Guild Wars 2

Fixing forum bug again.

Halvorn.9831:

I’d rather program this. It gives a better impression.

int run = 1;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < 84; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}
if (doubles >= 10) {
System.out.println(String.format(“Run Nr: %d, Doubles: %d”, run, doubles));
}
run++;
} while (true);

Make 84 random number between 0 and 25 and count how many times the current one is the same as during the last draw (a “double”). Output whenever there is more than 9 doubles (together with the run nr to get a feeling for how often it happens) among the 84 random numbers. Repeat this forever.

Sample result:

Run Nr: 201, Doubles: 10
Run Nr: 1195, Doubles: 10
Run Nr: 1877, Doubles: 10
Run Nr: 2336, Doubles: 11
Run Nr: 2505, Doubles: 10
Run Nr: 3225, Doubles: 10
Run Nr: 3501, Doubles: 14
Run Nr: 3615, Doubles: 10
Run Nr: 4350, Doubles: 11
Run Nr: 4380, Doubles: 10
Run Nr: 4672, Doubles: 11
Run Nr: 6234, Doubles: 10
Run Nr: 7821, Doubles: 10
Run Nr: 9148, Doubles: 10

Can you see that? In the first 10000 runs I already have an attempt with 14 doubles.
The reason is that 84 is much too small as a sample size. There is still way too much fluctuation in that. If you choose a sample size of 10000, the number of doubles is much closer to 3.8% (1/26).

Thanks for doing this, but using this method of counting I had 15 doubles, which didn’t occur in your 10,000 samples, right?

Nevets Crimsonwing.5271:

Halvorn.9831:

I’d rather program this. It gives a better impression.

int run = 1;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < 84; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}
if (doubles >= 10) {
System.out.println(String.format(“Run Nr: %d, Doubles: %d”, run, doubles));
}
run++;
} while (true);

Make 84 random number between 0 and 25 and count how many times the current one is the same as during the last draw (a “double”). Output whenever there is more than 9 doubles (together with the run nr to get a feeling for how often it happens) among the 84 random numbers. Repeat this forever.

Sample result:

Run Nr: 201, Doubles: 10
Run Nr: 1195, Doubles: 10
Run Nr: 1877, Doubles: 10
Run Nr: 2336, Doubles: 11
Run Nr: 2505, Doubles: 10
Run Nr: 3225, Doubles: 10
Run Nr: 3501, Doubles: 14
Run Nr: 3615, Doubles: 10
Run Nr: 4350, Doubles: 11
Run Nr: 4380, Doubles: 10
Run Nr: 4672, Doubles: 11
Run Nr: 6234, Doubles: 10
Run Nr: 7821, Doubles: 10
Run Nr: 9148, Doubles: 10

Can you see that? In the first 10000 runs I already have an attempt with 14 doubles.
The reason is that 84 is much too small as a sample size. There is still way too much fluctuation in that. If you choose a sample size of 10000, the number of doubles is much closer to 3.8% (1/26).

Thanks for doing this, but using this method of counting I had 15 doubles, which didn’t occur in your 10,000 samples, right?

But are 15 doubles just an outlier or statistically significant? THAT’s the question. And 84 trials is not a high enough sample size to say it is or that it is not with high confidence.

The reward system in general is terrible. Its always appeared to me at least, that they’re just driving players into purchasing gems or using real money to try and get what they want. Maybe less with fractals, but in general thats what it seems like to me. I haven’t really talked with anyone in the game or friends who play that actually like the loot system.

Seera.5916:

Nevets Crimsonwing.5271:

Halvorn.9831:

I’d rather program this. It gives a better impression.

int run = 1;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < 84; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}
if (doubles >= 10) {
System.out.println(String.format(“Run Nr: %d, Doubles: %d”, run, doubles));
}
run++;
} while (true);

Make 84 random number between 0 and 25 and count how many times the current one is the same as during the last draw (a “double”). Output whenever there is more than 9 doubles (together with the run nr to get a feeling for how often it happens) among the 84 random numbers. Repeat this forever.

Sample result:

Run Nr: 201, Doubles: 10
Run Nr: 1195, Doubles: 10
Run Nr: 1877, Doubles: 10
Run Nr: 2336, Doubles: 11
Run Nr: 2505, Doubles: 10
Run Nr: 3225, Doubles: 10
Run Nr: 3501, Doubles: 14
Run Nr: 3615, Doubles: 10
Run Nr: 4350, Doubles: 11
Run Nr: 4380, Doubles: 10
Run Nr: 4672, Doubles: 11
Run Nr: 6234, Doubles: 10
Run Nr: 7821, Doubles: 10
Run Nr: 9148, Doubles: 10

Can you see that? In the first 10000 runs I already have an attempt with 14 doubles.
The reason is that 84 is much too small as a sample size. There is still way too much fluctuation in that. If you choose a sample size of 10000, the number of doubles is much closer to 3.8% (1/26).

Thanks for doing this, but using this method of counting I had 15 doubles, which didn’t occur in your 10,000 samples, right?

But are 15 doubles just an outlier or statistically significant? THAT’s the question. And 84 trials is not a high enough sample size to say it is or that it is not with high confidence.

I’m going to ask you again. Can you please run a sample size estimator statistical test to show me that a sample size of 84 doesn’t have strong statistical power?

Miku.6297:

Just because the odds of getting something are higher then the odds of getting something else, doesn’t mean it’s not a RNG loot system.

This should be obvious, right? Rolling two six-sided dice and taking the sum will result in a 7 much more often than a 2. But the result is still (assuming fair dice) random. Nothing presented in this thread should lead one to believe that the RNG behind the drop of rings from fractals is broken. It could be broken, but no convincing evidence of it has been shown. It’s all tinfoil hat stuff so far.

Nevets Crimsonwing.5271:

Derigar.7810:

Is this true? I only get the ones in said attached file.
I’m hoping to get a Berserker ring sometime soon, but it seems like prestine relics will be the way to buy it.

Pretty sure their RNG is complete crap because many times you’ll get doubles of the same ring even though the chances of that happening are really low statistically.

This. I only ever got a few Berserker rings. I did however have many, many multiples of Megellan’s Whorll, Rurik’s Signet Ring, Vassar’s Band, and Rallena’s Band…..like a lotttttt of those.

Fact is that they have admitted to their drop system and reward system being broken in Fractals before. Even modifiers like potions for dungeons were broken for a very long time. In theory it is all RNG but RNG has to be based on script yes? I’ve thought for a while that they might have some numbers wrong and either are not sure how to fix it or don’t care to.

I think the chest piece and leg peice armor box drops are broken too. maybe it’s just me but I have only gotten head hands feet and some shoulders over the past like 6 months.

Saint.5647:

Nevets Crimsonwing.5271:

Derigar.7810:

Is this true? I only get the ones in said attached file.
I’m hoping to get a Berserker ring sometime soon, but it seems like prestine relics will be the way to buy it.

Pretty sure their RNG is complete crap because many times you’ll get doubles of the same ring even though the chances of that happening are really low statistically.

This. I only ever got a few Berserker rings. I did however have many, many multiples of Megellan’s Whorll, Rurik’s Signet Ring, Vassar’s Band, and Rallena’s Band…..like a lotttttt of those.

Fact is that they have admitted to their drop system and reward system being broken in Fractals before. Even modifiers like potions for dungeons were broken for a very long time. In theory it is all RNG but RNG has to be based on script yes? I’ve thought for a while that they might have some numbers wrong and either are not sure how to fix it or don’t care to.

Yeah, I remember when they released Fractured! the fractal weapon drop rate was borked for a long, long time. I’m sure if someone had posted about that, with a low sample size because weapons are much rarer than rings, the same people here would cry “RNG IS RNG.”

Then ANET came out and said, “Lel woops guys we broked it.”

And then didn’t fix it fully to what it was before.

Nevets Crimsonwing.5271:

Awesome, thank you so much.

So, if I’m reading this properly you’ve calculated the cumulative probability of getting 15 or more doubles out of 84, not just exactly total of 15.

Correct though I have the data for excact 15 as well in my chart (second collum).

Perhaps you can explain one more thing to me – why would this model be preferable to a model where you view 2 drops as a “trial” and count successes as 2 of the same rings (with odds of 1 in 26) and a failure as 2 different rings (25/26). Under this model you’d have 42 trials, 8 successes.

http://stattrek.com/online-calculator/binomial.aspx

Under this model you get a kitten prob of 8 or more of 1 in about 6,000 which seems a lot more likely to be “RNG is RNG” than 1 in a 1.6 million.

Thoughts?

First thing in statistics is that how larger the sample how more accurate the result.
Secondly because that’s not how a player will experience it, he will not compare every set 2 of the same. He will look at his current ring and think about his previous ring evvery time he gets a ring. take a look at how it is highlighted, he didn’t took sets of two and compared them. He took every occurence of a double (even triple) and marked them.

Seera.5916:

But are 15 doubles just an outlier or statistically significant? THAT’s the question. And 84 trials is not a high enough sample size to say it is or that it is not with high confidence.

Every result is statistically significant. Especialy when you want to verify if something follows a certain distribution, that’s what you are looking for: an outlier so far off the average that the probability of occurence in the suggested distribution is way too small.

Tim.6450:

Nevets Crimsonwing.5271:

Awesome, thank you so much.

So, if I’m reading this properly you’ve calculated the cumulative probability of getting 15 or more doubles out of 84, not just exactly total of 15.

Correct though I have the data for excact 15 as well in my chart (second collum).

Perhaps you can explain one more thing to me – why would this model be preferable to a model where you view 2 drops as a “trial” and count successes as 2 of the same rings (with odds of 1 in 26) and a failure as 2 different rings (25/26). Under this model you’d have 42 trials, 8 successes.

http://stattrek.com/online-calculator/binomial.aspx

Under this model you get a kitten prob of 8 or more of 1 in about 6,000 which seems a lot more likely to be “RNG is RNG” than 1 in a 1.6 million.

Thoughts?

First thing in statistics is that how larger the sample how more accurate the result.
Secondly because that’s not how a player will experience it, he will not compare every set 2 of the same. He will look at his current ring and think about his previous ring evvery time he gets a ring. take a look at how it is highlighted, he didn’t took sets of two and compared them. He took every occurence of a double (even triple) and marked them.

I made the image and I initially highlighted each of the doubles and the one triple, but as an afterthought I was thinking it would be better to look at them as pairs because then each trial is independent. If a single drop is used as the second comparison drop AND as the first initial drop, the drop is no longer “independent.” One of the assumptions of using a binomial probability distribution is that each “trial” has to be independent. If you’re using the same ring in two trials, it’s no longer independent and the assumptions of your test are violated.

Nevets Crimsonwing.5271:

I made the image and I initially highlighted each of the doubles and the one triple, but as an afterthought I was thinking it would be better to look at them as pairs because then each trial is independent. If a single drop is used as the second comparison drop AND as the first initial drop, the drop is no longer “independent.” One of the assumptions of using a binomial probability distribution is that each “trial” has to be independent. If you’re using the same ring in two trials, it’s no longer independent and the assumptions of your test are violated.

How is it not independant? Your chance of success is always 1/26.

Tim.6450:

Nevets Crimsonwing.5271:

I made the image and I initially highlighted each of the doubles and the one triple, but as an afterthought I was thinking it would be better to look at them as pairs because then each trial is independent. If a single drop is used as the second comparison drop AND as the first initial drop, the drop is no longer “independent.” One of the assumptions of using a binomial probability distribution is that each “trial” has to be independent. If you’re using the same ring in two trials, it’s no longer independent and the assumptions of your test are violated.

How is it not independant? Your chance of success is always 1/26.

I’m becoming more convinced that you’re correct, but for whatever reason I thought that you can’t use a single drop in two trials.

Nevets Crimsonwing.5271:

I’m becoming more convinced that you’re correct, but for whatever reason I thought that you can’t use a single drop in two trials.

you are confusing the statiscal independance with independance.

independance means that you’re previous result will not influence the next one which it does in a way, if you got a zerker ring in your first test and a rampager in your second. you would not have a double. so that previous roll influenced your result so it’s not independant.

In statistics the only thing that matters are the probabilities; The definition (from wikipedia) is:
“Two events are independent if and only if their joint probability equals the product of their probabilities.” Now let’s put this on my case:
What is the chance you get a double and the previous ring was a zerker? 1/26*1/26= chance previous ring a zerker*chance double. So the previous result and the chance of getting a double is independant.

Tim.6450:

Nevets Crimsonwing.5271:

I’m becoming more convinced that you’re correct, but for whatever reason I thought that you can’t use a single drop in two trials.

you are confusing the statiscal independance with independance.

independance means that you’re previous result will not influence the next one which it does in a way, if you got a zerker ring in your first test and a rampager in your second. you would not have a double. so that previous roll influenced your result so it’s not independant.

In statistics the only thing that matters are the probabilities; The definition (from wikipedia) is:
“Two events are independent if and only if their joint probability equals the product of their probabilities.” Now let’s put this on my case:
What is the chance you get a double and the previous ring was a zerker? 1/26*1/26= chance previous ring a zerker*chance double. So the previous result and the chance of getting a double is independant.

On the other thread I finally understood what you guys were saying, and I completely agree with you. When I did the math using an online calculator I got results of 1 in 1.63 million which is basically what you said.

So there you guys have it.

The odds of my number of doubles are about 1 in 1.63 million, which COULD be due to chance, though I’m inclined to believe otherwise, due to others getting similar results.

Nevets Crimsonwing.5271:

The odds of my number of doubles are about 1 in 1.63 million, which COULD be due to chance,

You still have less chance to win euromillions then getting this kind of result and that does happen as well.

though I’m inclined to believe otherwise, due to others getting similar results.

Well the problem is that the one who post about will get similar or bad result as well while the positive result are less likely to respond (especially since forums are for addressing problems therefore no problems=>no post)

Nevets Crimsonwing.5271:

Seera.5916:

Nevets Crimsonwing.5271:

Halvorn.9831:

I’d rather program this. It gives a better impression.

int run = 1;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < 84; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}
if (doubles >= 10) {
System.out.println(String.format(“Run Nr: %d, Doubles: %d”, run, doubles));
}
run++;
} while (true);

Make 84 random number between 0 and 25 and count how many times the current one is the same as during the last draw (a “double”). Output whenever there is more than 9 doubles (together with the run nr to get a feeling for how often it happens) among the 84 random numbers. Repeat this forever.

Sample result:

Run Nr: 201, Doubles: 10
Run Nr: 1195, Doubles: 10
Run Nr: 1877, Doubles: 10
Run Nr: 2336, Doubles: 11
Run Nr: 2505, Doubles: 10
Run Nr: 3225, Doubles: 10
Run Nr: 3501, Doubles: 14
Run Nr: 3615, Doubles: 10
Run Nr: 4350, Doubles: 11
Run Nr: 4380, Doubles: 10
Run Nr: 4672, Doubles: 11
Run Nr: 6234, Doubles: 10
Run Nr: 7821, Doubles: 10
Run Nr: 9148, Doubles: 10

Can you see that? In the first 10000 runs I already have an attempt with 14 doubles.
The reason is that 84 is much too small as a sample size. There is still way too much fluctuation in that. If you choose a sample size of 10000, the number of doubles is much closer to 3.8% (1/26).

Thanks for doing this, but using this method of counting I had 15 doubles, which didn’t occur in your 10,000 samples, right?

But are 15 doubles just an outlier or statistically significant? THAT’s the question. And 84 trials is not a high enough sample size to say it is or that it is not with high confidence.

I’m going to ask you again. Can you please run a sample size estimator statistical test to show me that a sample size of 84 doesn’t have strong statistical power?

It doesnt. You dont even NEED to run a test to tell you that.

• For starters, you dont even know what the RNG formula is, so any test is FUBARed and false-premises to begin with.
• Second, you dont know what information is used to determine chances of what loot (aside from any statistical values of your account being irrelevant [as stated multiple times by people who know a hell of a lot more than people on the forums with tinfoil hats]).
• Third, your micro-sample size is just too small to start with to even give serious consideration to. Is it 84 runs period (nice RNG getting rings every run)? Or is it just the 84 runs that you got rings (out of how many runs? which also skews your ‘proof’)? Or is it just 84 convenient rings in your inventory/bank you’re using?

Finally, and hopefully this puts this thread out of it’s misery, RNG is functioning absolutely fine because it’s creating outliers as based on OP’s poor sample size, not because the OP’s results ARE an outlier (absent about another 900+ run results). No outliers? Something’s funky or anet actually lessened the odds of outliers occurring (something JS pursued a thread about for a while).

edit: @tim, the drops are fully independent regardless of situation. You could drop a zerker ring, followed by a rampager ring, then followed by a magi ring. One does not affect the previous or following result in any form. Mind you, this is assumed from simple logic and the fact that we dont know the internals of the RNG system

edit2: also, sample size is a massive issue that you really need to understand. What seems abnormal and an issue at <100 “tests” will be normal at >100,000 “tests.”

Aidan Savage.2078:

• For starters, you dont even know what the RNG formula is, so any test is FUBARed and false-premises to begin with.

That doesn’t matter a true RNG has properties no matter the formula. One of them is previous results don’t affect the next result. You can test such things no matter the formula.

• Second, you dont know what information is used to determine chances of what loot (aside from any statistical values of your account being irrelevant [as stated multiple times by people who know a hell of a lot more than people on the forums with tinfoil hats]).

That doesn’t matter the thing the OP wants to prove is that the current way of determing the chances is not completely random. He wants to show this by pointing out the occurence of doubles is far higher then it should be. The only thing you need for that are results and the knowledge of the distribution he wants to disprove. Anet for all I care may use the position of the moon to determine the loot. The only thing that matters is what comes out of the loot, how often, and when.

• Third, your micro-sample size is just too small to start with to even give serious consideration to. Is it 84 runs period (nice RNG getting rings every run)? Or is it just the 84 runs that you got rings (out of how many runs? which also skews your ‘proof’)? Or is it just 84 convenient rings in your inventory/bank you’re using?

Prove it, statistics have formula’s to prove/disprove hypotheses , these formula’s are based on the size of the sample. Show me such formula and fill in the numbers, then you can talk about sample size.

edit: @tim, the drops are fully independent regardless of situation. You could drop a zerker ring, followed by a rampager ring, then followed by a magi ring. One does not affect the previous or following result in any form. Mind you, this is assumed from simple logic and the fact that we dont know the internals of the RNG system

But getting a zerker ring will influence what you need to get for a double which was what I wanted to point out.

Aidan Savage.2078:

Nevets Crimsonwing.5271:

Seera.5916:

Nevets Crimsonwing.5271:

Halvorn.9831:

I’d rather program this. It gives a better impression.

int run = 1;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < 84; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}
if (doubles >= 10) {
System.out.println(String.format(“Run Nr: %d, Doubles: %d”, run, doubles));
}
run++;
} while (true);

Make 84 random number between 0 and 25 and count how many times the current one is the same as during the last draw (a “double”). Output whenever there is more than 9 doubles (together with the run nr to get a feeling for how often it happens) among the 84 random numbers. Repeat this forever.

Sample result:

Run Nr: 201, Doubles: 10
Run Nr: 1195, Doubles: 10
Run Nr: 1877, Doubles: 10
Run Nr: 2336, Doubles: 11
Run Nr: 2505, Doubles: 10
Run Nr: 3225, Doubles: 10
Run Nr: 3501, Doubles: 14
Run Nr: 3615, Doubles: 10
Run Nr: 4350, Doubles: 11
Run Nr: 4380, Doubles: 10
Run Nr: 4672, Doubles: 11
Run Nr: 6234, Doubles: 10
Run Nr: 7821, Doubles: 10
Run Nr: 9148, Doubles: 10

Can you see that? In the first 10000 runs I already have an attempt with 14 doubles.
The reason is that 84 is much too small as a sample size. There is still way too much fluctuation in that. If you choose a sample size of 10000, the number of doubles is much closer to 3.8% (1/26).

Thanks for doing this, but using this method of counting I had 15 doubles, which didn’t occur in your 10,000 samples, right?

But are 15 doubles just an outlier or statistically significant? THAT’s the question. And 84 trials is not a high enough sample size to say it is or that it is not with high confidence.

I’m going to ask you again. Can you please run a sample size estimator statistical test to show me that a sample size of 84 doesn’t have strong statistical power?

It doesnt. You dont even NEED to run a test to tell you that.

• For starters, you dont even know what the RNG formula is, so any test is FUBARed and false-premises to begin with.
• Second, you dont know what information is used to determine chances of what loot (aside from any statistical values of your account being irrelevant [as stated multiple times by people who know a hell of a lot more than people on the forums with tinfoil hats]).
• Third, your micro-sample size is just too small to start with to even give serious consideration to. Is it 84 runs period (nice RNG getting rings every run)? Or is it just the 84 runs that you got rings (out of how many runs? which also skews your ‘proof’)? Or is it just 84 convenient rings in your inventory/bank you’re using?

Finally, and hopefully this puts this thread out of it’s misery, RNG is functioning absolutely fine because it’s creating outliers as based on OP’s poor sample size, not because the OP’s results ARE an outlier (absent about another 900+ run results). No outliers? Something’s funky or anet actually lessened the odds of outliers occurring (something JS pursued a thread about for a while).

edit: @tim, the drops are fully independent regardless of situation. You could drop a zerker ring, followed by a rampager ring, then followed by a magi ring. One does not affect the previous or following result in any form. Mind you, this is assumed from simple logic and the fact that we dont know the internals of the RNG system

edit2: also, sample size is a massive issue that you really need to understand. What seems abnormal and an issue at <100 “tests” will be normal at >100,000 “tests.”

Yeah, because in the vast majority of science literature they have 100,000 samples.

/sarc

I mean if you want to play ball with stats here, please do. But if you’re just going to make things up, please go make your own thread.

Nevets Crimsonwing.5271:

Yeah, because in the vast majority of science literature they have 100,000 samples.

/sarc

I mean if you want to play ball with stats here, please do. But if you’re just going to make things up, please go make your own thread.

But they run WAY more than 84 trials. Some run several HUNDRED trials. Your exaggeration doesn’t help your case.

The conclusions you draw from the results are based on a number of assumptions and confirmation bias (everyone suffers from confirmation bias to some degree). It’s also skewed due to lack of information on no ring drop runs. Your theory holds more weight if there is a low range of numbers of no ring runs between doubles (like 3, 3, 4, 2, 3, 4) than if some are small times and others are long (like 2, 6, 12, 5, 7, 0). Without the data for how many no ring runs you have, any conclusion isn’t that solid.

Also, your trials were all done on varying Fractal Levels. Which due to our lack of knowledge of loot tables, adds another variable: some levels may have a higher or lower chance for certain rings. Which we don’t know about. That varying probability for a specific ring affects conclusions that can or can not be drawn.

Do your results warrant further testing with better records of what you got in the ring department? Yes. Do they conclusively show that there is something wrong with the RNG and/or loot table? No.

I am not giving up.

int run = 1;
int samples = 84;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < samples; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}

float percentage = doubles/(float)samples * 100f;
if (percentage > {
System.out.println(String.format(“Run Nr: %d, Doubles percentage: %2.1f”, run, percentage));
}
run++;
} while (true);

This time the small program calculates the percentage of doubles on the whole sample and prints it out, if it is > 8 percent.

Output:
Run Nr: 4, Doubles percentage: 8,3
Run Nr: 18, Doubles percentage: 8,3
Run Nr: 40, Doubles percentage: 9,5
Run Nr: 88, Doubles percentage: 8,3
Run Nr: 116, Doubles percentage: 8,3
Run Nr: 125, Doubles percentage: 9,5
Run Nr: 148, Doubles percentage: 8,3
Run Nr: 151, Doubles percentage: 8,3
Run Nr: 196, Doubles percentage: 8,3
Run Nr: 211, Doubles percentage: 9,5
Run Nr: 228, Doubles percentage: 9,5
Run Nr: 232, Doubles percentage: 10,7
Run Nr: 234, Doubles percentage: 10,7
Run Nr: 259, Doubles percentage: 8,3
Run Nr: 267, Doubles percentage: 8,3
Run Nr: 279, Doubles percentage: 8,3
Run Nr: 339, Doubles percentage: 8,3
Run Nr: 412, Doubles percentage: 8,3
Run Nr: 422, Doubles percentage: 8,3
Run Nr: 430, Doubles percentage: 8,3

Now I am changing the sample size from 84 to 840.

Output:

Even after 10 minutes of looping not a single sample had more than 8% doubles. Because at this sample size the result is much closer to the expected average of 1/26, ca. 3.8%

Your sample size is at least factor 10 too small to contain any clues on anomalies in the RNG distribution. Even for an even distribution of random results (read: the RNG produces white noise when viewed as a frequency graph) your results are expectable because of your sample size.

Halvorn.9831:

Even after 10 minutes of looping not a single sample had more than 8% doubles. Because at this sample size the result is much closer to the expected average of 1/26, ca. 3.8%

Your sample size is at least factor 10 too small to contain any clues on anomalies in the RNG distribution. Even for an even distribution of random results (read: the RNG produces white noise when viewed as a frequency graph) your results are expectable because of your sample size.

First I assume this is java code, then I have to dissapoint you. Math.random() is a pseudo random number generator and therefore is made to not show such anomalies.

I knew somebody will bring this up. Somebody always brings this up.

And it is wrong.

First of all, I am working at a company which does hardware security modules. They have true random generators, I’ve also used one of them and there is absolutely no difference in the behavior (the code is just too clumsy to show here, since it is a network device call).

Secondly, the major difference between a prng and a trng is that the first one creates identical sequence of numbers when seeded identically. Apart from that, the statistical distribution of values is the same considering a huge sample size.

Which is what I said.

Halvorn.9831:

I am not giving up.

int run = 1;
int samples = 84;
do {
int lastRnd = -1;
int doubles = 0;
for (int count = 0; count < samples; count++) {
int rnd = (int) (Math.random() * 26);
if (rnd == lastRnd) {
doubles++;
}
lastRnd = rnd;
}

float percentage = doubles/(float)samples * 100f;
if (percentage > {
System.out.println(String.format(“Run Nr: %d, Doubles percentage: %2.1f”, run, percentage));
}
run++;
} while (true);

This time the small program calculates the percentage of doubles on the whole sample and prints it out, if it is > 8 percent.

Output:
Run Nr: 4, Doubles percentage: 8,3
Run Nr: 18, Doubles percentage: 8,3
Run Nr: 40, Doubles percentage: 9,5
Run Nr: 88, Doubles percentage: 8,3
Run Nr: 116, Doubles percentage: 8,3
Run Nr: 125, Doubles percentage: 9,5
Run Nr: 148, Doubles percentage: 8,3
Run Nr: 151, Doubles percentage: 8,3
Run Nr: 196, Doubles percentage: 8,3
Run Nr: 211, Doubles percentage: 9,5
Run Nr: 228, Doubles percentage: 9,5
Run Nr: 232, Doubles percentage: 10,7
Run Nr: 234, Doubles percentage: 10,7
Run Nr: 259, Doubles percentage: 8,3
Run Nr: 267, Doubles percentage: 8,3
Run Nr: 279, Doubles percentage: 8,3
Run Nr: 339, Doubles percentage: 8,3
Run Nr: 412, Doubles percentage: 8,3
Run Nr: 422, Doubles percentage: 8,3
Run Nr: 430, Doubles percentage: 8,3

Now I am changing the sample size from 84 to 840.

Output:

Even after 10 minutes of looping not a single sample had more than 8% doubles. Because at this sample size the result is much closer to the expected average of 1/26, ca. 3.8%

Your sample size is at least factor 10 too small to contain any clues on anomalies in the RNG distribution. Even for an even distribution of random results (read: the RNG produces white noise when viewed as a frequency graph) your results are expectable because of your sample size.

Why 8%?

15 doubles in 83 trials is 18%.

Also, to the person who said most science trials have hundreds of trials, I’m going to have to beg to differ. As someone who is a biochemist by trade with a number of publications myself, who consistently reads scientifikittenerature, that’s not been my experience. Also, before you say you don’t believe me, if you really want to read my articles, please pm me.

I was going to chime in to say the same thing along the same lines as well. I’ve spent the last few years working in research and very rarely have I ever seen a study with an abnormally high amount of trials. I don’t know where you have seen hundreds of trials run outside of simulations. In most situations it’s simply not feasible.

That said, my main concern with the Fractal rings is not the overall use of RNG. It’s that ANet may have written errors into the script. I wonder whether it is really independent or not. We already know that they have made errors regarding Ascended drops and Weapons in the Fractal RNG code. It isn’t reckless to assume that they may have made errors with the various ring types as well.

My thought on it is that they were unable to code a proper RNG system to they tied drops to occurring variables in the Fractal or the gameplay. This is much easier and sadly I don’t have the faith in ANet to suggest they wouldn’t do something like this.

Saint.5647:

I was going to chime in to say the same thing along the same lines as well. I’ve spent the last few years working in research and very rarely have I ever seen a study with an abnormally high amount of trials. I don’t know where you have seen hundreds of trials run outside of simulations. In most situations it’s simply not feasible.

That tells you something about the credibility of some research. This is a completely different topic though: there are some areas of “research” where there has to be a result, and if the sigma cannot be good enough, you just ignore it, rather than improving your sample numbers. There are lots of studies out there about the weaknesses of studies.

But that is not what we are talking about. What we are talking about is that, considering an even spread, the results presented as a proof of bad behavior are totally within the range of expectable and explainable results – given the small sample size.

I guess I am giving up on this topic. Conspiracy theories are successful because they appeal to weaknesses in our psyche. Secret rules in the code, the effects of PRNG, the aim to make more money through gem sales, it all makes sense suddenly…. not.

I look at MY code which shows me that the behavior is explainable without any dark secrets. That is my creed: it is all in the code.

Halvorn.9831:

Saint.5647:

I was going to chime in to say the same thing along the same lines as well. I’ve spent the last few years working in research and very rarely have I ever seen a study with an abnormally high amount of trials. I don’t know where you have seen hundreds of trials run outside of simulations. In most situations it’s simply not feasible.

That tells you something about the credibility of some research. This is a completely different topic though: there are some areas of “research” where there has to be a result, and if the sigma cannot be good enough, you just ignore it, rather than improving your sample numbers. There are lots of studies out there about the weaknesses of studies.

But that is not what we are talking about. What we are talking about is that, considering an even spread, the results presented as a proof of bad behavior are totally within the range of expectable and explainable results – given the small sample size.

I guess I am giving up on this topic. Conspiracy theories are successful because they appeal to weaknesses in our psyche. Secret rules in the code, the effects of PRNG, the aim to make more money through gem sales, it all makes sense suddenly…. not.

I look at MY code which shows me that the behavior is explainable without any dark secrets. That is my creed: it is all in the code.

Your code came up with zero examples of 18% doubles in hundreds of runs.

Halvorn, run your code until you see an output with 18% doubles. I’m curious to see how close to 1.6 million trials you’ll need.

The math sounds really bad to me. I’ve read through only the first thread.

This reminds me of a discussion I had with my friends after a gambling session. i believe we were palying roulette. Read about the Gambler’s Fallacy and see if that applies and helps you to understand:
http://en.wikipedia.org/wiki/Gambler%27s_fallacy

. . .probability of getting heads on a single toss is exactly 1?2 (one in two). It follows that the probability of getting two heads in two tosses is 1?4 (one in four) and the probability of getting three heads in three tosses is 1?8 (one in eight).

Now suppose that we have just tossed four heads in a row, so that if the next coin toss were also to come up heads, it would complete a run of five successive heads. Since the probability of a run of five successive heads is only 1?32 (one in thirty-two), a person subject to the gambler’s fallacy might believe that this next flip was less likely to be heads than to be tails. However, this is not correct, and is a manifestation of the gambler’s fallacy; the event of 5 heads in a row and the event of “first 4 heads, then a tails” are equally likely, each having probability 1?32. Given that the first four rolls turn up heads, the probability that the next toss is a head is in fact . . . 1/2.

While a run of five heads is only 1?32 = 0.03125, it is only that before the coin is first tossed. After the first four tosses the results are no longer unknown, so their probabilities are 1. Reasoning that it is more likely that the next toss will be a tail than a head due to the past tosses, that a run of luck in the past somehow influences the odds in the future, is the fallacy.

Also there are other things to come into play. there are 26 outcomes but maybe they are not all weighted equally. If you really wanted to test this ask people to open up their banks and show their rings. I have some duplicates but otherwise a great variety. The OP has only gotten 5 rings………. they aren’t even all the same 5. i would also suggest to run more fractals and tell us what your next 5 rings are? next 10 rings are? etc.

Benson.9063:

The math sounds really bad to me. I’ve read through only the first thread.

This reminds me of a discussion I had with my friends after a gambling session. i believe we were palying roulette. Read about the Gambler’s Fallacy and see if that applies and helps you to understand:
http://en.wikipedia.org/wiki/Gambler%27s_fallacy

. . .probability of getting heads on a single toss is exactly 1?2 (one in two). It follows that the probability of getting two heads in two tosses is 1?4 (one in four) and the probability of getting three heads in three tosses is 1?8 (one in eight).

Now suppose that we have just tossed four heads in a row, so that if the next coin toss were also to come up heads, it would complete a run of five successive heads. Since the probability of a run of five successive heads is only 1?32 (one in thirty-two), a person subject to the gambler’s fallacy might believe that this next flip was less likely to be heads than to be tails. However, this is not correct, and is a manifestation of the gambler’s fallacy; the event of 5 heads in a row and the event of “first 4 heads, then a tails” are equally likely, each having probability 1?32.

While a run of five heads is only 1?32 = 0.03125, it is only that before the coin is first tossed. After the first four tosses the results are no longer unknown, so their probabilities are 1. Reasoning that it is more likely that the next toss will be a tail than a head due to the past tosses, that a run of luck in the past somehow influences the odds in the future, is the fallacy.

Also there are other things to come into play. there are 26 outcomes but maybe they are not all weighted equally. If you really wanted to test this ask people to open up their banks and show their rings. I have some duplicates but otherwise a great variety. The OP has only gotten 5 rings………. they aren’t even all the same 5. i would also suggest to run more fractals and tell us what your next 5 rings are? next 10 rings are? etc.

I’d suggest reading the rest of the thread. Everyone dismissed the op so I sort of hijacked the thread with an ordered sample size of 84 rings, and multiple people have done the math on those to get odds of about 1 in 1.6 million.

Nevets Crimsonwing.5271:

Benson.9063:

The math sounds really bad to me. I’ve read through only the first thread.

This reminds me of a discussion I had with my friends after a gambling session. i believe we were palying roulette. Read about the Gambler’s Fallacy and see if that applies and helps you to understand:
http://en.wikipedia.org/wiki/Gambler%27s_fallacy

. . .probability of getting heads on a single toss is exactly 1?2 (one in two). It follows that the probability of getting two heads in two tosses is 1?4 (one in four) and the probability of getting three heads in three tosses is 1?8 (one in eight).

Now suppose that we have just tossed four heads in a row, so that if the next coin toss were also to come up heads, it would complete a run of five successive heads. Since the probability of a run of five successive heads is only 1?32 (one in thirty-two), a person subject to the gambler’s fallacy might believe that this next flip was less likely to be heads than to be tails. However, this is not correct, and is a manifestation of the gambler’s fallacy; the event of 5 heads in a row and the event of “first 4 heads, then a tails” are equally likely, each having probability 1?32.

While a run of five heads is only 1?32 = 0.03125, it is only that before the coin is first tossed. After the first four tosses the results are no longer unknown, so their probabilities are 1. Reasoning that it is more likely that the next toss will be a tail than a head due to the past tosses, that a run of luck in the past somehow influences the odds in the future, is the fallacy.

Also there are other things to come into play. there are 26 outcomes but maybe they are not all weighted equally. If you really wanted to test this ask people to open up their banks and show their rings. I have some duplicates but otherwise a great variety. The OP has only gotten 5 rings………. they aren’t even all the same 5. i would also suggest to run more fractals and tell us what your next 5 rings are? next 10 rings are? etc.

I’d suggest reading the rest of the thread. Everyone dismissed the op so I sort of hijacked the thread with an ordered sample size of 84 rings, and multiple people have done the math on those to get odds of about 1 in 1.6 million.

And just because you’ve hit that 1 in 1.6 million chances doesn’t mean there’s something wrong with the system.

As to studies that use several hundred trials: phase 3 drug testing. And I would imagine cheap, safe, and quick experiments in non-biology or non-biochemistry labs probably get a large trial size as well.

You said the trials were on the different fractal levels. And you haven’t given the data of what trials yielded no rings.

I would imagine the more uncontrolled variables you have the more trials you have to run to ensure that the uncontrolled, non-studied variables aren’t affecting the results. Levels 21-50 can give infused rings (you said all rings shown are infused, right? If so that’s another variable – non-infused rings). That’s 30 levels. That’s on average 2.8 trials per level. That’s not enough trials to ensure that the differences between their loot tables is not affecting the results. Average per level would be higher if you didn’t do specific levels but you didn’t mention which levels you did.

Seera.5916:

Nevets Crimsonwing.5271:

Benson.9063:

The math sounds really bad to me. I’ve read through only the first thread.

This reminds me of a discussion I had with my friends after a gambling session. i believe we were palying roulette. Read about the Gambler’s Fallacy and see if that applies and helps you to understand:
http://en.wikipedia.org/wiki/Gambler%27s_fallacy

. . .probability of getting heads on a single toss is exactly 1?2 (one in two). It follows that the probability of getting two heads in two tosses is 1?4 (one in four) and the probability of getting three heads in three tosses is 1?8 (one in eight).

Now suppose that we have just tossed four heads in a row, so that if the next coin toss were also to come up heads, it would complete a run of five successive heads. Since the probability of a run of five successive heads is only 1?32 (one in thirty-two), a person subject to the gambler’s fallacy might believe that this next flip was less likely to be heads than to be tails. However, this is not correct, and is a manifestation of the gambler’s fallacy; the event of 5 heads in a row and the event of “first 4 heads, then a tails” are equally likely, each having probability 1?32.

While a run of five heads is only 1?32 = 0.03125, it is only that before the coin is first tossed. After the first four tosses the results are no longer unknown, so their probabilities are 1. Reasoning that it is more likely that the next toss will be a tail than a head due to the past tosses, that a run of luck in the past somehow influences the odds in the future, is the fallacy.

Also there are other things to come into play. there are 26 outcomes but maybe they are not all weighted equally. If you really wanted to test this ask people to open up their banks and show their rings. I have some duplicates but otherwise a great variety. The OP has only gotten 5 rings………. they aren’t even all the same 5. i would also suggest to run more fractals and tell us what your next 5 rings are? next 10 rings are? etc.

I’d suggest reading the rest of the thread. Everyone dismissed the op so I sort of hijacked the thread with an ordered sample size of 84 rings, and multiple people have done the math on those to get odds of about 1 in 1.6 million.

And just because you’ve hit that 1 in 1.6 million chances doesn’t mean there’s something wrong with the system.

As to studies that use several hundred trials: phase 3 drug testing. And I would imagine cheap, safe, and quick experiments in non-biology or non-biochemistry labs probably get a large trial size as well.

You said the trials were on the different fractal levels. And you haven’t given the data of what trials yielded no rings.

I would imagine the more uncontrolled variables you have the more trials you have to run to ensure that the uncontrolled, non-studied variables aren’t affecting the results. Levels 21-50 can give infused rings (you said all rings shown are infused, right? If so that’s another variable – non-infused rings). That’s 30 levels. That’s on average 2.8 trials per level. That’s not enough trials to ensure that the differences between their loot tables is not affecting the results. Average per level would be higher if you didn’t do specific levels but you didn’t mention which levels you did.

A few things.

Stage 3 clinical trials are used to show super rare adverse affects that don’t occur in the normal population. They are only permitted AFTER a trial with a few dozen subjects convinces people that the drug is safe.

Again, a sample size of a few dozen or even smaller is enough for an alpha of .05 and power of .8.

All my data purport to show is that if you roll a ring, you’re not equally likely to get any of the 26. For whatever reason.

As to the rest of your post, I still see a bunch of claims being made with no numbers

Nevets Crimsonwing.5271:

Seera.5916:

Nevets Crimsonwing.5271:

Benson.9063:

The math sounds really bad to me. I’ve read through only the first thread.

This reminds me of a discussion I had with my friends after a gambling session. i believe we were palying roulette. Read about the Gambler’s Fallacy and see if that applies and helps you to understand:
http://en.wikipedia.org/wiki/Gambler%27s_fallacy

. . .probability of getting heads on a single toss is exactly 1?2 (one in two). It follows that the probability of getting two heads in two tosses is 1?4 (one in four) and the probability of getting three heads in three tosses is 1?8 (one in eight).

Now suppose that we have just tossed four heads in a row, so that if the next coin toss were also to come up heads, it would complete a run of five successive heads. Since the probability of a run of five successive heads is only 1?32 (one in thirty-two), a person subject to the gambler’s fallacy might believe that this next flip was less likely to be heads than to be tails. However, this is not correct, and is a manifestation of the gambler’s fallacy; the event of 5 heads in a row and the event of “first 4 heads, then a tails” are equally likely, each having probability 1?32.

While a run of five heads is only 1?32 = 0.03125, it is only that before the coin is first tossed. After the first four tosses the results are no longer unknown, so their probabilities are 1. Reasoning that it is more likely that the next toss will be a tail than a head due to the past tosses, that a run of luck in the past somehow influences the odds in the future, is the fallacy.

Also there are other things to come into play. there are 26 outcomes but maybe they are not all weighted equally. If you really wanted to test this ask people to open up their banks and show their rings. I have some duplicates but otherwise a great variety. The OP has only gotten 5 rings………. they aren’t even all the same 5. i would also suggest to run more fractals and tell us what your next 5 rings are? next 10 rings are? etc.

I’d suggest reading the rest of the thread. Everyone dismissed the op so I sort of hijacked the thread with an ordered sample size of 84 rings, and multiple people have done the math on those to get odds of about 1 in 1.6 million.

And just because you’ve hit that 1 in 1.6 million chances doesn’t mean there’s something wrong with the system.

As to studies that use several hundred trials: phase 3 drug testing. And I would imagine cheap, safe, and quick experiments in non-biology or non-biochemistry labs probably get a large trial size as well.

You said the trials were on the different fractal levels. And you haven’t given the data of what trials yielded no rings.

I would imagine the more uncontrolled variables you have the more trials you have to run to ensure that the uncontrolled, non-studied variables aren’t affecting the results. Levels 21-50 can give infused rings (you said all rings shown are infused, right? If so that’s another variable – non-infused rings). That’s 30 levels. That’s on average 2.8 trials per level. That’s not enough trials to ensure that the differences between their loot tables is not affecting the results. Average per level would be higher if you didn’t do specific levels but you didn’t mention which levels you did.

A few things.

Stage 3 clinical trials are used to show super rare adverse affects that don’t occur in the normal population. They are only permitted AFTER a trial with a few dozen subjects convinces people that the drug is safe.

Again, a sample size of a few dozen or even smaller is enough for an alpha of .05 and power of .8.

All my data purport to show is that if you roll a ring, you’re not equally likely to get any of the 26. For whatever reason.

As to the rest of your post, I still see a bunch of claims being made with no numbers

You asked me what experiments use several hundred trials. I answered. I could have been facetious and said phase 4. The keep an eye on the reported problems when the general population gets their hands on the drugs. But I decided to be serious.

And with the number of trials you’ve done and the varying variables that may affect the results, you can’t say that doubles are anymore likely than singles. We don’t know if the chances of a specific ring change from level to level or not. So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

ok i read through more of it and now i wish i hadn’t.

but that 1 in 1.6 million number… what does that represent the chances of? rolling duplicates? or rolling duplicates consecutively? i don’t get what you are trying to prove to be honest. Rolling duplicates – there is a very high chance of duplicates the higher and higher you get don’t you think? there’s only 26 outcomes. once you achieve all of the outcomes, you have 100% chance to get duplicates.

So i think you are trying to get the probability of obtaining the same ring consecutively with the math that Tim provided you and based on your picture; that is a very specific hypothesis and not the only claim you are making in this thread (and i’m not sure if you’ve even made that specific claim yet).

Why are you trying to figure this out? If you believe that if you get one ring you will then have a far greater chance to get it again on the next run, you should use this to your advantage – try to figure out if maybe you have higher chance to get the same ring on a sunday, or on back to back runs, or at night-time, or with your warrior, etc. if you get a ring you also want for another alt, then repeat and give yourself a better shot at it, if not come back later.

let me know if my logic anywhere if flawed. btw… you really put your rings in your bank in the order that you got them? and you haven’t touched them since? O_O

*edited to be more clear.

Halvorn.9831:

Secondly, the major difference between a prng and a trng is that the first one creates identical sequence of numbers when seeded identically. Apart from that, the statistical distribution of values is the same considering a huge sample size.

Which is what I said.

The major difference with prng is that there is one sequence that is constantly repeated. The seed is just where you start your sequence. Given this knowledge there are things that have no represenation in prng even though you could make your sample size big enough for that event to reliable occur in trng.

So taking a big sample size and then showing that an equivalent event did no occur, has little meaning in prng.

Benson.9063:

ok i read through more of it and now i wish i hadn’t.

but that 1 in 1.6 million number… what does that represent the chances of? rolling duplicates? or rolling duplicates consecutively? i don’t get what you are trying to prove to be honest. Rolling duplicates – there is a very high chance of duplicates the higher and higher you get don’t you think? there’s only 26 outcomes. once you achieve all of the outcomes, you have 100% chance to get duplicates.

So i think you are trying to get the probability of obtaining the same ring consecutively with the math that Tim provided you and based on your picture; that is a very specific hypothesis and not the only claim you are making in this thread (and i’m not sure if you’ve even made that specific claim yet).

Why are you trying to figure this out? If you believe that if you get one ring you will then have a far greater chance to get it again on the next run, you should use this to your advantage – try to figure out if maybe you have higher chance to get the same ring on a sunday, or on back to back runs, or at night-time, or with your warrior, etc. if you get a ring you also want for another alt, then repeat and give yourself a better shot at it, if not come back later.

let me know if my logic anywhere if flawed. btw… you really put your rings in your bank in the order that you got them? and you haven’t touched them since? O_O

*edited to be more clear.

Well what we were trying to (dis)prove (I’m not sure whether this thread is on track anymore) was that the rng for rings is not an independant. Meaning a ring’s chances does not change depending on the result before. Nevets tries this by placing a counter example. His example has 15 occurences where the drawn ring was the same as the previous ring drawn in a batch of 84 rings. The chance of this or worse (more “doubles”) occurring is 1 in 1.6 million.

Seera.5916:

You asked me what experiments use several hundred trials. I answered. I could have been facetious and said phase 4. The keep an eye on the reported problems when the general population gets their hands on the drugs. But I decided to be serious.

And with the number of trials you’ve done and the varying variables that may affect the results, you can’t say that doubles are anymore likely than singles. We don’t know if the chances of a specific ring change from level to level or not. So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

You asked me what experiments use several hundred trials

I never asked you that.

you can’t say that doubles are anymore likely than singles.

You haven’t run a sample size estimator and have provided no numbers here to make that claim.

So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

Yes, if you wanted to test the claim that different levels have different drop rates.

Tim.6450:

Well what we were trying to (dis)prove (I’m not sure whether this thread is on track anymore) was that the rng for rings is not an independant. Meaning a ring’s chances does not change depending on the result before. Nevets tries this by placing a counter example. His example has 15 occurences where the drawn ring was the same as the previous ring drawn in a batch of 84 rings. The chance of this or worse (more “doubles”) occurring is 1 in 1.6 million.

This. I haven’t made any claims as to WHY this occurs. There are various explanations.

It could be a poor RNG, it could be different tiers drop different types of rings at different rates, it could be that daily resets influence drops if you did a run before reset, it could be some rings have a higher chance at any tier.

All my data show is that only once in 1.6 million would you get the number of doubles I got IF any time the game draws from the “ring” bag, it draws randomly among the 26 possibilities.

The reason why I posted this in the first place is because so many people are like, “oh my gosh I got 3 spearguns in a row,” or “dude I got the same weapon box 5x in a row this week.”

I presented my data as I have it, and 84 rings is a large sample size, with high power, despite what people keep claiming here.

Nevets Crimsonwing.5271:

Tim.6450:

Well what we were trying to (dis)prove (I’m not sure whether this thread is on track anymore) was that the rng for rings is not an independant. Meaning a ring’s chances does not change depending on the result before. Nevets tries this by placing a counter example. His example has 15 occurences where the drawn ring was the same as the previous ring drawn in a batch of 84 rings. The chance of this or worse (more “doubles”) occurring is 1 in 1.6 million.

This. I haven’t made any claims as to WHY this occurs. There are various explanations.

It could be a poor RNG, it could be different tiers drop different types of rings at different rates, it could be that daily resets influence drops if you did a run before reset, it could be some rings have a higher chance at any tier.

All my data show is that only once in 1.6 million would you get the number of doubles I got IF any time the game draws from the “ring” bag, it draws randomly among the 26 possibilities.

The reason why I posted this in the first place is because so many people are like, “oh my gosh I got 3 spearguns in a row,” or “dude I got the same weapon box 5x in a row this week.”

I presented my data as I have it, and 84 rings is a large sample size, with high power, despite what people keep claiming here.

Just so we are clear, by “double” you don’t simply mean duplicate. so not just drawing a ring you already have – but furthermore you mean the exact ring you just got on your previous fractal run. I don’t think that is clear. Also, this does not account for the entire loot table. in between getting two of the same rings, you could’ve gotten a fractal skin, armor chest, weapon chest, absolutely nothing, a different quality ascended ring (ascended/not ascended).

And then to accept what you are saying you have to assume that Tim’s formula is accurate for this scenario, that the math is correct, and that your data is correct. i’m terrible at math so i’d question the 3rd: you really put your rings in the bank in the order that you received them and haven’t touched them since? have you ever used the rings you got from drops for your character? on your alts? are those all accounted for? edit: and did you ever buy rings? such as when you started fotm in the very beginning to get AR? did you include those in the 84?

Benson.9063:

Just so we are clear, by “double” you don’t simply mean duplicate. so not just drawing a ring you already have – but furthermore you mean the exact ring you just got on your previous fractal run. I don’t think that is clear. Also, this does not account for the entire loot table. in between getting two of the same rings, you could’ve gotten a fractal skin, armor chest, weapon chest, absolutely nothing, a different quality ascended ring (ascended/not ascended).

And then to accept what you are saying you have to assume that Tim’s formula is accurate for this scenario, that the math is correct, and that your data is correct. i’m terrible at math so i’d question the 3rd: you really put your rings in the bank in the order that you received them and haven’t touched them since? have you ever used the rings you got from drops for your character? on your alts? are those all accounted for?

Yes, I mean drawing the same ring as the last time I drew a ring. The intermediate drops shouldn’t matter if each “ring bag” draw is independent. That’s really the assumption I was testing – is each ring draw independent from the last.

Yes, I really put my rings in the bank in the order I received them and haven’t touched them since. I had a number of rings which I used on my characters – but this was before I started keeping track. Once I decided to start collecting data, I put in ring after ring, right after the previous one, so that’s the data you see.

Edit: The other thing to note is in the first tab in the middle you see a rainbow looking ring which can no longer drop since the number of rings possible decreased from 34 to 26. So some of that earlier data DEFINITELY had a different probability than the latter data.

But look at the last tab in particular. 24 drops, 23 trials, 5 repeats. That’s 21%, which given the smaller sample size is probably not statistically different than the 18% which occurred overall. But it’s DEFINITELY bigger than the ~4% we’d expect. (Might not be statistically different)

Nevets Crimsonwing.5271:

The intermediate drops shouldn’t matter if each “ring bag” draw is independent. That’s really the assumption I was testing – is each ring draw independent from the last.

sounds like you are assuming that which you are trying to prove.

I don’t know why i’m being contrarian. who knows,maybe you’re on to something. If you’re right then good – figure out the pattern and we’ll take advantage of it.

Benson.9063:

Nevets Crimsonwing.5271:

The intermediate drops shouldn’t matter if each “ring bag” draw is independent. That’s really the assumption I was testing – is each ring draw independent from the last.

sounds like you are assuming that which you are trying to prove.

I don’t know why i’m being contrarian. who knows,maybe you’re on to something. If you’re right then good – figure out the pattern and we’ll take advantage of it.

But the data tend to work AGAINST the assumptions. The idea is that 2 step process which people assumed drops work:

1. Reward = ring
2. Ring = X

Might not be the case.

The assumptions only hold if the data aren’t different from what would be expected. Since the data seem to be different from expected, that means the drops might NOT be independent, which is mildly interesting.

Nevets Crimsonwing.5271:

you can’t say that doubles are anymore likely than singles.

You haven’t run a sample size estimator and have provided no numbers here to make that claim.

So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

Yes, if you wanted to test the claim that different levels have different drop rates.

I don’t see how doing different levels that may very well have different drop rates of rings will prove that doubles are more likely.

Ring X may have a higher than average drop rate on Fractal Level A.

You do Fractal Level B first and get Ring X. You then do Fractal Level A and get Ring X. Is it because of the higher drop rate or is it because the RNG isn’t independent of the previous ring drop? And even if it isn’t independent, how do you know that it wasn’t the higher chance that caused it? Because it’s not 100% chance for doubles so there’s obviously something ELSE affecting the results. How do you that without the increased range that bias for Ring X would have been enough to get Ring X again?

Y = RNG

Where Y is the number of the RNG generator and S is the current seed.

X = (Y + X-1)/2

Where X is the number that they compare to the loot table, Y is the previous Y and Y-1 is the previous trial’s result. Rounded down.

Trial 1 – Running Fractal Level B
Y = RNG
Y = 10

X = (10+0)/2
X = 5

Trial 2 – Running Fractal Level A
Y = RNG
Y = 7

X = (9+5)/2
X = 7

Fractal Level B has a range of 1-5 for Ring X.
Fractal Level A has a range of 1-7 for Ring X.

Had you run the second trial on Fractal Level B, you would not have gotten a double. But since you ran on Fractal Level A, you got a double. But without the bias you wouldn’t have gotten Ring X at all.

See how differences in the loot table can affect whether you get a double or not?

You haven’t run enough trials on the SAME loot table to say with high accuracy that the number of doubles is due to a bias rather than a range difference on the loot table.

Enough numbers in my post this time?

Seera.5916:

Nevets Crimsonwing.5271:

you can’t say that doubles are anymore likely than singles.

You haven’t run a sample size estimator and have provided no numbers here to make that claim.

So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

Yes, if you wanted to test the claim that different levels have different drop rates.

I don’t see how doing different levels that may very well have different drop rates of rings will prove that doubles are more likely.

Ring X may have a higher than average drop rate on Fractal Level A.

You do Fractal Level B first and get Ring X. You then do Fractal Level A and get Ring X. Is it because of the higher drop rate or is it because the RNG isn’t independent of the previous ring drop? And even if it isn’t independent, how do you know that it wasn’t the higher chance that caused it? Because it’s not 100% chance for doubles so there’s obviously something ELSE affecting the results. How do you that without the increased range that bias for Ring X would have been enough to get Ring X again?

Y = RNG

Where Y is the number of the RNG generator and S is the current seed.

X = (Y + X-1)/2

Where X is the number that they compare to the loot table, Y is the previous Y and Y-1 is the previous trial’s result. Rounded down.

Trial 1 – Running Fractal Level B
Y = RNG
Y = 10

X = (10+0)/2
X = 5

Trial 2 – Running Fractal Level A
Y = RNG
Y = 7

X = (9+5)/2
X = 7

Fractal Level B has a range of 1-5 for Ring X.
Fractal Level A has a range of 1-7 for Ring X.

Had you run the second trial on Fractal Level B, you would not have gotten a double. But since you ran on Fractal Level A, you got a double. But without the bias you wouldn’t have gotten Ring X at all.

See how differences in the loot table can affect whether you get a double or not?

You haven’t run enough trials on the SAME loot table to say with high accuracy that the number of doubles is due to a bias rather than a range difference on the loot table.

Enough numbers in my post this time?

No, I see no sample size estimate in there at all.

But in either case I already admitted that different levels could affect drop rate of given rings as one possible explanation for so many doubles.

Nevets Crimsonwing.5271:

Seera.5916:

Nevets Crimsonwing.5271:

you can’t say that doubles are anymore likely than singles.

You haven’t run a sample size estimator and have provided no numbers here to make that claim.

So wouldn’t experiments to show that doubles are more likely have to be done on the same level?

Yes, if you wanted to test the claim that different levels have different drop rates.

I don’t see how doing different levels that may very well have different drop rates of rings will prove that doubles are more likely.

Ring X may have a higher than average drop rate on Fractal Level A.

You do Fractal Level B first and get Ring X. You then do Fractal Level A and get Ring X. Is it because of the higher drop rate or is it because the RNG isn’t independent of the previous ring drop? And even if it isn’t independent, how do you know that it wasn’t the higher chance that caused it? Because it’s not 100% chance for doubles so there’s obviously something ELSE affecting the results. How do you that without the increased range that bias for Ring X would have been enough to get Ring X again?

Y = RNG

Where Y is the number of the RNG generator and S is the current seed.

X = (Y + X-1)/2

Where X is the number that they compare to the loot table, Y is the previous Y and Y-1 is the previous trial’s result. Rounded down.

Trial 1 – Running Fractal Level B
Y = RNG
Y = 10

X = (10+0)/2
X = 5

Trial 2 – Running Fractal Level A
Y = RNG
Y = 7

X = (9+5)/2
X = 7

Fractal Level B has a range of 1-5 for Ring X.
Fractal Level A has a range of 1-7 for Ring X.

Had you run the second trial on Fractal Level B, you would not have gotten a double. But since you ran on Fractal Level A, you got a double. But without the bias you wouldn’t have gotten Ring X at all.

See how differences in the loot table can affect whether you get a double or not?

You haven’t run enough trials on the SAME loot table to say with high accuracy that the number of doubles is due to a bias rather than a range difference on the loot table.

Enough numbers in my post this time?

No, I see no sample size estimate in there at all.

But in either case I already admitted that different levels could affect drop rate of given rings as one possible explanation for so many doubles.

I could come up with 100 different examples to show the same thing. For brevity’s sake I showed only one as only one was necessary to demonstrate my point.

All your data shows, even if you had more samples with the same distribution of doubles to not doubles is that there is something odd with the system. It doesn’t show what it is.

There could be a bias in the system or it could be due to how the loot tables are set up across the different levels. Or it could be both.

And honestly, if I was designing an experiment or doing a case study to determine if there is a problem and I had a hypothesis of what might be causing it, I would design my experiment or case study to account for any variables possible. But that’s just me. And you seem to think it’s due to bias. But your data doesn’t prove it (it doesn’t disprove it either, I’m not that stupid to say that if it doesn’t prove it it disproves it). You haven’t eliminated all other possible reasons that you’re getting doubles. Nor really established that you just happen to be an outlier.

I would offer up my own results, but I haven’t run that many trials so my data set doesn’t prove or disprove anything.

As for why I’m not running simulations of my own, I’m not a super smart stat person so I wouldn’t know what a good stat simulator would be. And I don’t want to grab one that’s made by someone who just thinks they know stats. If you can give a link to a good one, I’ll be more than happy to start posting more numbers.

Seera.5916:

Nevets Crimsonwing.5271:

No, I see no sample size estimate in there at all.

But in either case I already admitted that different levels could affect drop rate of given rings as one possible explanation for so many doubles.

All your data shows, even if you had more samples with the same distribution of doubles to not doubles is that there is something odd with the system. It doesn’t show what it is.

As for why I’m not running simulations of my own, I’m not a super smart stat person so I wouldn’t know what a good stat simulator would be. And I don’t want to grab one that’s made by someone who just thinks they know stats. If you can give a link to a good one, I’ll be more than happy to start posting more numbers.

All your data shows, even if you had more samples with the same distribution of doubles to not doubles is that there is something odd with the system. It doesn’t show what it is.

Yes, I completely agree. Yay!

As for why I’m not running simulations of my own, I’m not a super smart stat person so I wouldn’t know what a good stat simulator would be. And I don’t want to grab one that’s made by someone who just thinks they know stats. If you can give a link to a good one, I’ll be more than happy to start posting more numbers.

I have a purchased copy of SigmaPlot. Otherwise I linked the StatTrek binomial probability calculator on this thread somewhere I think. But what you want is a sample size estimator which estimates how many trials you’ll need to see a statistically significant difference between two groups at an expected alpha, power level, and difference between groups. When I ran it earlier it was fewer than 80 trials, which is what convinced me I had enough drops to be confident in my result.

Nevets Crimsonwing.5271:

I have a purchased copy of SigmaPlot. Otherwise I linked the StatTrek binomial probability calculator on this thread somewhere I think. But what you want is a sample size estimator which estimates how many trials you’ll need to see a statistically significant difference between two groups at an expected alpha, power level, and difference between groups. When I ran it earlier it was fewer than 80 trials, which is what convinced me I had enough drops to be confident in my result.

See what I mean by not knowing which program to get? Does that StatTrek one have a sample size estimator? And simple enough that someone who hasn’t had a stats course in almost 10 years can figure out?

And when you ran it earlier, what were your settings? Would the settings be equivalent to 66%, 95%, or 99.99% level of confidence (exact percentages may be slightly off)?

Pity there isn’t a simulator that can access ANet’s RNG system and loot tables to simulate 100+ runs (to account for no ring runs). It would make testing hypotheses so much easier.

Seera.5916:

Nevets Crimsonwing.5271:

I have a purchased copy of SigmaPlot. Otherwise I linked the StatTrek binomial probability calculator on this thread somewhere I think. But what you want is a sample size estimator which estimates how many trials you’ll need to see a statistically significant difference between two groups at an expected alpha, power level, and difference between groups. When I ran it earlier it was fewer than 80 trials, which is what convinced me I had enough drops to be confident in my result.

See what I mean by not knowing which program to get? Does that StatTrek one have a sample size estimator? And simple enough that someone who hasn’t had a stats course in almost 10 years can figure out?

And when you ran it earlier, what were your settings? Would the settings be equivalent to 66%, 95%, or 99.99% level of confidence (exact percentages may be slightly off)?

Pity there isn’t a simulator that can access ANet’s RNG system and loot tables to simulate 100+ runs (to account for no ring runs). It would make testing hypotheses so much easier.

I don’t know the full range of StatTrek’s web resources – they’re free though!

SigmaPlot is kind of the industry standard besides Excel, but it does cost money.

My sample size estimator was run with an alpha of .05 (95% confidence) and a power of .8 (beta of .2).

Those are the “go-to” standards for most research.