Hey all,
Just to say I updated the site to the latest patch ( 2021-11-20 to 2021-12-13)
https://www.ageofstatistics.com
Also as mentioned in the above thread I’ve backed up prior copies to a zip on gdrive:
https://drive.google.com/drive/folders/14I7A0Xl_2upC-VwrzyDcD6HKfT1K1YsC?usp=sharing
I plan on implementing a more proper solution for comparing against historic snapshots sometime in the future when I get a spare moment.
Funny. I can’t really find any meaningful changes in the stats there.
Ofc the balance changes also were minor, but even the most changed civs in Burmese and Vikings are basically unchanged.
The only thing that changed is Poles, esepcially on closed map they aren’t as dominant as they used to. If this minor change to the Obuch is explanation enough for this or just a statisitical noise I can’t answer.
What is indeed a bit irritating is the strong deviation between the ladder and the kotd stats. Some civs seem to only work in the hand of pros like Sicilians and Vietnamese.
Any explanation for this?
Likely just the effect of small sample size in KotD. I don’t even mean total games of each civ, think about how important civ matchups are, and how few of those are represented even in a large tournament. Vietnamese for example are generally going to perform well against archer civs, but struggle against cav civs like Berbers. I’m sure the Franks play rate drags down the Vietnamese win rate in ladder games (just like it inflates Indians win rate), but this may not have occurred in KotD. Depending which civs they were up against, any civ can look good or bad.
Totally agree there. With the very high amount of Archer play in KOTD (Feudal was dominated by archers/skirms as counters) naturally all the archer counter civs get a boost.
Sicilians and Vietnamese both can be seen as “archer counter” civs. And Vietnamese are also above average archer civs themselves.
So you are right this is explanation enough why these civs perform way better at pro level than on the ladder.
Hey all,
Just to say I updated the site to include more data from the latest patch (updated to include data from 2021-11-20 to 2022-01-02), otherwise there are no further updates this time round. As always the site can be found at: https://www.ageofstatistics.com .
As a general FYI I’m currently working on re-writing the frontend in order to support both prior statistics (I.e. previous releases of the site) as well as eventually aoe4 statistics (dependent on the APIs being updated). Longer term I have planned on my agenda to implement separate Elo calculations for open & closed maps (to help boost the statistics accuracy) as well as doing a simulation of the effects of Elo drift due to asymmetric player retention.
If people have any other ideas / suggestions please let me know !
Edit: Also to add that I am currently considering dropping support for EW as there just simply isn’t enough games being played to draw useful inference for it. Am happy to change this decision if people wish for me to keep it …
Can we have a little bit more information on “Individual Civilizations” tab. For example in “Individual Civilization” tab, when we select a civ, there is just a chart of win rate of that civ against others. Can you please add win rate vs game length and win rate vs elo too? I know they can be found in a different tab but still it will be easier to find all the necessary information of a particular civ in one place.
@coolios9876 nice to see you tinkering around with the match stats.
But pls correct the elo ranges to chose from.
Currently high level player are evaluated together with <2k rm 1v1 players, which makes not sense.
First because the playstyle changes completly within the currently selectable range.
Second because there are much more games by lower rated players, so effects by high level players are less observable.
For me an absolute highlight is sliding window with stats for each civ, especially winrate over rating.
It’s nice to see a stat visualization with confidence interval for things we already guessed 
Btw this graph also runs out of rating range already.
And also the confidence interval should be a function of the rating, bc there are fewer games for higher rating.
For sure, I was looking to redo the individual civ bit as part of the rework and can definitely try and add some of those plots in as well, will add it to the TODO list!
Apologies, would you mind just clarifying what you are proposing, are you asking to create a separate category for the >2k players. For reference I did create a >1.7k category for the top 5% of players, it’s very hard to push the boundary beyond this though because of how few matches there are 
The current windowing function is based on percentiles of the Elo distribution which results in the same number of players being in each cutoff which is why the confidence intervals don’t increase in size (i.e. the Elo inclusion window is larger for higher ELOs) I used to do it as a fixed amount (like ± 50) but found it was just too unstable for the upper ELOs. Can and play with it again though if you think the current iteration isn’t clear enough though. For reference all data does get used though the last category plotted is the [0.8, 1.0] as it’s a ±0.1 window. I could try going smaller but I found last time it lead to instability as well
Personally i would make the groups of same distance in rating, like
[1000, 1199], [1200, 1399], …, [2200, inf]
Or at least have a resolution of 200 points for players rated >2000
There is not a reasoning to put the top 5% off all players in one group, and the play style changes the most within the top 5% bin compared to all other 5% bins.
You named the cohort ‘*pro’ but it’s not pro and it’s not a cohort.
But this is just correct data evaluation, the api/plots seem to work =)
Hi all,
Just to say that I’ve just released an updated version of the site. Changes include:
Please note that I’ve had to temporarily remove the individual civ v civ outputs to save storage & bandwidth. I am currently working on re-implementing these in a more efficient way. This update also paves the way for me to include aoe4 content though I am now waiting for the unofficial API to mature or for an official one to be published.
These changes required a complete overhaul of most of the code base so if you do spot any issues / bugs or just have any feedback please let me know!
As always:
Really great work! Thanks for your data and analysis
Because Mayans is #1 and the meme should read “Mayans is still the strongest civ”. And most likely will remain the strongest even after the porto DLC’s civs get released.
Laughs in Malians, Cumans, Burmese and Malay on release
Hi all, looking to get feedback on 2 proposed changes that I was thinking about making.
Instead of using fixed elo values cutoffs (currently 1200 & 1700) I was thinking I would derive it based upon quantiles of the active player base. In particular I was thinking 25% and 3%. Only issue is it it would no longer display the elo in the filter label, you would have to check the “criteria” page. It means it also may change between data cuts due to Elo inflation and people leaving the game .
I was thinking of removing the “no civ pick” filter and instead replace it with a dedicated plot within each other filter, i.e. Another sliding winrate plot where the X axis is the cutoff value for the “no civ pick” freshhold. My thinking is that some people find the excessive number of filters confusing so would be good to consolidate. (Also helps my bandwidth, storage and processing times).
Any input would be appreciated.
I have the same opinion.
I would prefer a quantile based approach than arbitrarily picked elo numbers.
It is also more robust against elo shift due to inflation or if you one say decide not maintain and update for some reason but everyone is 2k and the elo filters not usable.
Just to be clear all the outputs would still display exact Elo numbers it’s just the inclusion criteria that would change. I might even be able to make the filter label dynamic as well though it would then show some ugly numbers I.e. “RM, solo, open, >1276” without making it dynamic it would be like “RM, solo, open >25%”
