Greater Output/Crit rate?

[Norian]

Also, note that the item that you are crafting is a lvl 70 recipe and you are at skill level 84.

Some have said that your proc rate is max at 40 (or is it 41) levels above the recipe requirement. Your skill level relative to the recipe level has to be over some point.

Bad luck can happen though...look at some of the examples above. But, if you do the math, if you assume that your proc rate is 25%, if you determined the probability that you would not crit at all, in your case --> 75% = 3/4 --> (3/4)^14 = 0.0178179480135440826416015625 = 1.78% chance that you will not proc at all or 98.22% that you will proc at least once.

I know this is handicrafting, but getting a good feel/sampling for actual proc rate can be done with alchemy --> manastones. You can probably craft thousands of samples with the lower manastone conversion recipies and get a good ACTUAL rate with ~±1% error and minimal cost.

You could probably also do a no DP and 4000 DP trial as well.

Sounds like a job or a good test for me...

[Chillbro]

DP doesn't matter.
Skill level doesn't matter.

RNG matters.

Nothing else matters unless:

1. NCSoft comes out and gives you specifics in detail.
2. You perform an extensive, recorded controlled test with thousands upon thousands of trials. Which still would have to be taken at face value without NCSofts endorsement.

Stop posting your anecdotal "proof". You're giving people false hopes and contributing to the nonsense being posted on the wiki.

[Lerxt]

Originally Posted by Chillbro

DP doesn't matter.
Skill level doesn't matter.

RNG matters.

Nothing else matters unless:

1. NCSoft comes out and gives you specifics in detail.
2. You perform an extensive, recorded controlled test with thousands upon thousands of trials. Which still would have to be taken at face value without NCSofts endorsement.

Stop posting your anecdotal "proof". You're giving people false hopes and contributing to the nonsense being posted on the wiki.

for the most part, I completely agree with you.

One thing.. you don't need "thousands upon thousands" of trials to get a statistically significant conclusion. I'm no statistician, but I think the number of trials you need depends on the expected rate of occurrence. With an anticipated proc rate of 20%-25%, I'd believe you could get a statistically significant sample size with just a few hundred samples.

Some math/statistics major can come prove me wrong, of course. Not my field.

[Chillbro]

Originally Posted by Lerxt

for the most part, I completely agree with you.

One thing.. you don't need "thousands upon thousands" of trials to get a statistically significant conclusion. I'm no statistician, but I think the number of trials you need depends on the expected rate of occurrence. With an anticipated proc rate of 20%-25%, I'd believe you could get a statistically significant sample size with just a few hundred samples.

Some math/statistics major can come prove me wrong, of course. Not my field.

In an effort to lessen your +/- accuracy range you would definately need a metric **** ton of data. Granted, you can only lessen the +/- accuracy range to a finite number...

Without "official" data from NCSoft, I'd be hard pressed to accept any user submitted data that didn't involve thousands and thousands of trials. But perhaps that's just me.

[Kabasue]

Originally Posted by fungi

I just crafted 232 (50+50+50+50+32) gold rings, and 0 critical. Yes 0/232.
I am not sure whether I am bugged, I have send petition to NCsoft and waiting for reply.

BTW, this happened after new patch today.

-------------------------

More from friends today after patch:
0/113 in a row (food)
0/76 in a row (chain armor)
0/75 in a row (alchemy)

So now I am pretty sure there is something wrong with craft system after the new patch today.

Similar problem, 9 pairs of earrings, 0 procs. Decided to log off for a while and do something less painful like watch a movie.

[Primos]

Originally Posted by Chillbro

In an effort to lessen your +/- accuracy range you would definately need a metric **** ton of data. Granted, you can only lessen the +/- accuracy range to a finite number...

Without "official" data from NCSoft, I'd be hard pressed to accept any user submitted data that didn't involve thousands and thousands of trials. But perhaps that's just me.

It really depends on the amount of factors that would affect crit rate and by how much they affect crit rate. The former affects your ability to do ceteris paribus experiments (all other parameters being equal), the latter affects how easy it would be to detect deviant crit rates. If there for instance is a single variable that has a significant effect on crit rate, it doesn't require large amounts of data to figure out. You can use this binomial probability calculator to check the binomial probabilities. First you need a hypothesis. Let's say "being 40+ levels above the recipe lvl significantly ups your crit rate to 30%".
Given the general information the base rate would be 20%. Now let's say you do an experiment with a sample size of 200. Here are the cumulative probabilities for the success rate being greater than or equal to the given number of successes with the base rate:
40: 52.8%
45: 21.1%
50: 4.9%
55: 0.7%
60: 0.05%

Now let's take the cumulative probabilities of the number of successes being less than or equal to the given numbers with a crit rate of 30%:
40: 0.1%
45: 1.1%
50: 7%
55: 24.5%
60: 53.5%

If you'd repeat this experiment a few times, you'd easily detect such an effect on crit rate. If you repeating it 3 times and the crit rate is actually upped to 30%, the chance that all 3 experiments result in at least 45 successes is roughly (100%-1.1%)^3 = 97%. The chance of this happening when your crit rate is unaffected is roughly (21.1%)^3 = 1%. (These calculations aren't completely statistically valid. Ideally you'd use something like Fisher's method for combining the p-values, but that wouldn't really affect these rough numbers that much). This means that you only need a limited amount of data to notice such a significant increase in crit rate. This is because with a crit rate of 30% a few of these experiments would almost certainly result in outcomes which with a crit rate of 20% would be as likely as being struck by lightning even when you're very unlucky.

[Chillbro]

Originally Posted by Primos

It really depends on the amount of factors that would affect crit rate and by how much they affect crit rate...

But that's exactly it. There are no factors. At least none that people have accepted as valid at this time. As far as preliminary testing goes, all things are equal.

I mean, I guess you can say the chance of landing on Tails when you flip a coin is 50%. We know that because there are only 2 sides to the coin. But if you flip the coin 5 times and it lands on heads 4/5 times that doesn't mean the chance of landing on heads is 80%.

How many times should we flip the coin to "prove" the chance of landing on heads is 50% if we didn't know beforehand that there were only 2 possible outcomes?

Granted, there isn't an external factor pushing the coin one way or the other as NCSoft is doing for Crit or Non-Crit. But it does boil down to...Crit or Non-Crit. Two possible outcomes, the problem is NCSoft is using dice, and not a coin. Theres only 1 roll that constitutes a crit while every other roll constitues a non-crit.

We just don't know what kind of dice they are using and how many sides are in play.

For all we know, they could be using a single 100-sided die and if you land a Prime Number, you crit!

[Primos]

Originally Posted by Chillbro

But that's exactly it. There are no factors. At least none that people have accepted as valid at this time. As far as preliminary testing goes, all things are equal.

And how do we know this? Not because there are tons of statistical data rejecting every single hypothesis, but because even small amounts of data can clearly identify significant changes in crit rates.

I mean, I guess you can say the chance of landing on Tails when you flip a coin is 50%. We know that because there are only 2 sides to the coin. But if you flip the coin 5 times and it lands on heads 4/5 times that doesn't mean the chance of landing on heads is 80%.

How many times should we flip the coin to "prove" the chance of landing on heads is 50% if we didn't know beforehand that there were only 2 possible outcomes?

We know this for the mathematically ideal conception of a perfect coin, not for any actual coin. No actual coin will have a chance of heads of exactly 50%, although most actual coins only deviate from this chance so slightly that we don't notice a significant change. You can't really determine exact probabilities in empirical situations based on a priori deductions (i.e. a coin has two sides, therefore the chance of of landing on heads is 50%). In statistics you can't really "prove" the a priori chance of something happening, but you can calculate the likelihood of some outcome based on an assumed chance. If I assume that I'm dealing with a roughly perfect coin (roughly 50% heads, 50% tails) the chance of getting 570 heads or more is about 1 in 250,000, making the assumption pretty unlikely. Of course repeating an experiment 5 times won't give you a reliable indication of a priori probabilities. However, the law of large numbers that repeating the experiment more often fairly quickly results in the actual frequencies converging to the a priori probability.

Granted, there isn't an external factor pushing the coin one way or the other as NCSoft is doing for Crit or Non-Crit. But it does boil down to...Crit or Non-Crit. Two possible outcomes, the problem is NCSoft is using dice, and not a coin. Theres only 1 roll that constitutes a crit while every other roll constitutes a non-crit. We just don't know what kind of dice they are using and how many sides are in play.

For all we know, they could be using a single 100-sided die and if you land a Prime Number, you crit!

That is irrelevant in statistics. Let's say we have a company that produces computer chips. This is a process that sometimes results in broken products for a large number of different reasons. From a statistical point of view the only thing that matters is the chance of getting a product working or being broken. Frequencies in a few 1000 independent samples are quite a reliable approximation of this a priori chance, since (number of successes) / total production converges to µ. Now let's say we want to see if some random technical innovation has a significant effect on the chance of success. It's utterly irrelevant that there are dozens or even hundreds of ways in which the process can fail, properly collected data will allow us to accept or reject this hypothesis with near certainty.

The point is that statistical laws result in the fact that with properly collected data it's fairly easy to draw conclusions. On the other hand, most people have a notoriously poor grasp of statistics resulting in proclaiming for instance that there must be something causing those 4 crits in a row since with a 5% base crit rate, because the chance of that happening is 0.000625%. That would be really unlikely if those were the only 4 attempts at a crit ever made. However, loads of people are crafting and they're all attempting to crit frequently. This means we're going to see lucky streaks all over the place. Lucky streaks don't mean a thing, properly conducted statistical experiments mean a lot though.

[Dannick]

I think I know where all of your luck went. I am in the middle of crafting a set of Noble Rose Quartz Earrings for my alt, I figured 50 silver earrings synths would give me a reasonable chance of getting 2 double procs. 50*.2=10, 10*.2=2

First set of synths, 18/38 noble silver earrings, including 6 straight HQs. I'm kinda baffled atm, the odds of hitting 20% 6 times in a row is pretty out there. I suppose I fell on the right side of the bell curve this time. Cheers

[Chillbro]

Averaging all of my crafting attempts over the past 10 days (100+ necklaces, 100+ earrings, 50ish staves, 20ish bows) I'm getting about 12% proc rate from white to green... bad 10 days for me.

[GoldenArrow]

I've made 16 egrasi bows so far on my quest to expert one.

16 crafts
of which
5 green crafts
of which
zero blue ones.

But in my experience I've noticed that 8 is the magic number for a crit.
With 5 you usually get atleast 1 crit but 8 is pretty much guranteed :j

[Vurtox]

I don't think so. After experiencing 3 times 16 failures in a row (and running out of materials/money to keep trying) I've switched over to stop crafting after 4 failures, because whenever I didn't get an early proc, it only got worse.

[ehrie]

You get in such a hole when those bad streaks go on . I've spent myself dirt poot on the Expert Exam and I'm only on the first part. Can't even make the Staff. 0 for 11 on it. I've never ever made a blue on Handicrafting in well over 100 tries. Doesn't make sense, my crafting crit rate on Weaponsmithing is okay, usually around 30% or so, but I swear Handicrafting is around 5% or less for me.

[rawrXtina]

Seems to me like there's a Streak mechanic in play, though. You'll get a certain amount of low procs, then you'll get multiple procs

When you fail to crit 50 times, and you find out that plenty of other people are also failing to crit 50 times with a supposed 20% crit rate, that's more than just RNG

FYI that has something like a 0.5% chance of happening