Other players mining before you affects your hit rate

Entropia Universe Professions Mining & Resource Gathering
1

We just wrapped up the testing described here. Thanks to Leroy Casper Hunter and Leeloo Leeloo Mountain for helping out. There have been competing ideas out there on how other miners affect your hit rate, so I figured we’d try to some formal science on it.

To sum it up, this was designed like I would any scientific experiment at work to give this data a bit more rigor to have actual testable data. The three of us each dropped en/ore at the exact same coordinates with about a 5 minute buffer between us. Leeloo went first, then Casper, then myself. If someone caught up to the person ahead of them, we typically gave at least a minute or two buffer if not closer to 5 minutes. These were all done with F-101’s with no attachments, etc. so everything was standardized for depth, search radius, etc. There were 30 drops for each type, or 60 total drops per person.

Hit rate

The first miner had 26.7% HR, second was 6.7%, and third was 5.0%:

https://www.planetcalypsoforum.com/forum/index.php?media/feb23hitratetesting.333603/full

As you increase the number of miners (primarily having more than one miner dropping on recently mined areas), hit rate significantly decreases. For the formal research statistics end of things, I used a logistic regression with hit rate as a response, and miner number as the independent variable. One way to look the above graph is if the confidence intervals overlap, you cannot say those average hit rates are significantly different (i.e., they are only numerically different due to randomness). The confidence intervals basically mean that if you repeated this study 100s of times or had a much higher sample size (maybe 10k PED worth), we can say with 95% confidence that the “true average” falls within those limits without needing to actually do that much sampling.

Likewise, the logistic regression line had a negative slope (-1.2), and was also statistically significant p = 0.000982, also confirming that as you add more miners, hit rate goes down. P-values under 0.05 are generally the threshold for significance, so there’s next to no question that we ruled out randomness in this experiment and the differences we saw are due to the number of miners.

Hit rate however, was not zero for the 2nd or 3rd miners. Some claims could have been missed by the previous person because they found a claim closer where they dropped even though both were within their search radius. Another reason is claims “respawning”. As the third person in the chain, I did have one claim appear right next to me and not on the edge of the radius as you’d expect with the former example.

TT returns

I also analyzed the TT returns from these runs since it was possible claim size might be different depending on your hit rate. This gets a little trickier to analyze because that data doesn’t fit a normal distribution and are instead bimodal (a bunch of zeros for the NRFs, and another peak for actual claims). I won’t get into the nuts and bolts of that one, but in order from miner 1 to 3’s total TT: Leeloo 33.48, Casper 12.42, and myself 16.55 PED. There would be issues trying to graph that up due to the non-normal data, but I analyzed the total PEC with a zero-inflated poisson regression to take care of those. The short of it was that Leeloo’s average TT per drop was significantly higher than Casper (p = 0.04155) or myself (p = 0.00342). Casper’s and my TT per drop were not significantly different (p = 0.1988).

Conclusions

Essentially, TT returns declined with declining hit rate as more than one person mined an area. This go a good ways towards settling the question of whether other miners affect your hit rate or TT.

There are a few followups mentioned in previous threads now that we found there is a difference due to other miners:

[ol]
[li]Does the first miner only deplete claims at the depth they hit at, or at any depth within the radius (i.e., due claims exist in a disc, or cylinder)? This could be done in quick succession with at least two people at the same coordinates again. It would need to have enough range in depth treatments to not overlap much, so maybe like 200, 500, and 800m.[/li]

[li]What is the rough respawn rate? This could be done by repeating this again, but instead of 5 minute intervals, spread it out by increasing wait time of 30 minutes, 1 hour, 2 hours, etc. for each person to see when it’s “safe” to mine the area again.[/li]

[li]Do others in the same zone affect your mining if you don’t overlap drops? No need for staggering here again, just do 30+ drops in the same general vicinity and compare HR again.[/li][/ol]

I would like to take another weekend to test either 1 or 2 now that we have this wrapped up. I’m thinking March 9, but I’ll see what people think about what we have so far before setting up the next round of testing.

14 Likes
2

Nice bit of testing. I have always had a feeling that was the case.

3

Thank you very much King :slight_smile: hope we get plenty more testing to do !!!

I’ll link this thread also to my mining guide

Still some questions

What if different finders (read different depths)

Do deeper finders give the same results

Longer waiting times (I mostly work with 2 hour timers)

Different spawn times for different ores/enmatter?

As MA stated that skills do matter, we might do the same tests with higher (deeper) finders to compare, cause with that F-101 my HR was way of lol Maybe its because it’s the F-101 …

4

Was that a 16.55 PED or 165500 ped test because its late and I can’t read properly.
I hope its not the first case.I’ve been educated that a test should be conducted for a period of at least 6 months + with a very large sample of ped spent so you can remove any variation possible.
Anyway good luck :slight_smile:

5 Likes
5

there are different kinds of tests

like when you wanne see if you can find stuff on the same coordinates with autodropping without moving and you see that after the first hit the next 30 drops are all NRF then you dont need to do this test for 2 weeks and drop a million times. you can see the result after 2 minutes with a certain degree of certainty thats upwards of 95%.

if you wanne test the refresh rate you would have to do this for maybe few hours, depending on when you start getting hits again.

6

interestingly the second and third miner have roughly a fourth of the hitrate but have about half of the TT

you could also test this with doing ore once and enmatter once, with a few days inbetween. i assume the higher TT in regards to hitrate is due to more ore hits. could be luck but could also be a statistically significant difference

7

The sample looks a little small.

Repeat it 100 times, if the numbers are still the same I buy it.

1 Like
8

I’m pretty sure even if we did this 1K times, results would be the same, but I’m willing to go for it.

EDIT : that’s also the reason why I almost never mine during weekends

1 Like
9

Ty for doing this testing and surply the data. Yes its only small sample and there is other variables to consider. I would help, but on diffrent planet. I had some say its no point minging this area i alreay done. I did mine it and found gold and run was ok. The diffrence was his 105 522m depth and my milf 900m depth. Also i dont know if skill level has factor or just what tool you can use. Are all 3 of you simlier level?

10

As a hypothesis test I buy it 100%. The server/s keeps track of areas. But is this unique to mining or also true for hunting in same area or even crafting? I wonder how large the respawn areas are, if set to predetermined area sizes or dynamically determined. Is the return also a combo of 1. Local activity in the area + 2. Global activity on the planet? But I guess its not so easy to just say “More miners in an area = Less finds” if more acitivity in an area also means that players spend more in that area. Does that also mean that our spending in an area also determines how much there is to find there? Or is the entire spending on a planet spread out evenly on that said planet or even universe wide?

I have experienced fantastic hit rates and back to back globals and multis in more or less abandonded areas.

Maybe one way of do some further testing is to get in contact with some land owners. Don’t they have access to stats regarding mining activity in their areas? That way you could correlate activity vs returns maybe.

1 Like
11

with same setup cycle even 10k ped and results would be intresting to see.

Good data thank you.

12

When you’re using these statistical tests, they’re designed to use these sample sizes to make predictions of the “true average” you’d get if you did 1000+ drops instead. Those confidence intervals basically mean that if you repeated the study again, 95% of the time the averages would fall within that range. The higher the sample size, the more that confidence intervals shrinks, but the averages generally only bounce around within those limits as you increase sample size.

Something we never do in science is sample so much that we can get a near direct estimate of the true population size through sheer force. It’s expensive and time consuming. If the samples we pull come from a random distribution though (i.e., a random number generator with some set average), 30+ samples is typically enough as long as your comparisons are properly designed and you’re using the right statistical tests.

Likewise, those p-values I give already address that. There’s only a 0.0982% chance that 10k ped drops that the negative trend in the graph wouldn’t hold true. Similar for the TT test, Leeloo’s and my TT would still be expected to be different at that high of a sample size with about 99.7% confidence in that difference. That’s the fun thing about research statistics. You can determine what would happen at high sample sizes with much less.

2 Likes
13

I’m really cautious about directly comparing the actual numbers from TT because of the need for that zero-inflated model I mention. With that statistical tests, I can definitely say miner 1 had more than miner 2 or 3, but the way the model quantifies that doesn’t boil it down into simple math we can directly compare to hit rate. Those total TTs I gave pretty much can only give a general idea, but they have the same problem as averages not being “correct” because of all the zero data.

As for type affecting TT (just happening to get more ores with higher TT than enmatter claims), I forgot to mention I did look at that. I can actually account for that in the model as a covariate. That effect doesn’t change the model at (p = 0.1911), so the results pretty much stay the same.

14

Thank you (all three of you) for this interesting experiment. At the very least, it now satisfies my nagging query as to why I can sometimes get up to 20 NRFs in a row while mining planetside. :laugh:

Going off on a tangent now though

After reading this, two additional questions now pop up.
[ol]
[li]Is it safe to conclude (from this) that there is at least a 2D (x,y) spatial distribution to the mining claims? If yes, it would then lead me to wonder if there isn’t a 3rd dimension (z) to its distribution. Would going through with this again with everything done the same way but using three different avg. search depth finders, but with the same search radius, result in any difference in the hit rates? [/li][li]How much time in between (the drops) do you think it would require before there is no longer any interference (in hit rates) between the three miners? 5 minutes and 10 minutes is definitely out right? So maybe 30 minutes, or perhaps even 1 hour? [/li][/ol]

Lastly, it might probably just be smoke and mirrors, but I can’t help associating your hit rates in this particular manner/pattern.
[ul]
[li]1st Miner (Hit Rate): 26.7% [/li][li]2nd Miner (Hit Rate): 26.7% * 26.7% = 7.1289% (in comparison to the 6.7%) [/li][li]3rd Miner (Hit Rate): 26.7% * 26.7% * 26.7% = 1.9034% (in comparison to the 5%) [/li][/ul]
Its almost like having the probability of hitting a claim to be around 26.7% and then the 2nd/3rd Miners’ to be analogous to trying to hit 2/3 consecutive (in a row) claims in the exact same spot. And any discrepancy (in hit rates) can be explained off due to “luck” and “server’s regeneration of claims”. :scratch2:

As for the TT part

Would it be possible to separate the tally for number of claims and TT of enmatter and ores? Then do an average to see if there’s a difference in TT values (of the enmatt and ore claims) between the three miners.

It would be interesting to see if MA does any “compensation” to the 2nd and 3rd miners for their “reduced hit rates”.

:wink:

15

As King already mentioned we are going to do a lot more testing, the more questions, the more testing :slight_smile:

About the repop or respawn for claims, and now I speak from my own experience only, I wait 2 hours before redoing the same zone over and over again. I tried shorter times, but then my HR/TT return was less, a lot less.
If this is the same on any server or for other resources I never tested them all.
Those I know who mine the same way, try already after 1,30 hours but they go for specific ores most of the time.

Another thing that is worth mentioning is that you never can say that you didn’t find anything. Either you will find a claim some 1 has missed, most likely because he/she didn’t drop a 2nd bomb at the same place (rebombing) or there is a safety barrier that you will get a minimum if you keep mining in that zone.
Then I’m thinking of around 10-20% HR or even 25%
Either way if you get a HR of 25-30% you should consider stop mining there or keep a very close eye on your TT return so it doesn’t go under 90-80% (IMO)
These % are all for mining without amps/enhancers

16

Leeloo covered the rest pretty well. For the question of TT, I addressed this same question in my reply before yours. When you add resource type (ore vs. enmatter) as a covariate in the model, you can account for the fact that each type has a different base TT claim size. Accounting for that difference doesn’t change the results (p = 0.1911) on TT. So basically, no compensation.

I mentioned it in another thread, but when you pull random samples, you are going to get “streaks” as a part of that randomness (10 heads in a row in a coin flip). I’m pretty convinced that this is what is going on when people see “compensation” happening for a bad run. It’s a well known problem that people looking at raw data without appropriate statistical tests will see such trends in the randomness and think they truly exist. If someone repeated the exact same study a bunch of times or with a single higher sample size, they could test the “compensation” question further and rule out that randomess, but I’m planning to focus on the other things you mentioned that are also cheaper to test. I’ll probably draw up a new thread in a few days on the next round.

1 Like
17

Thank you very much for doing this experiment and for sharing the results.

Regarding this part:

That is only valid with some strong assumptions: independence of the observations, low variance and some sort of stationarity (i.e. that results do not depend on time and location). I am not sure any of these assumptions hold in this case, so I would not claim that the p-value for example is very accurate. Of course, the effect size is very convincing and would likely result in a high significance with a more elaborate model.

I find your analysis and conclusion quite convincing but I think it can be strengthened if this test repeated in different setting: different locations/times, wider time intervals and more avatars (easy to say, as I have barely time to play and cannot offer any help).

Again, thanks and keep up the good work.

1 Like
18

Thanks for reply and those test will be coming up in the next weeks when everybody can :wtg:

19

Some of that is addressed in my next round of testing thread, but the model assumptions are a big one (and I was the oddball who actually liked research statistics in grad school, so I love to talk them).

[ol]
[li]Independence of observations shouldn’t really be an issue unless someone is going to claim what you found in previous drops affects your later hit rate (or some form of autocorrelation). I didn’t find any evidence that something like that was going on when looking at the model data. [/li]

[li]Low variance isn’t a requirement, but maybe you meant homogeneity of variance is (i.e., not having highly variable variances)? The logistic regression is already tailored to dealing with this to a degree, and I didn’t find any evidence of overdispersion either as another check on that.[/li]

[li]The time factor would largely be minimized by us all testing in a fairly short window. For a location ( x% higher or lower hit rates for all of us), that would really matter for the purposed of this experiment. That or an interaction effect (miner 1 stays the same, but miner 2 and 3 have even lower hit rates), would get more into fine tuning if you wanted to really model exactly when and where someone should mine. If this were a full blown field study where I’d be putting out recommendations like that, I’d definitely want more locations. Just for showing that such an effect can happen though, this design did the trick, and it’ll be validated to some degree in the next round of testing.[/li][/ol]

I’m glad people are having fun poking around on this subject, so I’ll keep things going for a bit.

20

Thanks guys! Very interesting.