How to analyze loot

[Jimmy B]

Originally Posted by truantduck

4) Didn't MA at one point inform us that different types of damage will decay armor differently?

Interesting thoughts. Just in response to the above, all damage types decay armor the same at the moment. This may have changed from the past but that's how it works currently.

[falkao]

As mentioned above, we noticed that globals do seem to accumulated within an avatar, e.g. instead of observing them from time to time according to their relative frequency, they seem to come in groups with longer pauses in between.

I do have only limited data to answer this and therefore the results are preliminary.
To get something meaningfull I did the following. I have no data about number of kills between globals. Therfore I do use an internal clock namely the sequence of the globals itself. So all I do present later on has the interpretation of “how fast a single ava is in globaling contrasted to the others”. The data was transformed using the following procedure:

1) Ranked all recorded globals according to datetime.
2) Calc. the diff between consecutive globals within ava
3) Calc. mean rank diff within ava

With this we get a measure on how sparse gloabls are within an avatar. Here are the results:

In the following "n globals" refers to the number of globals an avatar has and "rank diff" refers to the difference in global ranks. To get a better feeling about the data both variables have been binned using quartiles (i.e. every bin contains 25% of all observations).

Tab. 1 Crosstab of Number of Globals vs Rank Diff


Click to enlarge

Avas with a low number of recorded globals (< 5) do have most of them in class 4 of the rank diff classes (1863+). With in median 2000 globals per day (roughly 80 per hour), we can conclude that in about 40% of cases there is gap of one to several days between globals. However, rank classes 1 and 2 are also quite frequently (nearly 50% in total) which shows that there is also another scenario where globals are quite close to each other.

For global class 2 and 3 we get a similar picture except for rank class 3. It tends to be higher than in global class 1. Rank class 3 starts at about 2-3 ours till one day. So it might be that rank class 3 and 4 do represent gaps between online periods. I can imagine that one group of players logs in nearly daily playing for 1-3 hours.

Players in global class 4 would then be the hardcore players, that are online several hours per day and every day per week. If we look at their rank classes we see, that most globals are in rank class 1 (e.g. about every half an hour). Nearly no globals are recorded in rank class 4.

So I do interpret this data as a distinction between player classes and that rank class 3 and 4 do express playing pauses. Therefore I did any subsequent analysis only for rank class 1.

Fig. 1

Click to enlarge

The mean of the rank differences within an ava shows a normal like distribution. Its mean is about 13 globals, so they are quite close to each other (within 10-15 min).


Fig. 2:


Click to enlarge

There are no differences in mean rank between global classes. This implies, that the time difference in globals is the same for the different types of players. Please don’t confuse this with the frequency in getting globals. This can’t be analyzed here. What we get is, that the same rules are applied to the different player classes. So if a player that plays from time to time will observe his globals in a short period of time quite like a hardcore player.

We can confirm again that globals do accumulate, what we need now is to understand why? Should this be some kind of adrenalin kick? Or maybe you have to spend first some money to get a global feeling? Who knows?

Edit: Team globals have been excluded from the analysis.

[Legion]

Originally Posted by falkao

As mentioned above, we noticed that globals do seem to accumulated within an avatar, e.g. instead of observing them from time to time according to their relative frequency, they seem to come in groups with longer pauses in between.

Couldnt this "build up" just be the timer that i speak of passes them by at that very moment?

[jdegre]

Originally Posted by Legion

Couldnt this "build up" just be the timer that i speak of passes them by at that very moment?

this comment seems to point again towards the Unified Loot Theory:

[Noodles]

Originally Posted by falkao

So I went into ava analysis and did the first steps.

The interesting thing about this is the distribution we get [edit] in this short period of time. Its normal to have an exp on waiting times, but when Globals seem to come short after each other for those type of avas, there must be something. [edit] The overall probability to get a global is about 2.5% in the worst case. To kill a mob with hp >1000 at least 1 min is needed. I do observe some avas having 6 globals in 1 hour. So they did kill between 30-60 creatures or mabye less within this hour. The prob. to get 6 globals out of them is less than .003. This is rather low, too low I would say. But how can this then be?

Getting caught up on this thread, apologies if the following comment has already been made.

I can't tell from your comments in the above post if you are advocating for a random global hit pattern, or one that has memory. But I might take issue with the comment of only 30-60 1000hp mobs killed in an hour. But both mod merc + A106 and foeripper+amp do more than 80 damage/sec (on average, not max). For a 1000hp mob, that's 15 seconds at most. So two mobs per minute should be easily done if the mob density is good. And therefore the 0.003 probability of 6 globals in one hour should be considerably higher.

I'll continue reading.

[falkao]

Originally Posted by Noodles

Getting caught up on this thread, apologies if the following comment has already been made.

I can't tell from your comments in the above post if you are advocating for a random global hit pattern, or one that has memory. But I might take issue with the comment of only 30-60 1000hp mobs killed in an hour. But both mod merc + A106 and foeripper+amp do more than 80 damage/sec (on average, not max). For a 1000hp mob, that's 15 seconds at most. So two mobs per minute should be easily done if the mob density is good. And therefore the 0.003 probability of 6 globals in one hour should be considerably higher.

I'll continue reading.

If triggering of a global is completely random according to its relative freq. (let's assume 2.5%), then we would expect one in mean in every 40 to 50 kills. Sure there might be by pure chance observations of some globals close to each other. However, getting 6 of them in a row (1 hour) would contradict pure randomness over time. With that I don't refer to the overall number of globals one gets, but how much one gets per time unit. The overall number is still the assumed rel. freq but instead of observing them sparse over 40-50 kills, we see rather long periods of no globals (say >100 kills) and then suddenly more than one of them in a short period.

So all is still random but biased in some way. To accomplish this, some kind of memory is needed, otherwise it wouldn't be possible to implement such a thing.

My own interpretation is, that there is some kind of memory of hp dmg done, like memory for skill gains. After you have reached a lvl you have access to globals. When overall global rate is 2.5% and if I do delay them by say factor 10, then I might trigger them with a rel. freq of 25% when the accumulated hp lvl is reached.
A rel. freq. of 25% would correspond to a global in every 4th kill. So I need a further memory that tells me how much globals have been triggered in their period, or how much dmg done was consumed.

The only explanation I have atm to justify such a system is, that it syncs payout to sustained costs. With a purely random system, I could have by chance the possibility to get a high payout in the first kill or kills. With many players, such a system will imbalance the return per ava. Money paid out for one ava is lost for the others. Having a memory for hp dmg done would prevent this. The system is still random but biased or corrected to accomplish consumed Peds.

[falkao]

Originally Posted by Legion

Couldnt this "build up" just be the timer that i speak of passes them by at that very moment?

When there is such a timer, then it is active within ava and not per server or area. Since we know now, that there is globals grouping, the observed area or server effects might come from this. It would be rather normal that one believes, that there is such a thing, when suddenly many globals begin to drop.

[Kolobok]

Originally Posted by falkao

So all is still random but biased in some way.

100% correct.

Originally Posted by falkao

To accomplish this, some kind of memory is needed, otherwise it wouldn't be possible to implement such a thing.

...

My own interpretation is, that there is some kind of memory of hp dmg done, like memory for skill gains. After you have reached a lvl you have access to globals.

My own experience (and probably Witte's, if he sees this and might want to comment on it) says that this is not the case. There is no such thing as accumulation of hp (that would be identical to "hunting only" loot pool). But it is possible to find a hot mob and hot area for oneself, except for some very bad days.

Originally Posted by falkao

The only explanation I have atm to justify such a system is, that it syncs payout to sustained costs. With a purely random system, I could have by chance the possibility to get a high payout in the first kill or kills. With many players, such a system will imbalance the return per ava. Money paid out for one ava is lost for the others. Having a memory for hp dmg done would prevent this. The system is still random but biased or corrected to accomplish consumed Peds.

I believe that the system tracks the total expenditures made in all three areas, hunting, mining, crafting, but rewards the player through one of these three activities dynamically. Sometimes it is better not to mine but to hunt, and vice versa.

[falkao]

Originally Posted by Kolobok

..

My own experience (and probably Witte's, if he sees this and might want to comment on it) says that this is not the case. There is no such thing as accumulation of hp (that would be identical to "hunting only" loot pool). But it is possible to find a hot mob and hot area for oneself, except for some very bad days.

we have atm 2 explanations for the global grouping. Area or intra ava effect.

Here some more details for Hogglo loots only. They are in one area and hence there should be some area effect if it exists.

binned wait ranks

wait rank		Percent
1  <= 29,50	0.36
2  29 - 190	0.28
3  190 - 1863	0.20
4  1863+		0.16

Seq	Percent	Exp 	Ratio
1.1	0.16	0.13	1.23
1.2	0.10	0.10	1.02
1.3	0.05	0.07	0.74
1.4	0.04	0.06	0.77
2.1	0.10	0.10	1.00
2.2	0.09	0.08	1.18
2.3	0.05	0.06	0.90
2.4	0.04	0.04	0.80
3.1	0.06	0.07	0.83
3.2	0.06	0.06	1.03
3.3	0.05	0.04	1.15
3.4	0.04	0.03	1.16
4.1	0.04	0.06	0.71
4.2	0.04	0.04	0.89
4.3	0.03	0.03	0.95
4.4	0.05	0.02	1.94

The table refers to the binned wait ranks. So for every ava the wait time between globals in ranks was calculated from the overall datetime ranks. The ranks have then been grouped into wait bins using quartiles. Team globals have been excluded thereafter. Therefore, the rel. frequencies per bin are not exactly 25%. For the intra ava wait seq I did extract the ordered 2 letter words. (Btw. this is what we do in phase diagrams).

So if an ava has the following wait seq
1121442 we will get the strings
1.1 1.2 2.1 1.4 4.4 4.2

In using the overall freq. of a bin, we can calculate the expected freq. of a word. So for example, wait bin 1 has a rel. freq. of 36%, and therefore if everything is purely random, the word 1.1 should have a relative freq. of 36%*36% = 13%. We do however observe 16%, so a value that is 1.23 times higher than expected.

If you check the single words, you'll notice that the words with the same letters are always higher than expected. The bins 3 and 4 might be longer pauses between gameplay. However, also there is some grouping.

The observed freq. are significantly different from the expected ones.

What can that tell us. The double letter words 1.1, 2.2, 3.3 and 4.4 are all more freq. than expected. The words starting with 1.x do decrease in freq. from 1.1 till 1.4. So if you are in wait bin 1 you probably will stay there (global grouping). If you are in 2.x then you'll also stay there for some time (2.2). It is more likely to pass then to a 2.1 contrasted to a 2.3 or 2.4.

I don't like the bins 3 and 4 very much. They are hard to understand. They might be real waits till the next global or real pauses. Let's take the 4.4. This means that wait time between globals is several days. So if one is there he will stay there for some time because 4.3 to 4.1 are less frequently. Hard to interprete, maybe a sporadic player, or some with bad luck.


We see that waits are not equally distributed. There is a trend to a repetition of the wait states in which one is atm. This can be explained by an area effect or some intra ava mechanism. To know what it is, I might need to identify avas hunting at the same time in the same area. So Hoggs might be good for that. If there is an area effect, then they should get a similar global wait sequence. If it's intra ava, then not.

So let's see if I'm able to get a good sample.

Edit: It might be combination of both effects as well. If my assumption on intra ava hp memory is correct, then thinking on taxed areas, this might apply as well. Why should there be a payout in an area before ped's are consumed there.

I doubt that the recorded global data might be sufficient to respond to this question. With global data alone you won't know if an ava pauses. So I guess we might need some experiments. But maybe I can find something.

[Immortal]

You refer to wait times between globals etc. You're assuming that it is one time-dependent function (we know it's not time dependent, but hp dependent, but we assume that the average hp_reduced is directly proportionate to time. I don't have access to the logged data yet so I haven't looked at this myself. However it would play along the lines of deconvolution of the global signal received {f(x),f(x+a)...f(x+a.n)}

In asking "why do globals occur close to eachother" you are assuming there is one function at play that delivers globals. Intra-avatar explanations aside (doesn't explain team globals at all, which is why I do not agree with any analysis that includes "avatar-based loot" but excludes team globals), the secondary, trinary global function (if mob-specific) may only be activated once a hp-reduction is reached - and these global function have a short life span. This may explain why large ath's are seen after a global - once again I don't have access to the logfiles yet to confirm or do any real analysis myself.
In short, the sequence proposed:
Avatar globals->Avatar globals again
may be
f(x)
f(x+a)
where a is hp_reduced interval

Or
Avatar globals->Avatar globals again
f(x)
g(f(x)) or g(t)/g(hp)
f(x) cyclical lifespan function, g(x) may ->0 as t->(1/0) or hp_reduced->(1/0)

Both convoluted function theories account for global spam on one mob. The second accounts also for global spam encountered during events. Both are avatar-independent.