Don’t know if it’s bugged for some or it’s just random bad luck. As an example I just emptied a lot of caches gathered yesterday, so I got a lot of blood shards.
Gambling for up to 30 pieces at a time Kadala one time at least gave 8 legendaries, several times 6 and 4, a some times less than that and few times zero.
I spent a lot of shards, and many times and the legendaries she gave me were the same as before. I have not noticed that she has given me more legendaries after doubling the chance
This is another fine example why the impenetrable multilayer RNG wall is such a dreadful design. The entire concept is unworkable. To give the away kingdom to one player and an empty bucket to another on a random basis is complete garbage.
But hey do the same thing in Diablo IV and expect different results.
So apart from those “many times” were there times where you got more than before?
No. For example, lots of times rolled for rings, and with the inventory filled with rings from Kadala, only between 2 and 4 were legendaries. And before doubling the chance, it was the same…
Yeah, rings feels bugged. Buying Quivers gives 20ish legendaries on average for a dump, rings 4 tops.
1 Like
It actually is.
A sample size of 750 is plenty to get a reasonable margin of error on a 20% chance.
This is actually a very common misconception about statistics, a proper random sample* of ~1000 is sufficient to get reasonable bounds on most things. People always look at poll results and say “wait they only sampled 1000 people, how can that possibly be enough to understand the population?” And while there can often be issues with polling data (e.g. political polls) those are almost never because of the sample size, but instead because getting a truly random sample of a human population is extraordinarily difficult. IF you can get a truly random sample of ~1000, it will give very reasonable bounds on most things.
Assuming it’s not some super rare event anyway. A sample size of 1000 will be insufficient for very low probability events, e.g. to estimate the rate of a very rare disease in a population, but for anything with an intermediate chance, a sample size of 500-1000 is sufficient.
But we’re not talking about super rare events here, we are trying to estimate something with an expected probability of 20%.
Do you know what the probability of getting 70 or fewer legendaries on 751 attempts would be, if the true chance were 20%? Hint: it’s astronomically small.
sum from x=0 to 70 (751 choose x)*0.2^x*(1-0.2)^(751-x) = 1.195 x 10^-15
That’s about 1 in 836 trillion.
It’s also very telling that the empirical probability is much closer to 10% instead of 20%. Not some other random value like 15% or 17%, but around 10%. Almost like, I don’t know, some item slots are bugged and only yielding 10% at Kadala instead of 20%.
17 Likes
Yeah I noticed the same and compiled some data on rings and it seems like it’s more like 10% as well.
I got 52 legendaries on 500 gambled rings, the probability of getting 52 or fewer legendaries on 500 attempts, if the true probability were 20%, is about 1 in 178 million.
Many other slots (e.g. sources) appear to be working fine, I get ~20% there.
But it does seem like a few slots may be bugged and still only give 10%.
11 Likes
I spent 1200-1300 shards for rings. I thought I might as well try to gamble before crafting and upgrading rings to get Krysbin’s Sentence for necro. Only two were legendary: LoN ring and primal ancient Krysbin’s Sentence.
You are forgetting one major factor -
It is one person. The sample size is 1. Now if 30 people all did the same experiment with 750 attempts, we would have some accurate data.
No, it’s not.
Not unless you think the true Kadala gamble chance varies between players, anyway.
There’s no reason to use a nested/stratified random sample (x players with y gambles) here unless you have good reason to think the population parameter (true Kadala gamble chance) varies between players.
For a (pseodo)-random sample from a Bernoulli process, there’s no difference between taking a sample of x*y gambles from a single player vs. x players doing y gambles, assuming gambles are independent between players. They are equivalent as long as the Kadala legendary gamble probability is constant between players. It almost certainly is.
10 Likes
yes, 751 trials is a very good sample size. We generally shoot for 10% of the population, and in cases where the source population is very large, we tend to cap the sample at 1000. Very few research papers even reach 200 participants / trials, let alone 1000.
Do you have a tentative list on what slots are bugged?
I’ve gotten more primal set pieces this season than ever before, and 3 primal sources. But I have not gambled for jewelry at all.
Me I say it’s just bad luck, and why because I’ve done over 150 gr 94’s and got 0 primal ancient items! Would love to get the last 3 go I can get this altar done.
No this isn’t accurate. 1 and 30 are essentially the same number when youre results are 1 in 836 trillion.
If the odds are truly 20%, and you have every person on earth rolling 750 attempts every 1 second, for every minute of every day for 4 years. You’d expect to have ONE event with 70 or less legos.
I kept track this afternoon while running speed 105s of my Kadala items. Here is my list of legendary&set items / total gambles (wizard helm):
10/48
7/40
11/50
12/57
11/50
11/50
10/50
9/62
12/50
5/25
11/50
7/50
10/50
9/50
5/25
10/50
11/50
4/24
165 / 831 = 19.8% legendary/set item rate
1 Like
Awesome. Please let us know if you try the same experiment with boots. Next time I play I’ll flip/flop between boots and helms to see if there is a pattern between the two. I haven’t gambled much besides boots so far this seasons.
1 Like
Well, I decided to do a quick experiment, comparing Seasonal drops with non-seasonal drops from Kadala.
10 runs in season (full 30 items, filling the inventory) = 64 out of 300 items (21%)
Only completed 6 runs in nonseason (because I got bored) = 14 out of 180 (8%)
So it seems like the double the chance of a legendary from Kadala is working for me.
1 Like
Yea, rings feels bugged.
Agree with this. Once I unlocked double legendaries, I was mostly gambling for rings. I stopped doing so when on the occasion I wouldn’t gamble for rings, the drop rate for gloves/belts did feel like it was doubled, but rings felt the same. It seems to be standard.
As an addendum to my last post, I wanted to confirm my strong suspicions on gambled ring drop rates not being 20%. In addition to the OP, I’ll add my data:
Total shards gambled on rings: 42,000.
42,000 / 50 = 840 gambles.
840 gambles @10% drop rate = 84 drops.
840 gambles @20% drop rate = 168 drops.
What do you think I got?
Results: 83 drops, or a ~9.88% drop rate.
That’s almost perfectly in-line with the normal drop rate. So please, go ahead and tell me my “sample size isn’t big enough” when this is clearly enough to show that rings being gambled (at least on my character) are being gambled at the standard rate. Some statistic guru can tell me the odds of this data being merely being “unlikely” at what should be a 20% drop rate.
To be perfectly clear, this character (wiz) is not the character which unlocked/completed the altar. However, the buffed gamble rate clearly works for every other piece of armour (with the possible exception of boots, which again simply do not seem to drop at the same rate).
This clearly isn’t a random aberration. The fact that the sample size is decent and that it just so happens to match the standard drop rate is too hard to ignore. It’s annoying when the “It’s just RnGGGGG” bros glibly respond to every post like this. Having played a game for thousands of hours, you get a feel for what is ‘normal’ and what isn’t.
Raw data:
Shards gambled >> number of drops
2000 = 4
2000 = 3
2000 = 3
2000 = 7
2000 = 5
2000 = 4
2000 = 2
2000 = 6
2000 = 2
2000 = 4
2000 = 6
2000 = 4
2000 = 3
2000 = 3
2000 = 3
2000 = 2
2000 = 6
2000 = 4
2000 = 3
2000 = 6
2000 = 3
At 20%, the average number of drops should have been around 8. You’ll notice that I never once hit that number.
4 Likes