Drop Rate Analysis and Treasure Hunter

[Mojo]

Finally got my PC fixed...sorta. I'll be doing some more TH1 later today. Thank you to everyone who has contributed thus far, doing a great job.

Carrion Crow (Killed 100 times)
       10 bird egg                     [Drop Rate:  10.00 %]
       18 bird feather                 [Drop Rate:  18.00 %]

Carrion Crow (Killed 384 times)
       31 bird egg                     [Drop Rate:   8.07 %]
       66 bird feather                 [Drop Rate:  17.19 %]

Did you happen to write down the number of double drops you got for both of these data sets? It's paramount to the foundation of this test.

[zantes]

Did you happen to write down the number of double drops you got for both of these data sets? It's paramount to the foundation of this test.

I don't really think that double drops would cause a problem with it. On these mobs, we know that they can drop both items, so it seems logical that each has it's own base percentage. Also, the data that Motenten posted for the Carrion Crow would have had a 1.39% chance of having a double drop (.0807*.1719).

[Toksyuryel]

I think he meant double drops as in dropping the same item twice. But the other thing is important to record too, simply because we DON'T know for sure and all data is needed at this stage to determine what's really going on. "Don't think" and "seems logical" are two phrases that have absolutely no place in any of this.

[Motenten]

Did you happen to write down the number of double drops you got for both of these data sets? It's paramount to the foundation of this test.

I didn't write anything down; it's all parsed. Also, it's only one dataset. The second batch is inclusive of the first. Since it's all parsed, I can go in and query the number of double drops. Here:

  Dropped  0 items   296 times ( 77.08 %)
  Dropped  1 items    79 times ( 20.57 %)
  Dropped  2 items     9 times (  2.34 %)

Edit: And if you mean the number of times the same item dropped twice, that never happened. Double drops in this case means 1 egg and 1 feather.

[Mojo]

Well, the premise of the study is to determine exactly how drops and Treasure Hunter tie into one another. There has been evidence to indicate items could possibly be pooled together, so establishing a no drop rate is important. That's impossible without recording the number of double drops, and I would much rather not calculate them.

Edit: Thanks for the update.

[Motenten]

You have an error of +/- 3%ish with such a small sample size.

Yes, yes, I should know better than to try to extrapolate trend lines from such a small dataset. I was just mildly concerned that the data never went -above- the semi-expected value of a 10% drop rate, while I would expect pure random variance to at least occasionally go both above and below the 'true' value.

Of course the next 200 kills had a 15% drop rate before falling back into the 9%-11% range, so...

[kwontess]

I know I should probably be focusing on other mobs after finishing up crawlers, but honestly I find dhalmel most interesting because of what they can prove. Here are my first 2 sets of data for dhalmel with TH2. Each set is out of 100 kills, they are separated by commas.

Nothing: 53,54
meat: 15, 17
meat+hair: 5, 1
meat+femur: 1, 4
meat+femur+hair+hide: 1, 0
meat+grass: 1, 0
meatx2: 9, 7
meatx2+femur:1, 4
meatx2+hair: 1,4
hair: 10, 7
hair+hide: 1, 0
hide: 3,0
hide+femur: 0, 1

Totals:
meat: 97
hair: 30
femur: 12
hide: 6
grass: 1

Now for some statistics. I am going to have to make an initial assumption that both meats are equal probability. Also note that the sameple size is way too small (I'm going to compound any error in the calculations), but I still find it rather interesting how far off things are.

There are 48 possible collections of drops assuming they are independent. I only whitnessed 13/48, although many of them are expectedly very low probability.

Edit: Orginally I had said there were 96 pools, but it should be 2^5+2^4 = 48

Meat had a 97/200 expected return per kill. If there are 2 possible meats each with equal probability then each one has a 24.25% drop rate. This gives the following:

Meat1: 24.25%
Meat2: 24.25%
Hair: 15%
Femur: 6%
Hide: 3%
Grass: .5%

Now to calculate the probability of each of my pools. The first number will be the probability assuming they are independent and my numbers are correct, the second number will be what I actaully whitnessed:

Nothing: 44.25%/53.5%
Meat: 28.33%/16%
Meat+Hair: 5.00%/3%
Meat+Femur: 1.81%/2.5%
Meat+Femur+Hair+Hide: .01%/.5%
Meat+Grass: .14%/.5%
Meatx2: 4.53%/8%
Meatx2+Femur: .29%/2.5%
Meatx2+Hair: .80%/2.5%
Hair: 7.81%/8.5%
Hair+Hide: .24%/.5%
Hide: 1.37%/1.5%
Hide+Femur: .09%/.5%

Now these numbers don't add up to 100% and they shouldn't as there are many other theoretical treasure pools which I didn't encounter. The interesting numbers are those of the no drop and those of just 1 meat dropping. I should have expected nothing to drop 44.25% of the time but in fact nothing dropped 53.5% of the time. On the other hand if both meats were independent of eachother and have equal drop rate we should expect a 28.33% chance of getting exactly 1 meat but instead we see a 16%.

Again this sample size is smaller than it should be, I will continue to expand upon it. Due to the way probabilities are calculated errors are going to compound. However two of the numbers are drastically off what should be expected assuming both meats have an equal chance of dropping.

[Motenten]

Not sure I agree with your base numbers on the 2 meat drops.

You had 97 meat drops, but only 71 actual kills with meat dropping (the difference being instances of 2 meats dropping). With a 24.25% individual chance of dropping, you would expect to have 85 kills drop at least 1 meat rather than the 71 you got.

If each meat has an equal probability of dropping, the probability P of at least 1 meat dropping should be (2x - x^2), where x is the probability for each meat to drop. So to solve for x, we take P = 71/200 = .355, and x works out to 0.197. I'll round that off and just say 20%.


Then the individual item percents would be:

Meat1: 20%
Meat2: 20%
Hair: 15%
Femur: 6%
Hide: 3%
Grass: .5%


Then redoing your other set, the totals would be:

Nothing: 49.4%/53.5%
Meat: 24.7%/16%
Meat+Hair: 4.9%/3%
Meat+Femur: 1.8%/2.5%
Meat+Femur+Hair+Hide: .01%/.5%
Meat+Grass: .14%/.5%
Meatx2: 3.1%/8%
Meatx2+Femur: .2%/2.5%
Meatx2+Hair: .1%/2.5%
Hair: 8.7%/8.5%
Hair+Hide: .27%/.5%
Hide: 1.5%/1.5%
Hide+Femur: .1%/.5%


With this number set, most results are pretty close, except for the single- and double meat-only entries. Single is 8.7% below expected, and double is 4.9% above expected. Sort of balanced, there, actually.

[Dawin]

Now for some statistics. I am going to have to make an initial assumption that both meats are equal probability.

97/200 meat total. Gonna round to 50% to make the rest easier.

If it works like you suggest, that implies each meat has a 25% of dropping on its own (25% + 25% = 50%).

In your actual results, you got meat x 1 ~ 25% and meat x 2 ~ 12.5%.

If it worked like you suggested, we would expect to get meat a (.25)(.75). Meat B would be the same %, so we would get 1 meat (2)(1/4)(3/4) = 6/16 = 3/8 = 37.5%.

We would get 2 meat (1/4)(1/4) = 1/16 = 6.3% of the time.

Instead, your 2 results are much more closer in probability which implies they are not 2 separate drop rates, but instead somehow linked. Either that, or TH2 has a different effect than we have been assuming.

Of course, the sample size is still small...

[Motenten]

Alternative possibility is that Meat (not MeatA/MeatB) has a chance of dropping, and that Meat itself has a probability set that indicates the odds of 1/2/etc of that type to drop.

If that's the case, then current results indicate a 35% chance of meat dropping, and a 65/35 split of getting 1 or 2 drops of that type.


Working out the probability sets if that's the case:

Nothing: 50.1%/53.5%
Meat: 27%/24%
Meat+Hair: 4.8%/5.5%
Meat+Femur: 1.7%/5%
Meat+Grass: .14%/.5%
Meat+Femur+Hair+Hide: .01%/.5%
Hair: 8.8%/8.5%
Hair+Hide: .27%/.5%
Hide: 1.6%/1.5%
Hide+Femur: .1%/.5%


That's significantly closer to what was observed. Further testing should indicate which of the two models seems to be closer in the long term.

[kwontess]

Not sure I agree with your base numbers on the 2 meat drops.

You had 97 meat drops, but only 71 actual kills with meat dropping (the difference being instances of 2 meats dropping). With a 24.25% individual chance of dropping, you would expect to have 85 kills drop at least 1 meat rather than the 71 you got.

If each meat has an equal probability of dropping, the probability P of at least 1 meat dropping should be (2x - x^2), where x is the probability for each meat to drop. So to solve for x, we take P = 71/200 = .355, and x works out to 0.197. I'll round that off and just say 20%.

There are multiple ways to go about this. But I am using a basic fact from statistics that the expected value will be twice the probability of each drop. What do mean by this?

Let x be the probability for each meat dropping. Then the number of drops per kill we can expect is as follows (assuming indepdence):

Exactly two meats drop: x^2
but if two meats drop x^2 of the time then we should expect 2x^2 total meats from these cases

Exactly one meat drops: x*(1-x)+(1-x)*x
which means we end up with exactly one meat 2x-2x^2 of the time.

Therefore the expected number of meats per kill is 2x-2x^2+2x^2 = 2x

I ended up with .485 meats per kill. Thus x = .2425

There are multiple ways to go about computing what x should be. I have looked at the expected number per kill because the math is easiest in that case.

If we were to use your numbers you would say that meat drops at a rate of 20% which means I would average .4 meats per kill instead of .485. That makes my meats/kill number substantially off (over 20% error on the largest of all the samples as total meats gives us the largest sample size).

If the data is truly independent and we have the numbers worked out any method we pick (no meats, exactly 1 meat, 1 or more meats, exactly 2 meats, average meats per kill) should predict the same probability. However the numbers provided show that no matter which prediction model you use to fit the data one of our numbers is drastically off.

Now the sample size is too small. I am going to work on getting more, but as it stands there is no way to adequitely fit the data.

Edit: When I say whichever prediction model you use I mean assuming that the drops are independent. The point of this test is to see if they truly are independent. If they are not independent then we have something like the OP outlined or Motenten in the post before this one.

[Motenten]

Ah, I see. You calculated to reach a total of meats dropped, while I calculated to reach a total of kills with meat.

Yes, if the drops are independant, the two results should be close to the same. Since each method yielded different results, and both predicted single/double drop rates that were notably different from actuality, there seems to be a very high probability that the two meat drops, at least, are not independant.

As I do more test runs, would you like me to do TH2 on dhalmels to supplement your data? Or do TH0 on dhalmels for baseline, plus secondary confirmation of the distribution pattern?

[kwontess]

Hmm... Either would yield useful information. I'm somewhat curious how TH2 is affecting things, but in truth what I'm most curious about is whether drops are independent. So I think I would prefer a little more TH2 data. I've actaully been at my computer doing work (non ffxi related) a bit too much the last couple days and am starting to have some severe wrist pain, so I'm probably going to take the rest of tonight and tomorrow off. As soon as I'm feeling up to it I'll try and get a couple hundred more data points.

[shulula]

I'm really excited to see this discussion and data mining going on. When I started looking at Treasure Hunter, there was very little information besides the few random comments from developers. At that time there were many people who thought TH was just a placebo, even among THFs I talked to. I knew that TH would be hard to test because unlike a trait like Fast Cast, it doesn't affect the job but rather the mob you kill. And we know so very little about mobs directly aside from what it tells you when you /check it.

My goal wasn't to explain Treasure Hunter, but rather to give a framework for testing it. The question I posed was whether Treasure Hunter had an effect on drops. The null hypothesis was that it had no effect and that killing with TH0 would get the same results as killing with TH4. That would prove that it was a placebo effect.

I received many positive comments about the testing. The main criticism was the small sample size. I can see where someone would say that 100 kills is way too few kills to prove the drop rate of honey. Which is very true. But the question was if TH had an effect, and going from 42 (TH0) to 95 (TH4) drops over about 100 kills seems larger than the margin of error.

Of course more data is always better, which is why I'm excited. I see TH as one of the oldest mysteries in FFXI. We've all heard the superstitions of keeping a 4-leaf mandy bud, eating certain food or killing on a certain day/time.

[Motenten]

Long. Two different conclusions reached as writing progressed. Leaving all writing in for basis in reasoning.

Expanded detailing of the Loot plugin for KParser to help sort this out. Will release 1.1.5 shortly so others doing testing can make use of it.

Thf/Dnc, TH2, W.Sarutabaruta [S]

Savanna Dhalmel (Killed 200 times)
        7 dhalmel hide                 [Drop Rate (#):   3.50 %]  [Drop Rate (any):   2.50 %]  [Max #: 2]
       13 giant femur                  [Drop Rate (#):   6.50 %]  [Drop Rate (any):   6.50 %]  [Max #: 1]
       96 slice of dhalmel meat        [Drop Rate (#):  48.00 %]  [Drop Rate (any):  37.50 %]  [Max #: 2]
        1 sprig of papaka grass        [Drop Rate (#):   0.50 %]  [Drop Rate (any):   0.50 %]  [Max #: 1]
       38 tuft of dhalmel hair         [Drop Rate (#):  19.00 %]  [Drop Rate (any):  19.00 %]  [Max #: 1]


       Dropped  0 items    97 times ( 48.50 %)
       Dropped  1 items    66 times ( 33.00 %)
       Dropped  2 items    26 times ( 13.00 %)
       Dropped  3 items     7 times (  3.50 %)
       Dropped  4 items     4 times (  2.00 %)

Combined with Kwontess's:

hair: 30 + 38 = 68/400 = 17%
femur: 12 + 13 = 25/400 = 6.25%
grass: 1 + 1 = 2/400 = 0.5%

hide can drop 2
hide (# ratio): 6 + 7 = 13/400 = 3.25%
hide (drop rate): 6 + 5 = 11/400 = 2.75%

meat can drop 2
meat (# ratio): 97 + 96 = 193/400 = 48.25%
meat (drop rate): 71 + 75 = 146/400 = 36.5%


Nothing: 107 + 97 = 204/400 = 51%

If we take drop rates as:
hair: 17%
femur: 6.25%
grass: 0.5%
hide: 3%
meat: 36%

Chance of 0 drops is: 48%


Total drops of 1 meat: 45 + 54 = 99 / 146 = 68% of all drops with meat
Total drops of 2 meat: 26 + 21 = 47 / 146 = 32% of all drops with meat
Drop rate of 1 meat: 99 / 400 = 24.75%
Drop rate of 2 meat: 47 / 400 = 11.75%


Using Kwontess's approximation of the individual drop rates of meat if they are two independant items, my parse shows 24%, almost exactly the same as Kwontess's 24.25%. If they are independant and equal, then the odds of getting 2 meats to drop is 5.76%, significantly below observed (11.75%), and outside the margin of error for the sample size. If they are independant with unequal drop rates than the chance of 2 drops is even lower.

The issue of hides shows a similar pattern. Given the overall low drop rate, the chances of two hides dropping at once is miniscule (should see about once every 3250 kills). Two instances out of 400 seems to imply that it is likewise a single slot with the possibility of multiple items being dropped.


Thus it seems likely that each indidivual item a mob can drop is given its drop percentage, and that once it drops a second check is made for the chance to drop multiples of that item (in the case of dhalmel meat, approximately 2/3 chance of 1, 1/3 chance of 2).



However, we then get to the drop groupings:

    Drop set: Nothing
       Count:  97  [ 48.50 %]

    Drop set: slice of dhalmel meat
       Count:  41  [ 20.50 %]

    Drop set: tuft of dhalmel hair
       Count:  21  [ 10.50 %]

    Drop set: slice of dhalmel meat, slice of dhalmel meat
       Count:  14  [  7.00 %]

    Drop set: slice of dhalmel meat, tuft of dhalmel hair
       Count:   6  [  3.00 %]

    Drop set: giant femur, slice of dhalmel meat
       Count:   4  [  2.00 %]

    Drop set: giant femur, slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
       Count:   4  [  2.00 %]

    Drop set: slice of dhalmel meat, slice of dhalmel meat, tuft of dhalmel hair
       Count:   3  [  1.50 %]

    Drop set: giant femur, tuft of dhalmel hair
       Count:   2  [  1.00 %]

    Drop set: dhalmel hide
       Count:   2  [  1.00 %]

    Drop set: giant femur
       Count:   2  [  1.00 %]

    Drop set: dhalmel hide, slice of dhalmel meat, tuft of dhalmel hair
       Count:   1  [  0.50 %]

    Drop set: dhalmel hide, dhalmel hide, sprig of papaka grass
       Count:   1  [  0.50 %]

    Drop set: giant femur, slice of dhalmel meat, tuft of dhalmel hair
       Count:   1  [  0.50 %]

    Drop set: dhalmel hide, dhalmel hide, slice of dhalmel meat
       Count:   1  [  0.50 %]

Of interest, as it caught my attention a couple times, is the combination of femur, meat (1 or 2) and hair. There were 4 instances of that set with 2 meat, and 1 instance of that set with 1 meat. 5 instances out of 200 for that particular combination.

If the items have an independant drop rate as calculated above, the odds of that set occuring is approximately 0.4%, as opposed to the observed 2.5%. Basicly, the observed results seem abnormally high (though within the margin of error).

However, the above hypothesis regarding numbers of drops implies the possibility that it might be extended further. Namely, that there is first a check to see if -anything drops at all-, and then the calculations are done regarding individual drops.

If that's the case, the above numbers would be reworked to approximately:

50% chance of dropping something.

If dropping, then:
hair: 30 + 38 = 68/196 = 34.7%
femur: 12 + 13 = 25/196 = 12.8%
grass: 1 + 1 = 2/196 = 1.0%
hide: 6 + 5 = 11/196 = 5.6%
meat: 71 + 75 = 146/196 = 74.5%

Approximating out to:
hair: 35%
femur: 12.5%
grass: 1%
hide: 5%
meat: 75%


Odds of femur/meat/hair combo with those rates are 3% (and thus 1.5% overall). In other words, probable.


And guess what? Suddenly the multiple drop items don't need to be treated specially anymore. 75% chance of dropping meat (if something drops), with 193 drops out of 196 kills that have drops implies a base drop rate of 50%. A 50% drop rate for each of two independant items works out to 25% not dropping, 25% dropping 2, and 50% dropping 1; that gives the 1/3 and 2/3 split for dropping 2 meat vs dropping 1 that was seen earlier.


So I can see three things that TH could possibly affect:

Odds of a mob dropping anything at all. For example it could be that dhalmels normally have a 25% drop rate, 37.5% with TH1, and 50% with TH2. This needs to be tested on mobs that can drop multiple items (such as the dhalmels) to minimize the influence of no drops happening while inside the "something can drop" section. In the case of dhalmels (using the above new percentages), that's about 13% (which implies the actual overall drop rate, if such is the case, is possibly 30%/45%/60%).

Odds of each individual item dropping within the "chance to drop" section. It could very well be that this can't be influenced. Can validate by comparing TH0 vs TH2 on dhalmels to see if the ratio of items dropped remains constant even if the number of drops changes.

The number of slots that can be filled. Looking at things in different ways here has made me see how this could be applied (it didn't make sense as described before). In my and Kwontess's tests, neither of us ever received more than 4 items from a single kill. Thus it could be that once we're in the "chance to drop" section, the game only tests for items dropped until it fills the open slots, and any further checks that might have been made are discarded. One possible way this could be done is have a max of 2 slots with TH0, 3 slots with TH1, and 4 slots with TH2 (varies by mob, obviously). This should be fairly easy to test at different TH levels on the dhalmels.

Note that changing the max number of slots would have no effect on mobs like crawlers which can only ever drop 1 item anyway.

And of course do all those tests again with the TH+1 items.

[hezzel]

All testing done with:
THF75/NIN37

Main: p.harpe
Offhand: thief knife
Hands: Assassin's Armlets
Rest various gear that was never swapped.


Target: Carrion Crow #1

Killed: 100
Drops:
Bird Feather: 44
Bird Egg: 8

Egg Drops are:
2 x Egg+Feather
6 x Egg Only drop

Target: Carrion Crow #2

Killed: 100
Drops:
Bird Feather: 52
Bird Egg: 15

Egg drops are:
5 x Egg+Feather
10 Egg only drop


Target: Crawler

Killed: 100
Drops:
Silk Thread: 21

[Motenten]

A short sample to see if TH0 data seems to follow the same pattern as predicted.

Savanna Dhalmel (Killed 100 times)
        2 dhalmel hide                 [Drop Rate (#):   2.00 %]  [Drop Rate (any):   2.00 %]  [Max #: 1]
        3 giant femur                  [Drop Rate (#):   3.00 %]  [Drop Rate (any):   3.00 %]  [Max #: 1]
       13 slice of dhalmel meat        [Drop Rate (#):  13.00 %]  [Drop Rate (any):  11.00 %]  [Max #: 2]
        2 sprig of papaka grass        [Drop Rate (#):   2.00 %]  [Drop Rate (any):   2.00 %]  [Max #: 1]
        9 tuft of dhalmel hair         [Drop Rate (#):   9.00 %]  [Drop Rate (any):   9.00 %]  [Max #: 1]

       Dropped  0 items    77 times ( 77.00 %)
       Dropped  1 items    17 times ( 17.00 %)
       Dropped  2 items     6 times (  6.00 %)

    Drop set: Nothing
       Count:  77  [ 77.00 %]

    Drop set: slice of dhalmel meat
       Count:   8  [  8.00 %]

    Drop set: tuft of dhalmel hair
       Count:   6  [  6.00 %]

    Drop set: slice of dhalmel meat, slice of dhalmel meat
       Count:   2  [  2.00 %]

    Drop set: giant femur, tuft of dhalmel hair
       Count:   2  [  2.00 %]

    Drop set: dhalmel hide, sprig of papaka grass
       Count:   1  [  1.00 %]

    Drop set: dhalmel hide
       Count:   1  [  1.00 %]

    Drop set: slice of dhalmel meat, tuft of dhalmel hair
       Count:   1  [  1.00 %]

    Drop set: sprig of papaka grass
       Count:   1  [  1.00 %]

    Drop set: giant femur
       Count:   1  [  1.00 %]

Dropped 'something' 23/100, 23% of the time. Initial prediction was ~25%.

Of the 'something' drops, percentage breakdown is:

        w/TH0    w/TH2
hair:     39%      35%
femur:    13%    12.5%
grass:     9%       1%
hide:      9%       5%
meat:     57%      75%

Given the low sample size, the ratios match up very well.

In addition, I never got more than 2 items to drop at a time. Hardly conclusive, but circumstantially supportive of the 'slot' idea, especially if the drop rate percentage behaves as hypothesized.

Will do another short run later tonight with TH1 to see if things line up there as well. If so, then that will provide a theory that we can specifically test against.


@hezzel: Out of curiosity, can you do a run on savanna dhalmels with thief knife/armlets and see if you ever get more than 4 items to drop at a time? If you do, and if the above theory holds, then that would imply that the TH+1 effect adds to the slot count.

[kwontess]

Great data Motenten. I'm going to try and get 200 more with TH2 tonight or tomorrow. That should bring us up to 600. I would like to get a solid 1000 for each tier of TH.

The math seems good and it seems to go somewhat with the OPs original conjecture of determining whether things drop first then establishing what does. Even if the drops are indepdent this is very interesting.

Motenten can you actaully focus on the TH0 data? I'll continue on with the TH2 data and fill it out. Your TH0 sample size is very fascinating but at the same time too small to say much, in particular its hard to tell whether TH2 increases the drop rate of any particular item after its been established that something drops. With a couple hundred more tests we should be able to say with a large degree of certainty whether more than 2 items can drop with TH0.

Edit: I'm a little bit confused on the meat drop rate when you say 75%. Saying the odds of getting any meat doesn't seem useful. Lets for the moment assume they are indepdendent. Then according to my numbers I had 97 meats and 93 drop kills. If we are going with indepedence after the initial roll determining something drops then we would expect the probability to be 52.15%~50% for each meat, or are you still suggesting that the second meat has different odds?

Looking at your data you had 96 meats and 103 drop kills. If we combine our data we end up with 193 meats out of 196 drop kills which would make each meat 49.2% chance of dropping if indepdent.

I think thats a very realistic way to look at it. Each meat is indepdent after its established you're getting something. Then each meat has a 50% chance of dropping. This also predicts an average of .5 meats per kill in line with the rest of the data.

[Motenten]

I'm a little bit confused on the meat drop rate when you say 75%. Saying the odds of getting any meat doesn't seem useful.

It's useful in describing the distribution results. If the two meats each have a 50% chance of dropping, then 25% of the time you get nothing, 25% of the time you get 2, and 50% of the time you get 1 (of which 25% is meat1 and 25% is meat2). Thus overall, 75% of the drops will contain some quantity of meat, and it explains the 2:1 ratio of getting 1 vs 2 meat drops.

It's also a different way of getting the answer to what the individual drop rates are, and any explanation we come up with has to satisfy both approaches. In the original methodology, your formula resulted in a 24% drop rate and my formula resulted in a 20% drop rate. That obviously indicates that there was a flaw in the approach we were taking. Using my current hypothesis, both your formula and my formula result in predicting a 50% drop rate, which lends greater credibility to the approach.

are you still suggesting that the second meat has different odds

No. As long as the two meat drops are considered independant, I've always assumed that they had equal odds.

Motenten can you actaully focus on the TH0 data?

I want to do a quick 100 kills with TH1 to see if the data still seems to line up (at least roughly), but after that, yes, I'll focus on doing TH0 (which is also easier for me since I can kill about twice as fast with my mnk as with my thf).

[kwontess]

Sounds good. I agree this model definitely gives more consistent results. I'll get us at least another 300 tomorrow for TH2.