Thark
Stalker
- Joined
- Jun 8, 2006
- Posts
- 1,671
- Location
- Berlin
- Society
- The Prodigy
- Avatar Name
- Shawn Thark Columbo
Hey,
BEWARE: some math down there but consider reading my other post containing a way larger data sample than the one beeing discussed here:
Lemme try show you what whiteknut and I are talking about:
I try to avoid math on this forum because it's hard for me to be as excact as one has to be when using nonbasic math because I don't know all the terms in english and because people get upset when i just assume for example that when i say avg, that i know it's for an infinite amount of times even though i don't have a sample that seize. But because you insinst, i'll give it a shot and just hope you know what I mean even if i use a wrong word:
Let's do a simpler version first: 90% Hit Rate (Normal + Crits) vs. 10% Evade, for 3200 trials:
The cumulative probability of a lower than 89% hit rate is: 3.2% (Your first Test 87.42+1.65=89.07%)
Addmittedly you are on the lower end of the hit rate scale, but nothing special there.
Since you are talking about halfs of percentage points here: the cumulative probability of a hit rate below 89.5% is 16.53%!!! It's just not feasible to talk about tenths of percentage points when having only this small of a sample seize.
Now:
Doing the same thing (90/10) with 70000 Trials and 89% hit rate results in a crazy low cumulative probability (1.635 x 10^-18)
For 70000 trials even a 89.8% hit rate yields only a cumulative probability of 3.865% of beeing lower.
That alone should show that 3200 is a very small sample seize!
Without going into why comparing 3 highly unreliable samples yields even lower probabilitys of beeing "correct", because the math behind that would be very unreadable without me beeing able to code the formulas in some way on this forum, let's look at the more complicated case with 3 variables: Hit, Crit, Evade:
So assuming you fired 3200 Shots total and have an avg hit rate of 87%, 10% misses/evades and 3% crits:
In your second test:
You had 3205 shots with 84 crits, assuming the real crit rate is 3% then the cumulative probability of you getting fewer than 84 crits is 9.3%!!! That also means that in 90% of the cases you will do more crits when you repeat this!
70000 shots fired with 3% assumed crit rate and a measured crit amount of 1834 (the crit rate you stated) would yield a cumultated probability of 8.24*10^-10 for beeing lower!
I could go on and i could also include the way the numbers influence each other and why but I guess since you are quite capable of understanding math and the reasoning behind my post I believe I don't have to.
I didn't mean to attack you when i said that the sample seize is too small and the conclusions unreliable because of that but it's just a fact!
Regards
Thark
BEWARE: some math down there but consider reading my other post containing a way larger data sample than the one beeing discussed here:
Lemme try show you what whiteknut and I are talking about:
I try to avoid math on this forum because it's hard for me to be as excact as one has to be when using nonbasic math because I don't know all the terms in english and because people get upset when i just assume for example that when i say avg, that i know it's for an infinite amount of times even though i don't have a sample that seize. But because you insinst, i'll give it a shot and just hope you know what I mean even if i use a wrong word:
Let's do a simpler version first: 90% Hit Rate (Normal + Crits) vs. 10% Evade, for 3200 trials:
The cumulative probability of a lower than 89% hit rate is: 3.2% (Your first Test 87.42+1.65=89.07%)
Addmittedly you are on the lower end of the hit rate scale, but nothing special there.
Since you are talking about halfs of percentage points here: the cumulative probability of a hit rate below 89.5% is 16.53%!!! It's just not feasible to talk about tenths of percentage points when having only this small of a sample seize.
Now:
Doing the same thing (90/10) with 70000 Trials and 89% hit rate results in a crazy low cumulative probability (1.635 x 10^-18)
For 70000 trials even a 89.8% hit rate yields only a cumulative probability of 3.865% of beeing lower.
That alone should show that 3200 is a very small sample seize!
Without going into why comparing 3 highly unreliable samples yields even lower probabilitys of beeing "correct", because the math behind that would be very unreadable without me beeing able to code the formulas in some way on this forum, let's look at the more complicated case with 3 variables: Hit, Crit, Evade:
So assuming you fired 3200 Shots total and have an avg hit rate of 87%, 10% misses/evades and 3% crits:
In your second test:
You had 3205 shots with 84 crits, assuming the real crit rate is 3% then the cumulative probability of you getting fewer than 84 crits is 9.3%!!! That also means that in 90% of the cases you will do more crits when you repeat this!
70000 shots fired with 3% assumed crit rate and a measured crit amount of 1834 (the crit rate you stated) would yield a cumultated probability of 8.24*10^-10 for beeing lower!
I could go on and i could also include the way the numbers influence each other and why but I guess since you are quite capable of understanding math and the reasoning behind my post I believe I don't have to.
I didn't mean to attack you when i said that the sample seize is too small and the conclusions unreliable because of that but it's just a fact!
Regards
Thark
Last edited: