on Arabia. I’m not saying this is the case, just an example of an effect that if it did exist, would make the situation even more complex to analyse.

This is true, though I tried to account for this by splitting out between “open” and “closed” maps. Although it doesn’t perfectly solve the problem it should account for the worst of it as the “open” maps should all have a similar play style in theory.