Welcome to the Forum Archive!

Years of conversation fill a ton of digital pages, and we've kept all of it accessible to browse or copy over. Whether you're looking for reveal articles for older champions, or the first time that Rammus rolled into an "OK" thread, or anything in between, you can find it here. When you're finished, check out the boards to join in the latest League of Legends discussions.

@Riot: 10% Crit chance is actually 8%

White Mage Kris

Senior Member

Holy ****, this is in-depth analysis

Riot Morgageddon

Player Support Tech Specialist

Quote:
Yurixy:
I'm not good with these math related problems.. but
Is it really that way it works? I mean % can't be trusted even after millions tests. There will be always a margin of error. A quick example would be:
If a coin is flipped 1 million times, the side chosen could be 490.000 instead of the 500k expected.

I'm not sure how these error margin works. If anyone knows, please explain better to me, so I can learn

An error margin is just a fail/success margin.

Think of flipping a light switch. Flick up, it's on, flick off, it's off.
An error margin on that switch working (in this case) is 1%.
In 1% of all cases, flipping that light switch will NOT cause the light to go on or off, requiring a reset back to original position before working again.
As time goes on (1000, 10000, 100000 flips), the rate of failure should be at 1%.
Numbers that large are statistically significant. So, if after 10000 flips, you find that your sample is actually failing at 4%, you can conclude one of two things.
Either the overall stated error margin is wrong (4% instead of 1%) or your switch has an error margin of 4%, because the overall pool of every light switch in the world is 1%, and thereby many other switches are below 1% to even it out.

The first is provable by testing many switches.
The second has been proved by your test.

Does this help?

March of Dimes

Senior Member

Interesting, will come back later +1.

Yurixy

Senior Member

Quote:
Morgageddon:
An error margin is just a fail/success margin.

Think of flipping a light switch. Flick up, it's on, flick off, it's off.
An error margin on that switch working (in this case) is 1%.
In 1% of all cases, flipping that light switch will NOT cause the light to go on or off, requiring a reset back to original position before working again.
As time goes on (1000, 10000, 100000 flips), the rate of failure should be at 1%.
Numbers that large are statistically significant. So, if after 10000 flips, you find that your sample is actually failing at 4%, you can conclude one of two things.
Either the overall stated error margin is wrong (4% instead of 1%) or your switch has an error margin of 4%, because the overall pool of every light switch in the world is 1%, and thereby many other switches are below 1% to even it out.

The first is provable by testing many switches.
The second has been proved by your test.

Does this help?

And this error margin is also applied on your tests? Or there is no way to determine that? (sorry i'm noob lol)

M0b1us

Senior Member

Glad to see your using some math, but you are forgetting a single fundamental concept. In probability, there is no set number of trials that should be completed until you can make a definitive conclusion. Think about a coin in a perfect world so there is a 50/50 chance to land on heads. Now let's say you flip a coin 1000 times. For each and every combination of coin, there is a (1/2)^1000 chance they will all land on heads, all land on tails, 500 will be heads and 500 tails and literally every combination.

You would need to perform the experiment enough times so that any more experiments would change the average by miniscule amounts. Let's say you performed it 1000 times and have an average of 11%, do you think doing it again will change that percentage much?

Good experiment, but you just need to run through it several more times. Probability 101.

Ozbirta

Recruiter

@Rickless

Come help us in this awesome math thread. I know you can't resist it....

Riot Morgageddon

Player Support Tech Specialist

Quote:
Yurixy:
And this error margin is also applied on your tests? Or there is no way to determine that? (sorry i'm noob lol)

Well, I don't know the error margin.
For instance, on a standard graphing calculator, there is a lovely function called rand() of some form or another.
It randomly gives you a number between 0 and 1. Theoretically it should average out to .5 with a long enough time frame. However, if we were to test a lot of calculators, and find they all were around .52, then the error in that random function is 2%.

I might be able to guess at it from the final tally of data, however all I can say at the moment is that 10% crit is ~2% lower after the duration, and that the PROBABILITY of my run's 10% crit chance ACTUALLY being 10% is around .0000000001% (It's around 33 standard deviations away, or absurd to be blunt).

Ozbirta

Recruiter

Quote:
M0b1us:
Glad to see your using some math, but you are forgetting a single fundamental concept. In probability, there is no set number of trials that should be completed until you can make a definitive conclusion. Think about a coin in a perfect world so there is a 50/50 chance to land on heads. Now let's say you flip a coin 1000 times. For each and every combination of coin, there is a (1/2)^1000 chance they will all land on heads, all land on tails, 500 will be heads and 500 tails and literally every combination.

You would need to perform the experiment enough times so that any more experiments would change the average by miniscule amounts. Let's say you performed it 1000 times and have an average of 11%, do you think doing it again will change that percentage much?

Good experiment, but you just need to run through it several more times. Probability 101.

He did explain that he is going to be running it several more times, he seems to be very aware of basic probability. lol

Riot Morgageddon

Player Support Tech Specialist

Quote:
M0b1us:
Glad to see your using some math, but you are forgetting a single fundamental concept. In probability, there is no set number of trials that should be completed until you can make a definitive conclusion. Think about a coin in a perfect world so there is a 50/50 chance to land on heads. Now let's say you flip a coin 1000 times. For each and every combination of coin, there is a (1/2)^1000 chance they will all land on heads, all land on tails, 500 will be heads and 500 tails and literally every combination.

You would need to perform the experiment enough times so that any more experiments would change the average by miniscule amounts. Let's say you performed it 1000 times and have an average of 11%, do you think doing it again will change that percentage much?

Good experiment, but you just need to run through it several more times. Probability 101.

I did say I was intending to run this several more times, since I was only time limited on the Dominon map. I'll be updating with each test.