Top Tier Korean ZvT and TvZ TLPD statistics - Page 17

On March 16 2012 23:11 LoliSquad wrote:


Halving attack damage and doubling attack speed of units actually has (at least) 3 rather drastic effects. First, weapon upgrades and enemy armor upgrades become far more relevant for the unit (units with +2 or more damage per upgrade would be fixed by scaling them down, but armor would still be twice as strong against them (not just upgrades, base armor as well)). Second, units with projectiles (such as roaches) which as of today overkill a ton (would be cool to have "wasted damage from overkill"-statistics in the game) would overkill a lot less, making them a lot more effective. Lastly, it would lower burst damage which would let more units counterattack after the first volley and change unit dynamics a lot (hard to predict exactly how).

I think the lowered overkill would make the biggest impact. And it would do so in all leagues, not just Korea GM.

@Overkill, i do not only mean the wasted 3 damage 3 roach shots (16dmg) have against a 45hp marine (0 armor). In addition to this, a lot of roaches can attack the same marine as it dies only when the 3rd shot hits it, leaving other roaches open to attack it while the 3 projectile is in the air, potentially making marines "tank" over 100 hp worth of attacks.

A possible "micro trick" zergs can employ is to make roaches spread fire instead of letting them automatically acquire targets and potentially wasting hundreds of damage.

On March 16 2012 22:54 Fubi wrote:


1) It's a common misconception that higher rank players should have ungodly win ratio: you're forgetting the fact that the better you are, the further you advance in tournaments, and therefore the better your opponents are as well. Most of the % are inflated above 50% simply due to all the lower level players they've played, but as you can see, it gets harder and harder to climb higher: it isn't a linear increase relative to your skills

2) 64% win rate is VERY VERY good, it means you win 2/3 of your matches, so you win most of you Bo3's, so I don't know what you mean "only" 64%

3) Where did you read from the OP's post that "most" of the zergs have less than 50% win rate vs T? He listed like 5 out of 50...

4) There is more than your two possible reason to explain this; one of them being "by pure chance". If you flip 20 fair coins, in theory you should get 10-10 head/tails, but in reality, it's not rare to get 9-11 or 8-12 or even some more variations. This is simply due to chance: something the OP didn't take into consideration at all, and the whole point of statistics is to prove your results to be actually true due to real variables and not simply by chance.

On March 16 2012 23:59 Otolia wrote:
@OP : I am starting to question the quality of your education when you decide that 20 is a relevant number. Try to write a paper on the behavior of 20 electrons in a cold plasma (random example) and the world will laugh at you. And that's because 20 is a ridiculously small number for a system where a mole contains 10E23 more atoms. I could argue that 100 games is a better number and none of us would be wrong but neither of use would be right either. Unless you can somehow prove that beyond a certain number micro-phenomenon (in this case : mis-micro, illness etc) do not have a relevant influence on the outcome, you will have a hard time convincing anyone that what you have is proof. I agree that 20 is good arbitrary limit in case of binary system but games are hardly binary. 1 loser / 1 winner doesn't make the game of equal probability.

That being said, the tendency is certainly clear and intriguing. But I suspect the environment to be largely responsible for this as it's evident that Terran is the most represented race in Korea. Maybe you could compare your results with the playhem database to have a larger sample.

Did you consider that the time span of your sample might just be too big (too many patches) for your results to be relevant ?

On March 16 2012 23:32 TheBB wrote:
In BW, 64% was about as good a winrate as anyone could hope to have in a single matchup, barring a select few outliers. Now all of a sudden it's "not really that good."

On March 16 2012 23:32 TheBB wrote:
In BW, 64% was about as good a winrate as anyone could hope to have in a single matchup, barring a select few outliers. Now all of a sudden it's "not really that good."

On March 17 2012 02:26 1st_Panzer_Div. wrote:


And in BW there were more top level players and the difference in skill between top players was smaller. It's also a different game. I could say in Football 64% is not that great, so what's the big deal?

On March 17 2012 02:37 m0ck wrote:
Is there really any doubt that, at the very highest level, TvZ is T favored?

On March 16 2012 19:43 neoghaleon55 wrote:
Ugh why do I feel like everytime I bring up stats in teamliquid, I need to teach a whole course of statistics to satisfy the whiners. This is the reason why I didn't want to spend time explaining earlier...but here goes.

Ok here's the breakdown:

The argument: The sample size is not large enough...

We have to understand why this is a problem in the first place.
This is related to the coin flip test...which is a comparison between True theoretical probability and actual probability.
Everyone knows that in a truly balanced coin (yes I know tails land more because head is heavier, but let's assume that the coin is fully balanced) the chances of heads or tails is 50/50.

However, if you flip the coin 8 times, you might get 5 heads and 3 tails, or even 7 heads and 1 tails. The reason this happens is because the actual outcome does not approach the theoretical outcome until very high number of samples are gained. This is related to the question at hand:

Are the number of games played by these top koreans high enough for their theoretical skill level to show?

I answer yes, 20 coinflips or greater tend to be the magical number in which the standard deviation improves significant enough for the gaussian distribution to be acceptable. Thus 20 games or greater is enough to probe how well a pro-gamer is skilled at a single matchup, as the chances of random deviation should decrease significantly when we attain 20 games or more. All these statistics presented in the opening post has more than 20 games. We are pretty safe to say that they matchup well with the player's capabilities.


So here are some pictures

Number of Tails
8 coinflips


16 coinflips


32 coinflips


As you can see, the gaussian distribution gets "slimmer" the more coinflips there are.
This sliming down of the curve can be numerically expressed by the standard deviation. The bigger the sample size, the slimmer the standard deviation, which means the closer the actual probability approaches the theoretical probability.

The calculation for standard deviation is dependent on the inverse of sample size N...
The greater the N samples, the less fluctuations one is likely to see (meaning the standard deviation is smaller...which is what we want)

So why 20?
Because anything greater than 20 is great, but the total impact of N, itself, to the statistics decreases significantly over 20.
Call it diminishing returns.

Those of you who are telling me to go calculate the confidence interval have no idea what you're talking about.

On March 16 2012 23:32 Antisocialmunky wrote:
BW players play against each other in multiple leagues and rack up tons of games. Speaking from a statistical point of view its probably too early to tell what the relative win rates will be.

Also:

MMA vZ: 21-5 80.77%

Yeah, Boxer's Heir indeed.