I mean unless I misunderstood, isn’t ideally to nothing means no farm?
Yeah in feudal age. Given that you can purchase around 1500-2000 food before reaching the marginal cost (market) = marginal cost (farming) optimal stopping point criterion this isn’t exactly difficult to do for many builds. The 1500-2000 depends on the assumptions of the model but to get numbers below 1500 the assumptions get pretty unbelievable pretty quick.
And ideally to nothing is still constrained by whether it’s actually optimal to do so. Ideally the math works out that we need 0 farms because then we don’t have to test the build order. If we do need some farms then you need to run through the build order and test it and in general its much more of a hassle to make sure you don’t come up short and unable to afford castle age before reaching the optimal stopping point criterion.
I think there is a big assumption that unless you don’t stay in feudal age (aka not making units in dark+feudal age) which is pretty wrong most of the time on new arabia?
200 starting food + ~720 food from sheep + 600 food from boars + 750 food from berries + 1500-2000 food from the market = 3770-4270 food. Many builds can get to castle age on this budget. If you happen to be using a build that cannot get to castle age on this budget then just throw down more farms.
I mean I guess you could have interpreted what I said as “optimal use of Saracen market is (on average) to dramatically reduce feudal farm usage, because the optimum is nothing.”
If I were more careful I would say “The optimal use of Saracen market is (on average) to dramatically reduce feudal farm usage and instead buy food as needed until the price is 150-165 or so.” I would leave off the ideally because the ideal is not related to the optimum, its related to practically implementing it.
Great explanation, I feel like you’ve touched on all of the relevant points of consideration when it comes to variables that effect win rate. I feel that this is a good resource for players to view. at least once. as many of these realizations help people realize that good is relative and it really depends on a lot of things. Its because it depends that Age 2 is such a good strategy game.
Just to play devils advocate, win rate can still be an occasionally relevant data point. Given enough time/games played and taking into consideration the factors that you’ve listed I think that a civ performing under 50-48% in all of these area’s could benefit from a discussion about buffs.
I think there are reasonable arguments that can be made about a civilization’s strength that use winrate as a component for said argument. If that were the majority of threads I wouldn’t deem this thread necessary.
Unfortunately, that’s not the reality of these forums.
1 Like
You’re point was absolutely right about Arabia introducing a win-rate bias but this is a bad example. Franks have very high win rate across a wide variety of maps except pure water maps. Such water maps are very unpopular (have a very low play rate at all elo levels) and only a few civs are usually picked on those maps like Portugese, Vikings, Italians, Koreans. Most other civs would just die easily to these civs on water maps, unlike land maps where a lot of civs are usable. Franks are indeed a powerhouse. They’re the most picked pocket civ in ranked tg, one of the most drafted in tournaments, most drafts have them picked early and they get banned or sniped often.
While balance shouldn’t be Arabia based, it should be based on a pool of top 5 or 7 mostly played maps.
A solid example.
I just see balance changes as shuffling the deck of top tier civs. Cumans when DE released, then they moved to bottom but Khmer came up then they went slightly below and Tatars came up and so on. Ya its great that the balance team has converged upon an awesome 46-53% winrate range at higher elo. But still the most powerful civs in the game are the ones from 1999 - Chinese, Mayans, Aztecs, Vikings, Franks, Britons (a subset of these on most maps). I really wished the community doesn’t call for nerfing certain newly buffed civs and units right away when they become the new top tier.
Again right on point about draft viability but Vietnamese are a bad example for tournaments.
I dont think less than 10% play rate qualifies as a “ton”. They’re one of the least drafted and played civs in HC3, KOTD3, HC4, TOC. And even in redbull series they weren’t seen much. Mostly good pickrate only in TG tournaments because they’re a top 5 archer civ but hugely unpopular in mixed map or open map 1v1 tournaments.
This is a great point. Statistics from small data are very unreliable for analytical purpose.
In general, I agree that balance shouldn’t be based on blind civ win rates but based on all factors like you’ve mentioned. But I do want certain civs like Chinese, Mayans, Aztecs, Vikings, Britons, Franks to get a nerf.
2 Likes
I hear you about the Franks. My point there has less to do about an overabundance of Arabia, but rather underrepresented map styles in the data pool. There are a good number of maps that clearly give Franks some pains. Maybe the specific example for specifically Arabia is a bit misleading, but what I wanted to portray is that the Franks literally cannot float on water maps. You could design a tournament, within the already established game modes and map types in the game, in which the Franks would not be picked. That’s not my idea of a powerhouse.
On the subject of Vietnamese, they don’t see a whole lot of play, but they do get picked pretty often. Another contributing factor for me is that I don’t hear a lot of lamentation from the players about the Vietnamese. It’s really just a lot of things I’m considering.
(Edit just to say that I’ve only read the OP and not any of the follow up posts so apologies if the points I’ve made have already been made before / discussed)
I think I partially agree with you but I think your post would be better worded as “why not to take naive win rates such as those on aoestats at face value”. I would believe there is still a lot of value & utility in win rate statistics if used properly and within context. Some points I would argue:
- Confidence intervals allow us to measure the uncertainty due to low sample size
- LSmeans / emmeans allow us to weight towards specific scenarios of interest (i.e. equal weighting to each map * wr combination)
- Sensible cutoffs (like both players being > x) avoid the aforementioned issues of win rate imbalance due to the cutoff
- Restricting to higher elos reduces noise from players not using civs correctly (at the cost of noise from lower sample size)
- Modelling allows us to account for other factors such as imbalances in player Elo / skill
Don’t get me wrong even after all this win rates alone are not a golden bullet for balance decisions but I think to write them off completely as a “terrible idea” is a bit too extreme in the opporsite direction.
One thing I do agree with that is that not all civs have to, or even should be, good on 1v1 Arabia (which is where 70% of the data comes from). It is important to keep the usecase / context for which the civ was built for in mind and then look at the statistics / data that correspond to that context.
1 Like
At this point I stood up and I clapped. Great post
1 Like
Unless its a full water tournament, Franks will definitely get picked. DM WC picked, RBW series got picked a LOT, KotD-3 3rd most drafted, Hidden cups used a fair amount of times. And pretty sure it will be prioritized a lot in the upcoming History hit, Holy cup and KotD4 as well. Full water tournaments are generally not going to be popular because of the lack of variety in water map play and only a few civs being usable.
No they don’t. Stats at all elo levels in ranked, on multiple maps, across multiple tournaments show that Vietnamese are one of the least picked civ, very poor win rate against most civs. Only tg tournaments have Vietnamese and that’s just obvious - Even for 2v2, a best of 5 draft will have at least 10 archer civs both sides combined and obviously Vietnamese will be present there. But for 1v1,
RBW5 just happened. Viper picked it in the semis and won, Liereyy picked it in the Finals and lost to Viper. Of the three final matches in RBW, it was picked and played in two of them.
Do you not count RBW?
LOL. there were 4*7 sets before playoff, 7 playoffs. A total of 35 sets, a civ gets picked 4 times and you consider that as played often? Did you count number of picks/bans of Franks, Vikings, Britons, Cumans, Berbers, Magyars, Tatars, Bulgarians? Those were seen in almost every set. Vietnamese were one of the least drafted. They were picked more than Japanese, Lithuanians, Mongols, Italians sure. But if you take a look at Hidden Cup those civs were seen a lot. Vietnamese had just 3 or 4 picks there as well. Consistently least preferred across all tournament formats and yet never addressed.
They’re not a “never” picked civ but sure they are one of the least in all tournaments.
Less than half the civs are drafted in any given set. I consider the fact that both players have 39 choices, and still it gets picked 2/3 sets is notable. You think I mean “frequent” when I say “often” and when I say “often” I mean exactly that. We saw them picked and played in two consecutive sets, and yes, I see them played often. Is often enough to warrant shrugging off lackluster overall play? Maybe, maybe not. I know for a fact that there are a ton of civs that almost never get picked and the Vietnamese are not one of them. That alone should tell you something is more to the Vietnamese than a bad archer civ unless you claim competitive incompetency.
Granted. I don’t have a lot of long-winded conversations with people in the top competitive echelon of play, so I don’t know what drives them to pick it. I just know it gets picked, and otherwise it’s practically dead silent in terms of people talking about the balance of the civ.
I would argue that tournament picks and pros opinions in general are given too much weight in terms of balance discussions. Don’t get me wrong they are going to understand the game better than most but that doesn’t change the fact that they are equally prone to bias, misconceptions and regression to the meta. There is no golden metric for assessing balance but the idea that tournament picks & results are meaningful whilst win rate statistics aren’t is absurd to me.
Edit: in addition to this, just because a civ is balanced at the pro level doesn’t mean it’s balanced across the entire player base.
2 Likes
Quite the opposite instead.
If a civ is weak on Arabia 1v1, it’s never going to get used at all. At high level even less.
Take Italians for example, they are one of the best civs on water, yet water maps don’t even account for FIVE % of maps played.
Italians are useless really. If they were removed nobody would even care, same goes for Portuguese, another super broken civ on water.
Too bad water is good only for a very low number of players.
Meanwhile nobody would pick them on Arabia without going random/full random.
We have the top picked 10 civs that account for almost 50% of the play rate (Franks almost 9% alone), while the bottom 10 civs can’t even manage to represent 15%. Sure are old data, pre Poles+Bohemians, but depict a clear image imho.
I blame the mess that is the water gameplay for this, at least in part.
1 Like
It depends - do you want to balance civs across all maps or Arabia only?
If you want to balance on all maps, tournament performance is better than aoe stat, imo, as most of the maps are not played in ranked at all.
1 Like
Well at least tournaments pick rate and results don’t take into account all the times where one side started making mistakes until they lose.
Like seriously watch your own gameplay, the one of your friends or some low/mid elo streams, it’s quite eye opening. Why do people say cheap Franks castle are OP when they forget building their first one until they have 800 stone and will float stone in the late game? Why do they think Burmese are auto lose against archer civs when they forget to research leather archer armour with other civs anyway? The list goes on.
Or as an alternative you can take the raw win rate data and calculate the statistics for the maps / elos / civs of interest weighting according to how you wish to balance it. Don’t get me wrong, tournament / pro picks of Civs are very insightful, the point I was trying to make is they aren’t the be all and end all and still need to be evaluated within context and with data / scenario assessment as they too are also prone to bias and misconceptions.
I mean you are right, but only to an extent.
Let me start with an extreme example. Imagine you have a unit that dominates the game and completely drives the meta but that can be reliably countered with perfect micro. From the pros perspective you can argue this is balanced as they have the skills to micro and counter the unit however for the remaining 98% of the player base the game is now broken. The point is balance needs to be considered across all levels as the pros aren’t the only ones who play the game. A more relatable example is quick walling. If you have an OP unit that can be countered via quick walling should this be considered balanced? Sure the pros can do it and it won’t disrupt the tournament scene but a very high % of the player base now has this OP unit that they can’t do anything about…
I do concede to your point though that theres only so far you can go with this as after a while the issue is more people having a lack of fundamental understanding of the basic mechanics and having poor macro level decision making; however to me that doesn’t mean that we simply ignore >90% of the player base when making balance decisions.
This is why when I have generated statistics in the past I have focused on the >1200 ELO group as I argue this provides a good balance between player skill whilst still representing a decent chunk of the player base.
My final point would be that tournament results have such a low sample size that they really can’t be used as anything more than anecdotal evidence. For example let’s say civ X vs civ Y was played out 3 times and that civ X won all 3. Does this mean civ X is better than civ Y ? If so by how much? If we repeated the matchup 1000 times would civ X always win or would it average out to 50%? Again tournament play does give us useful information but I strongly believe its value is massively overstated in discussions on these forums.