Thread: 7 failures in a row at 62% success crafting == 11/hundreths of a percent chance

Not to sound cynical, I swear, but you do realize this is just a level 17 piece of armor we are talking about?

Just sayin'.

To sound totally cynical, I was kinda erked the other day at a 72% chance to make a piece of armor with four failures. So ticked I said screw the extra %, will run without the stupid weed in there.. and promptly made 4 purple .. so 72% with your item to increase in the mix vs not running it, that particular moment - without won.

*bang head on wall, repeat*

Originally Posted by Naesraen

i just did bring proof - and insofar as i don't understand programming, you've just shown an equal lack of knowledge of statistics.

the question is NOT 'is it possible to get an extremely unlikely streak' - it is, after all, possible to win the lottery, even though the odds are 30 million to 1 against or more.

the question IS, 'based upon a claimed proportion p1, can we reject or fail to reject a claimed proportion p'.

we can test that; i just did, the claimed proportion is rejected = the crit chance, whatever it might really be, for the OP, was LESS THAN the 62% the tool-tip stated. what causes that is beyond my expertise, absolutely - but determining the statistical significance of the result most certainly is within my expertise.

Do you REALLY believe the test applied to the data was appropriate? A small set of data, selected not at random but rather noticed to be anomalous by a user, is used to make the claim that "the conclusion to be drawn from the OP's results is that the crit chance was less than the indicated 62%.". I am not a statistician by any means, but did do research in biology for many years and have used statistical tools and know how to design experiments reasonably well. This does not pass the sniff test. Not for the math being performed, but for the applicability.

I repeat my question. If 1000 people flip coins 10 times, and one of them gets 10 heads and reports that to you, and is the only reporter, do you then apply your analysis and say "the conclusion to be drawn from that person's results is that the coin is heavily biased and doesn't have a 50-50 chance of heads and tails".

I wonder where the quote "lies, damned lies, and statistics" comes from?

Originally Posted by Baryn2

Not to sound cynical, I swear, but you do realize this is just a level 17 piece of armor we are talking about?

If the crit system isn't working perfectly, then it can easily affect other much more important crafting pieces.

Originally Posted by Baryn2

Not to sound cynical, I swear, but you do realize this is just a level 17 piece of armor we are talking about?

Just sayin'.

This is interesting because this lvl 17 math can be applied to lvl 65 items as well

Each roll is an independent event. So, the first roll is 62%. The last roll is 62%. Over time, you will win out. Even 100 rolls won't bear out a 6 of 10 result every time. 7 rolls can and will result in failure from time to time.

Just one of those unlucky moments, I had one recently, had a slightly higher chance of 72% and failed to crit four times in a row. I was rather annoyed, but that's just life. Take it on the chin and try not to dwell on those failed attempts, you may find next time you're crafting the dice will roll in your favour and you'll have crit after crit instead.

Random stuff sucks, but every time I've paid attention over a decent sample size (50+ or so), the crit rate has worked out to be about correct.

Originally Posted by DelgonTheWise

Originally Posted by Naesraen

i love these threads too, and seeing as i want a break from marking my students' statistics exams, i thought i'd make a hypothesis test.

our null hypothesis is that P(crit) = 0.62
our alternate hypothesis is that P(crit) < 0.62

we've got a binomial, and a sample of 7, so we use a test statistic z. no confidence margin is indicated, so we'll take the standard 95% confidence interval.

our test statistic z works out at -3.380; our critical z value we read from a table (left sided test, let's not make that mistake) and is -1.645.

now, if mag (z) > mag(z-crit), we reject the null hypothesis: 3.380 > 1.645, thus we reject the null hypothesis that P(crit) = 0.62.

or to put it another way: yes, the conclusion to be drawn from the OP's results is that the crit chance was less than the indicated 62%.

i posted a similar calculation using my F2P account some months ago, and it had the same conclusion: the crit chance is only indicative, and if you have a streaky run, stop and restart, because whatever seed value Turbine uses to approximate a random crit chance has failed you.

Drawing any conclusions from a sample size of 7 seems bizarre. Note that this is a self-reported anomalous small sample size run. People tend to notice, remember, and mention anomalies, and see patterns where there are none. If I give a truly random number series of ANY size, there will be patterns and oddities, that is the nature of the beast. Not questioning your math, but rather its application. For example, if 1000 people flipped a coin 10 times, and only one reported to you and told you that they got 10 heads in a row, would you have concluded by the above methodology that "the coin is broken"? How is that different than what you have done?

From the statement "whatever seed value Turbine uses to approximate a random crit chance has failed you" I do gather that you don't know much about computer pseudo random number generators. A "seed" is normally used to initialize a random number generator, not to approximate anything. Given an identical seed, a PRNG will produce the same exact series of numbers always. However, it would not be normal practice for Turbine to be re-seeding a RNG repeatedly. And for decent RNGs, the period at which the sequence repeats is really really big...

If you are saying that the RNG is broken, OK, bring the proof. Show me a series of RNG results and be able to predict the future based on it. Show that streaks/clumps are REALLY more than would be predicted by random chance. Or believe that, as in the so called "hot-hand" in sports, streaks are all significant rather than an inherent property of random chance...

Sorry bout the mess, but my brain just exploded all over this thread

It all evens out. Set up the experiment again and see if you do not get 4 of 7 crits with a 62% chance of success. My 6/7 with 58% chance (which I have noted more than once) evens out your bad run.

Just remember, NO computer RNG is precisely random. The randomness of the RNG takes hundreds of thousands (if not more) samples to approach the programmed parameters.

People only remember the bad runs especially (or good runs). The forget about the thousands of times things did not appear out of normal. To cherry-pick a bad experience and apply statistics to it is poor science.

Originally Posted by Dotlbeme

Just remember, NO computer RNG is precisely random. The randomness of the RNG takes hundreds of thousands (if not more) samples to approach the programmed parameters.

While your first statement is true of all pseudo RNGs, I don't understand your second statement. I believe it is not accurate, or would apply equally to a truly random sequence...

I missed an 82% crit chance 13 times in a row once. Calculate them odds.

A year later I hit two tier-9 relic crits in a row. Started with two tier-8 relics, ended up with 5 tier 9s.

It does even out. Look up "regression to the mean" for the statistical details.

Originally Posted by Dotlbeme

Just remember, NO computer RNG is precisely random. The randomness of the RNG takes hundreds of thousands (if not more) samples to approach the programmed parameters.

Thank you, you have made this thread complete in my eyes

Originally Posted by DelgonTheWise

While your first statement is true of all pseudo RNGs, I don't understand your second statement. I believe it is not accurate, or would apply equally to a truly random sequence...

It means that the wonkyness of the computer RNG will get close to what it is supposed to do given enough samples. In the short run you may see anomalies that are not within the bounds of the distribution's parameters (mean and variance). The variance in the short run, may allow for 6/7 fails more often than is supposed to happen with a uniform distribution (62% success, variance=.62*.38), hence an upset player.

I stand by the 'notice' factor. You really only notice what you think is a breakdown in the RNG, you never notice other times, so any analysis of the 'noticed' runs is not exactly statistical.

Oh, and 'The RNG hates me!" IS a legitimate observation. It probably does

Random number generators work AT RANDOM. Odds are that if you flip a coin 100 times, you'll get 50 heads and 50 tails. But the fact that you got a head last time *has no influence* on what you're going to get this time. You had bad luck, but the RNG isn't broken. It is Working As Intended.

Originally Posted by Dotlbeme

To cherry-pick a bad experience and apply statistics to it is poor science.

But...it does make for good forum drama!

Originally Posted by Dotlbeme

It means that the wonkyness of the computer RNG will get close to what it is supposed to do given enough samples. In the short run you may see anomalies that are not within the bounds of the distribution's parameters (mean and variance). The variance in the short run, may allow for 6/7 fails more often than is supposed to happen with a uniform distribution (62% success, variance=.62*.38), hence an upset player.

I stand by the 'notice' factor. You really only notice what you think is a breakdown in the RNG, you never notice other times.

My point is that over the short haul you'd see the same variance issues with a truly random series of numbers. A modern pseudo-RNG is for the most part indistinguishable from truly random by any analysis I (and probably you) could do.

I agree on the notice thing, as said up thread. Folks notice and post about anomalies, don't notice and don't post about things being about what is expected...

Meh, I've seen some things not add up right. While it makes sense to understand the numbers, the systems themselves can have errors. I find that some things do crit more. Polishing gems on my tinker seems to have between a 30 and 40% crit rate even with just 25% tools. I have gone 0 for 10 with an 83% chance to crit on fishing poles. I also once caught 10 point fish in a row (back when you had to catch a specific fish to level up). 10 casts, 10 points, proficiency maxed for the day And, yes, I posted about it (later purged), was considered probably lucky because I didn't stand around and cast another 10,000 times to bring real data.

IMHO, the RNG doesn't belong in crafting. It takes great pain and effort to just get a chance at something. I'm thinking the books and craft drops should increase the item quality and not the chance of the item itself.

I wonder if something like this would work better... it would certainly raise the desire for tools, scrolls, books, etc...

normal cloak - +10 ICPR
normal cloak with +5 mastery - +15 ICPR
normal cloak with +20 tools and mastery - +35 ICPR
normal cloak with +45 item and + 25 tools and mastery - +80 ICPR
normal cloak with +45 item and + 25 tools and mastery and +13 scroll - +93 ICPR.

Originally Posted by LadyDena

If the crit system isn't working perfectly, then it can easily affect other much more important crafting pieces.

Originally Posted by TharbadThief

This is interesting because this lvl 17 math can be applied to lvl 65 items as well

My post was just solely in reference to the fact, that someone who has played the game for four years had spent about ~500 silver coins to make some armor and was so upset that they didn't crit in 7 chances that they came to the forums to voice their frustration.

I could understand if you just spent 100 Lothlorien Branches and 7 Mithril Flakes with a 62% chance why you would be slightly upset and may be upset enough to warrant making a whole post about how much it upsets you.

That's all I was saying. The horse has been beat to death, and the RNG is just that, a RNG. It's not broken, there's not a secret to it, doing the /backflip emote before you craft doesn't secretly add to your percent. It's random. Knowing that, I only saw the need to point out the above.

Originally Posted by hucklebarry

Meh, I've seen some things not add up right. While it makes sense to understand the numbers, the systems themselves can have errors. I find that some things do crit more. Polishing gems on my tinker seems to have between a 30 and 40% crit rate even with just 25% tools. I have gone 0 for 10 with an 83% chance to crit on fishing poles. I also once caught 10 point fish in a row (back when you had to catch a specific fish to level up). 10 casts, 10 points, proficiency maxed for the day And, yes, I posted about it (later purged), was considered probably lucky because I didn't stand around and cast another 10,000 times to bring real data.

IMHO, the RNG doesn't belong in crafting. It takes great pain and effort to just get a chance at something. I'm thinking the books and craft drops should increase the item quality and not the chance of the item itself.

I wonder if something like this would work better... it would certainly raise the desire for tools, scrolls, books, etc...

normal cloak - +10 ICPR
normal cloak with +5 mastery - +15 ICPR
normal cloak with +20 tools and mastery - +35 ICPR
normal cloak with +45 item and + 25 tools and mastery - +80 ICPR
normal cloak with +45 item and + 25 tools and mastery and +13 scroll - +93 ICPR.

Personally, I'd rather have the RNG in situations where you could make a better item, than in another game where the RNG determines whether you get a point in advancing your crafting in the first place.

The chances of getting 7 failures in a row is 0.11% only if you do exactly 7 craft attempts. If you do 100 attempts, the chances of getting 7 in a row is far, far more than 0.11%. Even better, if you accumulate all the random things you do each day there are bound to be certain things that appear unlikely. "Oh wow, exactly 2 people got off of the bus at the last 5 stops, what are the chances! Amazing!"

A common error people make when trying to simulate a random string of 100 coin flips is not putting in enough streaks.

From one who is taking STAT 411 right now in summer term.. you have pretty much ensured that I will fail miserably...

Originally Posted by Waxe

Keep in mind a 62% chance doesn't mean 62 out of 100 items, it means each item has a 62% chance.

Which means that if you make 100 items, 38 of them will fail (give or take). Increase that number to 1000, you get closer to the 38% failure.

So if you try to crit something at 50% 15 times, you would think that at least one would crit. I think that the % is made up as I can craft less desirable items 20 times and receive crits more often than if I crafted some better item 20 times.

I'm pretty sure that having a vendor window(to buy needed crafting items) open while you are crafting affects your crit chances. I'm not saying it negates it completly, but it sure lowers it a lot; possibly to just 5%

I crafted 30 shields with a 17% chance to crit; None were a crit.

Another time I crafted 25 tools with over a 60% chance to crit and only 1 was a crit.

I did it twice, by mistake. I have since learned.

Originally Posted by BigFree

I'm pretty sure that having a vendor window(to buy needed crafting items) open while you are crafting affects your crit chances. I'm not saying it negates it completly, but it sure lowers it a lot; possibly to just 5%

I crafted 30 shields with a 17% chance to crit; None were a crit.

Another time I crafted 25 tools with over a 60% chance to crit and only 1 was a crit.

I did it twice, by mistake. I have since learned.

I always craft with the vendor panel open, and my inventory too. My metalsmith has some awesome luck because I critted armor 7 tries -> 6 crits at least 3-4 times with 58% crit chance.

Failing will always stick out more then normal result or even great luck.

Originally Posted by Dotlbeme

It all evens out. Set up the experiment again and see if you do not get 4 of 7 crits with a 62% chance of success. My 6/7 with 58% chance (which I have noted more than once) evens out your bad run.

IS that an example of the RNG equalling out though? Or is it the RNG getting stuck repeating the same result and not tripping over a new one? Another example of the RNG drops not working right, was when a regular non-special enemy dropped two Armor of the North-Drake's for me...

Originally Posted by BigFree

I'm pretty sure that having a vendor window(to buy needed crafting items) open while you are crafting affects your crit chances. I'm not saying it negates it completly, but it sure lowers it a lot; possibly to just 5%

I crafted 30 shields with a 17% chance to crit; None were a crit.

Another time I crafted 25 tools with over a 60% chance to crit and only 1 was a crit.

I did it twice, by mistake. I have since learned.

I'm with Uath. I ALWAYS (not sometimes, not usually, every single time) craft with a vendor window open. I have good streaks and bad streaks, but overall have results that are pretty much in line where they are supposed to be.

I once hit one-shot crits 13 times in a row, the last two with out crit ingredient or scroll since I had run out. However I didn't feel the need to start a thread pointing to Turbine's lack of ability to implement a random number generator in their code. Interesting, isn't it?

Nobody has mentioned the single most important factor, which is whether the OP was wearing the horseshoe when crafting. As we all know, it has an absolutely massive in game effect on probabilities.

Originally Posted by BellusDuFenna

After reading this thread so far, only one thing comes to mind: Statistics are fun, and most people don't understand how they work.

Much like magnets.

So... statistics are fun, and most magnets don't understand how they work?

I guess I agree.

Originally Posted by DelgonTheWise

Nobody has mentioned the single most important factor, which is whether the OP was wearing the horseshoe when crafting. As we all know, it has an absolutely massive in game effect on probabilities.

You are totally going the wrong direction

The largest single factor by far that affects the results are dependent on
A.) are you a Dwarf
B.) are you wearing your're "Beer Glasses"

Originally Posted by Aefa

The chances of getting 7 failures in a row is 0.11% only if you do exactly 7 craft attempts. If you do 100 attempts, the chances of getting 7 in a row is far, far more than 0.11%. Even better, if you accumulate all the random things you do each day there are bound to be certain things that appear unlikely. "Oh wow, exactly 2 people got off of the bus at the last 5 stops, what are the chances! Amazing!"

A common error people make when trying to simulate a random string of 100 coin flips is not putting in enough streaks.

Another common error people make is assuming the RNG in games is actually generating random numbers.

One streak that's in no jeopardy...any time someone is unhappy about the game and posts to vent their frustration, there will always be a large group of people waiting to pile on with "The problem is you/everything is WAI/stop whining and shut up" replies.

Originally Posted by Naesraen

i love these threads too, and seeing as i want a break from marking my students' statistics exams, i thought i'd make a hypothesis test.

our null hypothesis is that P(crit) = 0.62
our alternate hypothesis is that P(crit) < 0.62

we've got a binomial, and a sample of 7, so we use a test statistic z. no confidence margin is indicated, so we'll take the standard 95% confidence interval.

our test statistic z works out at -3.380; our critical z value we read from a table (left sided test, let's not make that mistake) and is -1.645.

now, if mag (z) > mag(z-crit), we reject the null hypothesis: 3.380 > 1.645, thus we reject the null hypothesis that P(crit) = 0.62.

or to put it another way: yes, the conclusion to be drawn from the OP's results is that the crit chance was less than the indicated 62%.

i posted a similar calculation using my F2P account some months ago, and it had the same conclusion: the crit chance is only indicative, and if you have a streaky run, stop and restart, because whatever seed value Turbine uses to approximate a random crit chance has failed you.

Great math, improperly and incompletely applied.

You conclusion should have been "the conclusion to be drawn from the OP's results is that FOR THIS SAMPLE, the crit chance was less than the indicated 62%".

Put another way, I'd be more concerned about Turbine's PRNG if every sample of 7 events had a perfect distribution. THAT would be much more indicative of improper operation than the OP's case.

Originally Posted by Solien

One streak that's in no jeopardy...any time someone is unhappy about the game and posts to vent their frustration, there will always be a large group of people waiting to pile on with "The problem is you/everything is WAI/stop whining and shut up" replies.

Another safe streak is that every time somebody rationally points out how the poster might, in fact, be incorrect, there will be people making the ad hominem suggestion that the reply is actually motivated by knee-jerk irrational loyalty.

Burned through 25 crit ingredients the other day and critted 10 times (70% crit chance). 40% actual is pretty bleh for 25 attempts. Luckily I was giving the stuff away, otherwise, this can only happen so many times before you start getting accused of thieving mats when crafting for others :/

I'm much more of a fan of increased crit % turning into increased gear or something that can be a little more dependent. But its only a minor frustration.

But I'm still a fan of the feast or famine theory with a twist of the broken dice theory. Mostly because that's what I see from my seat. Too many instances with 40 to 50 items dropping for 6 players to split and I leave with nothing but a broken club because I looted a corpse. I do win stuff... but I don't ever win 1/6th or 1/12th respectively, and I rarely reach crafting percentages unless I craft small and crit early. I also understand that because others get better rolls, its very easy to assume everyone else is bringing fixed vision to the discussion.

If I had to choose between better housing and better dice... I'd go with housing.

If I had to choose between better dice and constant character revamps that change things that seem to be working ok... I'd want a new set of dice.

I wish I had the foresight to have turned on my combat log to submit some numbers to Turbine. I was deeding the other day. Worms 20 levels below me. Full red traited with Mini and crit legacies in WS. I easily cast Piercing cry 400 times. Not only did it not crit 1 time in over 400 casts but it only ever hit for 2 different numbers. That is quite odd considering the range of the base skill plus the added range of both crit and devastate. I stand by my claim that at least my toon (because that is the only one I can see the numbers on) periodically saves a roll for a bit before re-rolling.