Drop Rate Analysis and Treasure Hunter

[pchan]

Comments about sample size.

Let's say you kill a mob X that drops Y. If the droprate is Z and the sample size is N, then the error is as follows :

with 95% chance, +/- 1.64*s/sqrt(N)
with 90% chance, +/- 1.28*s/sqrt(N)
With 80% chance, +/- 0.84*s/sqrt(N)

And s is the square root of the variance of your sample. For example if you test the drop rate of one item (and only one), you just record 0s and 1s,

s= sqrt(Z*(1-Z))

(sorry for the maths)

example :

droprate around 50%, therefore s is aprroximately 1/2

error(95%)<1% only if N=6700 kills or greater
error(80%)<1% only if N=1700 kills or greater


droprate around 80%, therefore s is approximately 0.4

error(95%)<1% only if N=4300 kills or greater
error(80%)<1% only if N=1100 kills or greater


droprate around 10%, therefore s is around 0.3

error(95%)<1% only if N=2400 kills or greater
error(80%)<1% only if N=635 kill or greater

Killing 130 mobs and making conclusion that TH is +/- 1% doesn't make sense.

[Vail]

for those who don't have TH2, and wanna see some proof that it is in fact a separate trait

(note: both traits have the same help text description: "Increases chance of getting treasure."

[Enedin]

I have a buttload of data waiting to be posted, and I didn't like the data! Although if I had more time, I'd test more.

Gigantobugard : lizard. Outside area.
Stirge : bird. Dungeon.
Tiger : beast. Outside area.

Add in these results, and it might show that TH is different in dungeons/outside areas, and not according to family, but that is WAAY too early to tell:

http://users.telenet.be/jandebruyn/THTEST1.png

Also, I didn't want to test it anymore, since 150 with such a high drop rate is enough for me, but if you want more tests, I can kill a lot of Stirges in a short amount of time, place is full of em. Anyway, FULL MOON IS A LIE!:

http://users.telenet.be/jandebruyn/THTEST2.png

Aaaand here's more testing for ya. I believe that we can all assume that higher level monsters drop more items, just go kill some raptors outside Jeuno, and then kill some in Kuftal Tunnel. Same with Lizards.

http://users.telenet.be/jandebruyn/THTEST3.png

I'll be posting more.

EDIT: I stopped testing Lizards in YuhJungle, since they can drop more than one item, and that skewed testing. It's still up there, but interpret it with care.

Also I'm wondering what mob to do next. I need something in a dungeon not belonging to the bird, beast or lizard family, that is easy enough to kill, that there's plenty of and has a high enough drop rate. /sigh

[Shalafi]

for those who don't have TH2, and wanna see some proof that it is in fact a separate trait

(note: both traits have the same help text description: "Increases chance of getting treasure."

First impression would be that it was forgot in development.

But to explore further...

When you look at kick attacks, the equipment that has '"Kick Attacks" +' is said to increase the activation rate of kick attacks. How this applies to TH? I don't know.




Would be an interesting question to ask in the next interview.

"Why are TH & TH2 listed separately on the Job Traits list?"

Then depending on the answer.

"How then does '"Treasure Hunter" +' affect the job trait(s); does it affect only Treasure Hunter or both?"

[Spider-Dan]

The explanation for the separate TH2 trait is not that complex, I think:

Given the playerbase's general superstitiousness and paranoia, it seems quite reasonable that at some point, Squenix added TH2 as a separate trait just to prove beyond suspicion that yes, it is an actual trait.

Unlike, say, Resist Slow trait tiers, the confirmation of TH tiers is quite powerfully relevant to the game in general (and THF in particular).

[Jeryhn]

Given the playerbase's general superstitiousness and paranoia, it seems quite reasonable that at some point, Squenix added TH2 as a separate trait just to prove beyond suspicion that yes, it is an actual trait.

You do realize we're talking about SE here, creators of Almighty Apkallu, Parade Gorget, Absolute Virtue and Sage Sundi... right?

[Kimiko]

Are we also taking into consideration the use of thief as a main job vs. thief as a subjob in these findings?

[Zellgadiss]

The explanation for the separate TH2 trait is not that complex, I think:

Given the playerbase's general superstitiousness and paranoia, it seems quite reasonable that at some point, Squenix added TH2 as a separate trait just to prove beyond suspicion that yes, it is an actual trait.

Unlike, say, Resist Slow trait tiers, the confirmation of TH tiers is quite powerfully relevant to the game in general (and THF in particular).

If only they released an item that said: "Enchances: Treasure Hunter II" it'd be so much easier to test. Maybe they could have not been a-holes and called it Treasure Finder or something like that.

I've found that while some things may seem random and arbitrary in this game, they're not.

TH and TH2 affecting two different things sounds like a good theory to me.

[kwontess]

I'm fairly certain all TH1 data is /thf. All TH2 data of course is thief main. I don't know of any job traits that behave differently as a subjob. Its possible there is an effect on it, but it may be more difficult using thf main below 45 to do these tests in an effective manner. Moreover the most important thing to discover is TH2/+1/+2 and TH0. How do these relate?

Edit: Although it is a good question do TH1 and TH2 do completely different things? I doubt /thf is going to change this versus thf/ at a level less than 45. When would anyone have TH1 realistically that isn't /thf?

On another note, we need some more people testing. This requires too many kills for just 2-3 people to be contributing data. Data on other mobs is nice to see if it fits the pattern at the end, but if anyone is doing any additional testing it would be great if they could follow the outline Mojo set up.

I should have 1000 TH2 crawlers finished in a couple hours if I don't die of boredom first.

[Cutriss]

I don't know of any job traits that behave differently as a subjob.

As it happens, Sneak Attack and Trick Attack do exactly that.

[The Blackrose]

As it happens, Sneak Attack and Trick Attack do exactly that.

Traits, not Abilities.

[fndragon]

Are we also taking into consideration the use of thief as a main job vs. thief as a subjob in these findings?

You think there's a difference in job trait abilities between being a THF37 and /THF37? Is there any other job trait that differs when it is a subjob?

[Uubu]

All this is interesting data and I applaud you guys for the effort. But we know TH increases drop rates already, but I haven't seen any discussion on how TH works in party/alliance etc. This would be much harder to test but could actually be applied to the end game community><. What I am trying to get at is if it matter if thfs are in the party with the kill shot or just need to be in alliance or just need to have some form of hate on an enemy to gain the benefits of TH? Does multiple TH in a party/alliance have no benefit or diminishing returns? Maybe we could use all this data as a baseline but the latter is what I would really be interested in.

[Cutriss]

Traits, not Abilities.

As soon as I loaded this thread to read the replies, I realized I had done that. Oops.

[kwontess]

Uubu,

If we believe them SE already answered this. Once the thf is placed on the hate list for the mob TH is on it.

This is actually not that hard to test and verify, it could be done in a few hours. You can do an alliance with 2 chars, have the thf in one pt and the other pt have some misc job, thief takes first hit and you let the other job finish off the mob. Once we have base TH rates established for one mob very firmly something like this takes almost no time to test. For instance we see that crawlers are ~10% drop rate w/o TH, so go out and do 200-300 kills in this manner (thf can just pull with ranged) and you should be able to establish one way or another.

Thats whats nice about established baselines for a collection of mobs, then testing any other theories about TH superstition becomes a lot easier.

[Mojo]

Yeah, I've studied Shulula's data many times. When the testing first came out, I looked over it a lot and I replicated his tests in my own fashion on the same monster. I didn't test for the same thing, I simply tested each level of Treasure Hunter assuming independant drpos, with sample sizes varrying from 200 to 300 for each level. (I wish I still had this data floating around somewhere... I might...)

I did not record how many "No drop" outcomes I ran into; however, the major flaw in coming to any based on Shulala's data conclusion is the sample size. It's small, as you admitted, and unfortunately far too small to draw any conclusions even with a difference of 24 drops.

I broke down her data into the individual drops, and came up with this:

TH      Drop 1      Drop 2      Drop 3      Drop 4   Ttl   Kills

 0      (0%) 0    (6.6%) 7  (14.2%) 15  (18.9%) 20    42    106
 1   (0.99%) 1    (5.9%) 6  (13.9%) 14  (37.6%) 38    59    101
 2      (3%) 3      (8%) 8    (13%) 13    (55%) 55    79    100
 3      (7%) 7      (9%) 9    (31%) 31    (41%) 41    88    100
 4      (2%) 2    (15%) 15    (26%) 26    (52%) 52    95    100

Hopefully that formats properly. Looking at this, it seems quite clear that the wildly varying results between TH2, TH2+1, and TH2+2 are caused by the extremely small sample. It's no surprise that TH2 and TH+2 ran into the same amount of no drop outcomes.

Don't forget that while 88 to 95 is an 8% increase, it's also being stacked four times. In a perfect world, if you have a monster with 4 different drops and increase the drop rate on each one by exactly 1%, your total increase in drops is going to be more than 1%. If we use TH2 -> TH2+1 as an example, it would go from 79 to 83 drops (adding 1 drop to each item), which is a 5% increase. If we use TH2+1 -> TH2+2 as an example, it would go from 88 to 92 drops, which is a 4.5% increase.

With this in mind, an 8% increase is not at all out of the realm of possibility (in fact, it's very closely in it).

I feel like I've rambled on, but my main point is that Shulala's testing is far too small in terms of sample size and inconclusive to even consider trying to draw a hypothesis from.

One thing a lot of people don't realize is that if you're testing for a difference of 1%, it doesn't matter whether you use an item that drops at a 1% base rate or an item that drops at a 50% base rate. You'll need the same sample size to draw an accurate conclusion.

I will admit that the sample size is small and that if you were trying to pin it down to an exact percentage you would need a much larger amount of data, but that simply isn't the case here. The closer some occurrence is to 50%, the less room for relative variance there is. The base drop rate on Shulula's test seemed to be around 1/3, close enough to 50% to be somewhat reliable, and followed Enedin's pattern of a 50% increase from TH1 and another 50% increase for TH2. The drops continued to rise with TH2+1 and TH2+2, but the base drop rate remained a constant. A 20% increase in total drops with a constant no drop rate is unlikely. From the data presented, the chance that it was just variance(I'm not denying that it could have been) is lower than the chance that it was not. Either way, that's the purpose of this thread, to record data and attempt to figure all of this out with a reasonable amount of accuracy.

I'm glad to see that so many people are interested, although one thing I do ask is that you don't hold the model I presented accountable to the percentage. It's just a model that fits all of the data within reason, although it is by no means that accurate. The reason the model does go down to a percentage(Treasure Hunter +1 items giving +1% to the base drop rate) is because divisortheory's 6600+ sample seemed to indicate that it did. Hopefully we can see if it holds true.

Anyways, any help with the testing method presented in the OP would be greatly appreciated. My PC broke last night(I fear the motherboard fried or the CPU chipped....), so I may be out of the running for a few days.

PS: Maybe I'm stupid but I have no idea what I'm even looking at with those charts Enedin. The way the data is organized is quizzical and unclear to me.

[Enedin]

It's quite easy to read once you know how to. Each zone is split into mob, each mob is split into every item I looked at.

http://users.telenet.be/jandebruyn/THTEST1.png

The Left-Top number is always the amount dropped, and the right-top number is always the amount killed, the Left-bottom number is the percentage dropped. The Right-down number is the mean amount of mobs you have to kill to get one drop. And all this is done for TH0, TH1 and TH2.

Regarding the other columns right next to it, the black figures are REAL, the RED and BLUE ones on the right (in big font) are APPROXIMATIONS.

So the (TH% increase) columns show how much every TH trait (TH1 = TH1, THII = TH2-TH1, TH2 = TH2) increases drop rate (calculated off TH0). The (TH1% increase) column shows how much TH1 improves drop rates as opposed to TH0, the (THII% increase) does the same for TH2-TH1, so NOT TAKING INTO ACCOUNT TH1. For total increase (all TH), need the (TH2% increase column)

So...

(TH1) = Treasure Hunter 1.
(THII) = Treasure Hunter 2 but NOT having Treasure Hunter 1. Impossible.
(TH2) = Treasure Hunter 2+1.

I want you to understand so you can use this data and this clearly shows that there's a difference between tigers/lizards in outdoor areas on one hand, and bees in dungeons on the other hand.

The former has 50/50/100 increments, the latter has 113/37/150 increments, and sample sizes for both bugards and stirges are large enough to give a high enough accurate result.

[Dawin]

I could be wrong about this since it's just based on observations of EXP party vs solo...

but aren't drops increased by having more players in party? If this is a possibility, you really should confirm that all the testing is done by one solo player.

[Cutriss]

but aren't drops increased by having more players in party? If this is a possibility, you really should confirm that all the testing is done by one solo player.

I highly doubt this is the case. In fact, my observation tends to be that the opposite is true. I would guess that the reason you feel this way is because Signet/Sanction/Sigil is adding in crystal drops, plus the occasional BS/KS from >=EP mobs, thus the loot pile always looks bigger.

[Genosync]

I highly doubt this is the case. In fact, my observation tends to be that the opposite is true. I would guess that the reason you feel this way is because Signet/Sanction/Sigil is adding in crystal drops, plus the occasional BS/KS from >=EP mobs, thus the loot pile always looks bigger.

Pretty sure that about 3 years ago we got information (from SE) that party size increases the possibility to get a drop, and I'm also pretty sure that from my own experiences over the course of playing FF, that it is definently true.

If you've seen the opposite, you're definently playing a game that's not FFXI. Sorry.