How to analyze loot

[falkao]

Originally Posted by Noodles

I'm going to take a walk for a moment, and see where it goes. Another way to look at this data is to make a plot of minimum loot versus HP (or better, lowest possible damage done to a mob, or something similar). This plot may be linear in some regions but is definitely not linear over all mobs. The instantaneous slope of this plot is then 1/(B*(1-empty rate)), where B is probably the 20 from your HP/20 figure. However, one doesn't need to know either B or the empty rate--the observed slope (and changes in the slope) can be due to either B or the empty rate (and changes in these values).

thats absolutely correct.

Originally Posted by Noodles

.. Some of the question are then: how does empty rate vary with mob? with HP? Is there some sort of smooth function that maps empty rate to HP? To which mobs might this function apply?

In the VU9.3 release notes, MA was talking about mob templates. So I think empty rate is also part of that template. There might be per HP class a suggested value, however, If you take Ambu (1010 HP) and Formidon (1100 HP), Ambu has no empties but Formidons do have one of 30%. So I think it's set manually as well. Since loot is empty corrected, it has no implication on return itself, but you need to kill more mobs to get the mean.

Originally Posted by Noodles

It also seems like we could in fact determine an upper limit for the empty rate of a mob with few kills, if we get "lucky". We would assume B=20 and the 1/(B*(1-empty rate)) functionality. A "lucky" loot would be a really low one. Then set

low loot = HP/(B*(1-empty rate))

(or HP replaced with damage done) and solve for the empty rate--for example, my Snable Male Guardian data (only 5 kills, but one small loot) might qualify for such an estimate. Anyway, maybe an interesting idea.

good idea but there are some limitations. If you fix B then the empty rate can be estimated. However, you still need a sufficiently large dataset to get a valid estimate. Take the case of 0 empty rate. If you observe a small loot then there might still exist a smaller one. Hence, if a large dataset is need, then empty rate can be estimated directly and so there is no need to use the formula. However, if only loot values are recorded and no empties, then this approach would give indeed some good estimates.

Originally Posted by Noodles

Your comments about the rounding (regular rounding versus truncation), may also reflect reality. If you look at the data in Post #402, regular rounding improves the match for the 30 hp and 70 hp mobs (round up instead of down). It may also imply that x.50 is rounded down (e.g. 20 hp mobs), and that to get minimum loot in such cases, no regeneration can occur. Regular rounding may or may not make sense from a MA point of view--it is more fair to the player, but truncation could also be a hidden source of "decay"--MA would always win. The relative amount won would be insignificant for large mobs, but nominally noticeable for smaller mobs.

Btw, how are globals given? In the chat there are only PEDs but there are some PECs as well.

I think, as you already mentioned, there are different types of rounding. If PED/PEC are displayed, there might be truncation (auction), for loot there might be other rules, like round business, or roundup till 1 PED and down when > 1 PED. I would have expected a bias in our formulas of about .5 PEC. Atm it’s only 0.05.

Originally Posted by Noodles

Finally, I'll include my (or Jimmy's) observed empty rates for the mobs with 79 kills or more:

Mob	  	# Kills	  HP	Empty Rate
sabakuma young	338	20	61%
sabakuma mature	260	40	58%
snable female y	79	40	53%
Plumatergus y	239	40	59%
sabakuma old	115	70	57%
combibo  young	87	80	51%

If I use the new empty rates, I do get a slope of .064 with mob HP, .061 with HP * 1.05 and .058 with HP *1.1. The bias (intercept) is for all three HP setups -.34. So independent of HP we get the same negative intercept, an indication of overestimation that is close to the above mentioned +/-.5. So now it seems as there is truncation. Without intercept the slopes are .058, .055, .053. Moreover, we don’t know if loot is rounded after or before empty correction. I guess after, but that’s only a guess.

If the dataset is from normal hunting then HP*1.1 is a good estimate for effective dmg done. If the low lvl mobs were killed with high end gear, then HP might be better. This would lead to a further interesting test. Loot should be slightly lower with high end gear.

Edit: Forgot to mention. The model for min loot should be estimated using empty corrected loot, however, to explicitely state it, empty correction should be undone, i.e.

if min = HP/20, then

real min = HP/20/(1-empty_rate)

The intersting thing is, that with the same approach I do get the same min for ambus, formidons and hoggs as with your data. So min loot might be valid for every mob hp class.

edit 2:
Noodles,

to get a feeling of the real empty rate, it's always good to check the confidence interval. For the saba y I do get 55.7% to 66%. So, although having killed 338 mobs the error i still some 5% (abs 10%).

To get the CI I can recommend this site. The normal approximation is rather unreliable for small samples. Furthermore, you'll get some possible estimates which are all more or less 61%.

[falkao]

I'm still testing Hoggs. Here a first direct evidence for a loot bias, i.e. a relation between waiting time to next global and its value.

fig. 1: Waittime to next global vs. value of global: Data Hogg n = 21


Click to enlarge


The correlation coefficient is statistically significant (r = .435 Pearson, p = .049; r = .524 Spearman, p = .015), which indicates that there is probably some kind of memory. Due to the small sample size, p-Values are quite weak.

I’m not sure if the globals with the short waittime (10-20) and low value are related to global grouping.

edit : interpretation
to observe a correlation between waiting time and loot value, something must get stored. It could be the wait time itself (I doubt on this) or the PEDs spent during this time (this is more plausible). This memory is then used to influence the loot value. The correlation has a lot of variation and so this explain the loot value only partially.

Moreover, there is again some indication about global grouping. Instead of getting a high one on long wait times, there might be triggered several globals with lower value.

[falkao]

Originally Posted by Doer

Here's another example of a prolonged, personal "spike" in loot (about 1.5 hours). ...

There was a break of probably about an hour between the global at 00:51 and 2:39. I hope this is useful.

to Doer,

Would it possible to you to take note of the number of kills between globals? I'm pretty sure you'll notice a correlation between the number of kills and global value, as I do with Hoggs. I guess about 25 to 30 globals are needed. It seems further, that this is valid for team hunting as well.

[Jimmy B]

Missed the 07-07 post, interesting falkao. My gut feeling is that loot does have some sort of memory so nice to see some evidence for it.

[falkao]

Originally Posted by Jimmy B

Missed the 07-07 post, interesting falkao. My gut feeling is that loot does have some sort of memory so nice to see some evidence for it.

indeed, but what I don't get is that nobody mentioned it so far. We had everything like eights, areas etc. but nothing that did relate it to some intra ava effect. We had a first weak indication in the global data, so I started testing it and I would recommend to do the same to everyone interested in the matter. You'll be surprised how precise this balancing is.

[Jimmy B]

Originally Posted by falkao

indeed, but what I don't get is that nobody mentioned it so far.

It just gets drowned amongst all the "got to find the hot area" posts. Personal loot cycles and the like have been discussed for years, just no-one has found solid evidence of it because people have just given descriptive reports rather than recording the stats

[Dion Red Strike]

Originally Posted by falkao

indeed, but what I don't get is that nobody mentioned it so far. We had everything like eights, areas etc. but nothing that did relate it to some intra ava effect. We had a first weak indication in the global data, so I started testing it and I would recommend to do the same to everyone interested in the matter. You'll be surprised how precise this balancing is.

some of us know that the system has a precise balance

Although, I would like to know what you think of the preciness of the situation. How do you think the loot is balanced? Or, in which way is the loot balanced?

~D

[falkao]

Originally Posted by Jimmy B

It just gets drowned amongst all the "got to find the hot area" posts. Personal loot cycles and the like have been discussed for years, just no-one has found solid evidence of it because people have just given descriptive reports rather than recording the stats

that might be the point. However, also on area we have only descriptive reports, but that found believers. Well, I'm always surprised about what people tends to believe.

[kalanen]

Originally Posted by falkao

indeed, but what I don't get is that nobody mentioned it so far..

Ahum, nobody listens to a village idiot

This has repeated (at least) since I have been keeping an eye on it - from HA 64 to 75: For every cycle of prof ranking there has been one grand peak and kind of "echo" of the big one, which occurred after short dry period. Bigger the peak was, dryer the short dry phase was. For me that weird little phenomena seems like the system aims for in = out minus little something for the house.

Hypotethical (hell not a suggestion) example: You have hidden looter-animal, looter-mutant, looter-human and looter/robot -professions and all of them are built up by several separate skills with several weight, like all professions are. All separate skill ranks have been divided to segments with different "unlock values", say [.5][1][2]. Every time you loot something, you gain skills that contributes to corresponding looting profession. More you have relevant skills overlapping at [2] phase, more of final loot's [*2] multipliers you can "access" when you loot something.

Simply put: system offers the potential but it's you who loots.

Loot ability, anyone?

[falkao]

Originally Posted by Dion Red Strike

some of us know that the system has a precise balance

Although, I would like to know what you think of the preciness of the situation. How do you think the loot is balanced? Or, in which way is the loot balanced?

~D

With preciseness I had the balancing rule in mind, but still analyzing it.

It was interesting to see, when I was 50 kills late for a global, that soon after I've got a really big one or several smaller ones close to each other (mean about 5-10 kills). So I was asking myself, what kind of criteria might be behind. What decides if I have to wait or not? Is it random or does it depend on something else? Does it exist to accumulated loot for a high TT items in loot, or should it give a further kick to globalling?

There is another thing that I noticed on Hoggs. As already mentioned class c3 is nearly missing with Hoggs. However, it was more frequent when beauty items did drop. Had quite some loots with them but atm they don't drop for me and so no c3 loot. So it seems that there is some relation.