SC2 ladder + math = Zerg sucks

Based on the MMR distribution of the SC2 ladder (https://i.imgur.com/ojB2Nw8.png), we can calculate the odds of a grandmaster occurring for each race, which is P(x>4860). The mean & standard deviation for each goes as follows:

Mean:

P T Z R
2791.858355 2696.590539 2886.363453 2769.670307

Standard Deviation:

P T Z R
769.1231371 782.1677696 741.3906068 691.8325611

P(x > 4860) for each race:

P T Z
P(z > (4860 - 2791.858355)/769.1231371) = P(z > 2.69) = 0.003573 P(z > (4860 - 2696.590539)/782.1677696) = P(z > 2.77) = 0.002803 P(z > (4860 - 2886.363453)/741.3906068) = P(z > 2.66) = 0.003907

The ratios of Grandmaster should then be:
P(x > 4860) / (P(ProtossX > 4860) + P(TerranX > 4860) + P(ZergX > 4860))

which is:

P T Z
(0.003573 / 0.010283) * 200 = 0.347*200 = 69 (0.002803 / 0.010283) * 200 = 0.27*200 = 54 (0.003907 / 0.010283) * 200 = 0.38*200 = 76

What are the actual Grandmaster statistics? I am glad you asked:

P T Z
56 86 45

What’s the probability of this occurring? That’s an easy one:

P T Z
2.62% 0.01% 0.01%

TLDR

Zerg’s representation in Grandmaster is about 40% smaller than it should be, Protoss is about 20% smaller, Terran is about 60% larger than it should be; the confidence of these results exceeds 97%, 99.99%, and 99.99% in each case respectively. Terran is drastically overpowered, Protoss is slightly underpowered, and Zerg is drastically underpowered.

2 Likes

Interesting topic, let me talk further on this.

I had a look, and Zerg is also the lowest represented at the low level, bronze, silver and gold. So it appears that zerg is happily average.

Another, seperate point, is the lower amount of zergs total. This points to zerg not being fun to play. I will blame it on tiring gameplay, such as creep spreading, possibly queen inject? And having to anxiously wait on larva to pop.

1 Like

You know what’s great? I can actually ā€˜prove’ the opposite with this image: imgur(Link Remove Me).com/gallery/3JMjyJ4

Now, according to this, there are FAR more terrans in low leagues than zerg.

Grandmaster players account for less than 1% of the population. So, to 90% of players, zerg are OVERPOWERED lol.

Here’s a cool statistic:
Diamond/plat are even distribution (60,000~ players)
Now, gold/bronze/silver: (70,000~ players)
Zerg has 24% dist.
Protoss has 33% dist.
Terran has 43% dist.

So, based on distribution alone, you are around 50% more likely to be in a low league as Terran, 25% as Protoss.

Or, putting it another way, as Zerg, you are 16% more likely to be dia/plat than gold/silver/bronze.
A Terran is 26% more likely to be gold/silver/bronze than dia/plat.
Protoss is roughly in the middle of them.
Masters has 5% less zerg, Grandmasters has 25% less zerg, but grandmasters are less than 1% of the playerbase. Who do we balance for?

Zerg’s mean MMR is about 100-200 higher than Terran/Protoss. I was surprised that the difference was that low.

Being overpowered would widen the variance and lower the average which is literally the exact opposite of what is observed for zerg. The problem on these forums is that people don’t know how to read a basic probability density chart.

If you have two groups of people and one group has an advantage over the other group, the variance of each group would be identical and this is crystal clear if you understand, for example, the binomial theorem. The difference in variance of the two groups can’t change as a function of the average performance level of the groups, e.g. if it’s tight for one then it will be tight for the other; if it’s loose for one then it’s loose for the other. A difference in the variance can only occur as a result of group selection, e.g. if there is a correlation between group performance and and the skill level of the player who picks which group he wants to be in.

This creates the effect that, when a group has a disadvantage, their average performance goes up while the variance of their performance shrinks. A group that has an advantage has a lower average and a larger variance. This makes sense, intuitively, since an increase in performance lowers the skill floor, making it more accessible to more people (in particular the lower performing ones). This is saying that as the performance level of the race goes up, the performance level of the group goes down, and this happens because the demographics of the group changes as a function of the race’s performance.

Mathematically, this means that group performance varies by racial strength and group demographics, but group demographics varies by racial strength. In other words, racial performance is a derivative of both group demographics and group performance, but group demographics is also a derivative of group performance. That makes racial performance both a first and second derivative of group performance.

Plotting this over time, the first derivative (racial performance) will create an increase or decrease in group performance for a short period, but it also incites a change to another derivative (group demographics) which will quickly outweigh its own impact direct impact on group performance.

This means that average group performance goes up for a short period after becoming overpowered, but goes down long-term due to the second derivative aka demographic change. This predicts that a race that is underpowered will have a higher average and lower variance. A race that is underpowered will have a lower average and higher variance. That’s exactly what is observed in all three cases with all three races – each one perfectly fitting this trend.

What I am describing is known as the ā€œLotka-Volterraā€ equations. They are also known as the ā€œpredator/prey equationsā€, which gives us a way to visualize this more effectively. Think of balance like rabits vs wolves. Wolves have to kill rabits to survive. So if wolves are doing well, they are by definition killing lots of rabbits. Which rabbits are the wolves killing, though? They are killing the slowest rabbits, which increases the average speed of rabbits. It’s the same exact process in effect. When a zerg gets absolutely wrecked by a terran, he quits playing or switches races. The ones most susceptible to being wrecked are the lowest skill ones, which means the average zerg performance goes up as a result of terran being overpowered.

Literally every facet of ladder data perfectly fits this description. Zerg’s play fewer games, are less numerous on the ladder, have higher average performance, and a tighter variance – all predictions of the Lotka-Volterra equations.

Hmm, not sure about that, the simplest explanation to me is that Zerg must be easier to win with (below Masters). EDIT - a few days later: I wanted say Zerg should be easier at EVERY rank, but I was being nice.

Continued: I mean, it’s the most logical explanation as to why so many Terrans are stuck at bronze/silver/gold. They can’t all be bad, r-right?

If wolves are eating rabbits, that would increase the average performance of rabbits, because the wolves eat the slowest rabbits, while decreasing the number of rabbits and also decreasing the variance in performance of the rabbits (the rabbits left alive will have a reliably higher average performance).

If zerg is easier to win with, Terran and Protoss would have higher performance as a group, there would be fewer Terran and Protoss than there are Zergs, and the Terran/Protoss groups would have a lower variation in their performance. None of these predictions align with the measurements. Your theory is wrong.

SC2 is a predator/prey tournament system which is very similar to many biological systems. The behaviour here is well defined. We know exactly how it works, and if it is doing something we know what it means. If you can’t understand simple ā€œwolf eats rabbitā€ examples, then you should probably check out tic-tac-toe instead of SC2.

This is very strange, we are talking about casual ranks, there is no incentive for terran to try harder because they got eaten. It’s just non-sensical. The majority of zergs are at diamond rank, they all won for some reason.

What on Earth are you even talking about. This has nothing to do with ā€œtryingā€ harder. Players are more likely to quit the game or race switch when they lose. People who lose more are more likely to quit or race switch than someone who loses less. If a race is underpowered it will lose more which means more quitting and more race switching. It’s the same as a rabbit dying to a hungry wolf. This is how tournament systems operate. The lowest performers are kicked out. I can’t believe I have to explain how a tournament works.

You don’t dude, the fact is that terran are losing way more games at the low levels. Race switch or not, they lost to zerg somehow.

That’s a straight up lie.

How can it be? There’s twice the amount of terrans than zerg at low level, but only the same amount at diamond. Theoretically, it should be spread out more.

So you have a hard time separating racial performance from group demographics. That takes legit brain damage to get those two mixed up. Racial performance is the design of the game. Group demographics would be who plays what race.

Ah, so you are using the argument that terran is the ā€˜noob’ race. Perhaps, but of course there’s no proof really, and the game is 11+ years old now? Maybe.

That’s absolutely wrong. Did you read anything I posted? The widening variance means it’s played by more noobs AND more veterans, which is exactly the case with terran (highest in Bronze and highest in Grandmaster). This is the widening of the variance that I keep talking about.

That’s a straight up lie. Blocked. The ladder data perfectly conforms to this model (and no other) with about a dozen verifiable predictions (each of which are accurate) as I detailed in previous posts. To say that there is ā€œno proofā€ is a blatant lie. Literally all the available data conforms to this theory. It’s unreal that I can spend 1,000 words detailing all the ways in which this theory robustly describes the observations, and someone will waltz in here and say ā€œof course there’s no proof reallyā€. Do you have eyes? Absolutely unreal.

1 Like

There are drastically more Terrans in bronze, silver and gold league (half the playerbase) and only more Terrans in Grandmaster league, slightly masters.
Zerg must be doing really well, on average.

For anyone reading, the OP is trying to assert that the majority of diamond/plat zergs are just ā€˜better at the game’ than the gold/silver rank terrans. No imbalance here.

The worst part about Terran hogging GM is that it’s all low GM terrans so they account for 60% of your games. I also think you’re using the NA server because it fits the narrative.

This is a 5300 mmr terran. Link to any other person to ever beat a GM with 1 base swarmhost and I will delete my account and never post again. Balance is not relevant to guys like me. I’m in a different universe in terms of skill. Everyone else is stuck in the stone age while I am colonizing mars.

Accusations of bias just don’t work on me. Do you think I need help from SC2’s balance team to win? Lmao. Come to think of it, I am biased – I want Blizzard to nerf zerg even more.

It’s sad when a player thinks that a 1-base push is an indication of skill, but anyway.