Career Officer
Join Date: Jun 2012
Posts: 1,500
# 41
02-07-2013, 08:00 AM
the results Borticus posted sound logical to me.

but i would suggest to take a closer look at the Random Number Generator that STO uses too.

"Random" or "Shuffle" are often just not random enough.

Take a Winamp list of 5000 Songs, always listen to that same playlist in Shuffle mode, after a few weeks/months you will notice that you hear the same few songs over and over again, even in the same order, and notice that other stuff is never played at all.

-> my money is on the RANDOMIZER to be just not *random* enough.


Probably the same reason the DOff Upgrinder spits out more Purple Bartenders, Chefs and Sci Officers (for 5000 dil a pop) than anything.
Career Officer
Join Date: Jun 2012
Posts: 442
# 42
02-07-2013, 08:06 AM
Quote:
Originally Posted by diogene0 View Post
You forgot something extremely important: the human factor. As a geography teacher myself i'm not completely new to probabilities, but to understand this you often need to get your head out of the %, especially if humans are involved. Here it's not just coins you're flipping in an ideal environment.

This game has 2M players. I'm too lazy to make the math now, but with this amount of players, you'll get obviously a significant amount of unlucky people. These people will tend to come on the forums more often than the happy ones, and they will post together in such threads to complain.

I'm not saying that everything is fine and that it wouldn't require further studies, but this topic whehe 10 or 20 people complain is clearly not enough to say that there's something wrong. Because these 10 people could very well be among the most unlucky guys in the game. As a general rule, when people have the freedom of speech, they will use it to voice their concerns before saying what's fine for them, if they ever say it.
The human factor should not apply here because each event is independent of the other. One console craft = 1 roll one the loot table and my roll does not affect yours. Therefore it is just a "coin-flip". Unless the developers have constructed such a complex loot system where everyone's action affect others. Yes, you can have unlucky "streaks" but the probability of having such an unlucky streak is minuscule in some of those tests. With 300 replicates, the OP's data, if accurate, do not portray a uniform drop-rate.
Career Officer
Join Date: Jun 2012
Posts: 442
# 43
02-07-2013, 08:11 AM
Quote:
Originally Posted by deusemperor View Post
I make 21 consoles a day across 7 toons. (Toons are purple critical trait, resolve, and engineer 24/51/24)
My console spread purple blue and green right now:

p=210,
b=395,
g=175,1,1,
Taking your information here (and I assume you mean you got 177 green consoles), and applying the same simple analysis as above we would expect your to get:

Purple = 782*0.24 = 187.68
Blue = 782 * 0.51 = 398.82
Green = 782*24 = 187.68

We do a quick test to get a chi-square = 3.299, df = 2, p = 0.1921.

That tells us your observed drops do not significantly differ from your expected drops. Thus, the quality of the console drops is occurring as expected or correctly.

Now if we do a post hoc power analysis (which really should only be done a priori) we find the following power (1-beta) of the test above:

Small Effect Size - Power = 0.850
Medium Effect Size - Power = 1.00
Large Effect Size - Power 1.00

So in this case, the sample size is large enough to determine even a small difference between the distributions.

Last edited by commodoreshrvk; 02-07-2013 at 08:25 AM.
Lt. Commander
Join Date: Jun 2012
Posts: 190
# 44
02-07-2013, 08:20 AM
The human factor isn't, I think, as critical. What are the odds that a player will track the results of every console created? That will seriously reduce the sample pool, and make it unlikely for the player to be an unlucky outlier. And if there are some "unlucky" players, where are the lucky ones? Someone pulling MKXII purple disruptor coils every other console would be crowing about it on the forums! (with STF old-style loot, there were definitely those who were unable to find that last piece, while others found all three in a few STFs)

I am curious how the loot tables are created and handled in the code. Are the consoles listed in alphabetical order? Or in chronological order? Perhaps there is a problem in how loot is generated, and things in certain positions on the list have a skewed probability of dropping; ie rounding or something is causing things later in the list to have a lower chance of dropping. Bort isn't going to share code, but perhaps he could test flipping the order in a way that should not change the results.

Another possibility is how the random number to jump into the list is generated. RNGs are hard to do well, and its possible that someone developed what they thought was a good RNG (hell, could even be a problem in the compiler), but it has some flaw in it causing a skew in the results. It would be interesting to run the same RNG algorithm in a separate environment and histogram the results. Could be that with a small sample size the RNG looks random, but as you run up more samples the RNG starts showing some peaks and valleys in the histogram.
Career Officer
Join Date: Jun 2012
Posts: 442
# 45
02-07-2013, 08:59 AM
For those of you still worried about sample sizes for the OP's data, here are the corresponding post hoc Power Analyses:

Science Consoles
Effect Size = 1.118
alpha = 0.000213532
Sample Size = 131
df = 9
Power (1-beta) = 1.000

Engineering Consoles (I had to put in an observed probability of 1x10^-8 for the consoles the OP did not obtain)
Effect Size = 1632.99
alpha = 4.36x10^-22
Sample Size = 94
df = 14
Power (1-beta) = 1.000

Tactical Consoles (I had to put in an observed probability of 1x10^-8 for the consoles the OP did not obtain)
Effect Size = 883.88
alpha = 1.69^10-9
Sample Size = 88
df = 15
Power (1-beta) = 1.000

All three tests have sufficient sample sizes to determine the observed effect sizes.
Cryptic Studios Team
Join Date: Jun 2012
Posts: 1,144
# 46
02-07-2013, 09:32 AM
Quote:
Originally Posted by zerobang View Post
-> my money is on the RANDOMIZER to be just not *random* enough.
If (great big IF) anything is truly amiss with this process, THIS would be the culprit.

All the additional hooplah over "hidden weighting" is, quite frankly, a pile of ... stinky things. The data is, what the data is, and I've already publicly stated exactly what that data is.

So, if (again, this is a hypothetical ONLY) there is anything off-kilter about the overall distribution of rewards from these tables, the only remaining culprit is the core RNG, about which I have approximately zero knowledge or input.

I'll send out a feeler to investigate, though, just for peace of mind. Don't get your hopes up about results, though.

Before I go, I'll leave you all with one more parting thought: http://en.wikipedia.org/wiki/Confirmation_bias

The numbers are what they are, but I think that at least some of the outcry being posted here is largely motivated by an emotional reaction to those figures, due to the effect they have on your in-game currencies (including Time as a more abstract currency).
-=-=-=-=-=-
Jeremy Randall
Cryptic - Systems Design
"Play smart!"
Quote:
Originally Posted by Me
Kurland here...
Career Officer
Join Date: Jun 2012
Posts: 442
# 47
02-07-2013, 10:04 AM
So how the loot roll occurs may be different as well, whether there are 3 rolls - quality, category, console (equals 9 end loot tables per tier) or 2 rolls - quality, console (equals 3 end loot tables per tier). In the prior, each loot table would consist of a quality/category combination, e.g. Uncommon Science Consoles, and in the latter each loot table would consist of just each quality, e.g. all Uncommon Consoles.

For everything to be even, and following the OP's quality proportions we get the following percent chances to get a specific console on a given roll:

For three rolls which would be for the Quality (OP's values), then for the Category (1/3 for Science, Engineering, Tactical), and then for the console (1/10 for science, 1/16 for Engineering, and 1/16 for Tactical), the percent chances of getting a specific console are:

Purple Science - 24.0%*33.3%*10.0% = 0.80%
Purple Engineering - 24.0%*33.3%*6.3% = 0.50%
Purple Tactical - 24.0%*33.3%*6.3% = 0.50%

Blue Science - 51.0%*33.3%*10.0% = 1.70%
Blue Engineering - 51.0%*33.3%*6.3% = 1.06%
Blue Tactical - 51.0%*33.3%*6.3% = 1.06%

Green Science - 24.0%*33.3%*10.0% = 0.80%
Green Engineering - 24.0%*33.3%*6.3% = 0.50%
Green Tactical - 24.0%*33.3%*6.3% = 0.50%

For two rolls would be for the Quality (OP's values) and then for the Console (1/42 = 0.024), the percent chances of getting a specific console are:

Purple - 24.0%*2.4% = 0.57%
Blue - 51.0%* 2.4% = 1.21%
Green - 24.0%*2.4% = 0.57%

So out of the 314 consoles the OP obtained there should have been 314*0.005 = 1.57 or 314*0.0057 = 1.79 Very Rare Phaser Relays, SIF Generators, Sensor Probes etc?

Last edited by commodoreshrvk; 02-07-2013 at 10:06 AM.
Career Officer
Join Date: Jun 2012
Posts: 71
# 48
02-07-2013, 10:08 AM
There is something wrong with RNG, though it might be a Doff only issue (or was, as I do not have it happen now as often as it used to). If I had a dollar for the number of times I have mass failed or mass critted 4+ Doff assignments in a row (as in a straight line, not 4+ randomly out of 20 total), I might have enough to get all 3 versions of the NW account packs. In the cases of mass failures, I normally had like 1% to 4% chances of failures for each, so having 4 or more in a group of 20 is possible but 4+ in a row would not seem reasonable, especially since it was pretty much daily at one point (or several times a day when I was overdoffing).
______________________________________________
Many thank yous to the The G&T Show for the Holo-Leeta doff!
Rihannsu
Join Date: Jun 2012
Posts: 12,506
# 49
02-07-2013, 10:15 AM
Hmm.... I've seen some interesting posts in the past that made me wonder about the RNG in general.
HAIL HYDRA!

-=-=-=-=-=-=-=-=-=-
I can haz joystick!
MMOs aren't charities. Corporations are supposed to make a profit. It's what they do.
Starfleet Veteran
Join Date: Jun 2012
Posts: 105
# 50
02-07-2013, 10:37 AM
One of the changes I'd make to the game if I could, would be to split the final console-making assignment into three - one for an attempt at making tactical consoles, one for engineering consoles, and one for science.

This would increase the chance of people getting an console they'd actually use, and far fewer MK XII greens cluttering the exchange.
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off


All times are GMT -7. The time now is 02:06 AM.