If you wouldn't mind posting one more time, what do you think is wrong with that value? Its just an illustration of where your method can fail.Originally Posted by Yokai
I'm actually a graduate student in statistics, so these sorts of questions interest me. I'm also interested in finding (relatively) simple ways to explain statistical intricacies (like the one we're discussing), so I appreciate you input, if perhaps not your tone.
I hate to be a **** here but you need to cull your data and start over with a better system. Anything past 4 in the options menu is creating false data from a bunch of trolls who have nothing better to do.
Thumbs up for the idea and the work done ... keep up ....
playing on PC the real gaming platform"if you rush me ... we all go boom... if you yell at me ... we all go boom..."
The sources for those samples are really not trustworthy, making statistics in a non regulated environment is always fail.
This thread's really blown up- far bigger than I expected when starting it, but great to see... And the discussion has gotten much deeper than my understanding of statistics goes. :S
So from reading the responses, are these the changes you guys want to be made?
- Use validated frequency as basis for new calculations
- Add estimate for margin of error. (with forumla 1.96*sqrt(p*(1-p)/Number of Items for Tier) ?)
- Change 'Drop rate / Cost per try' to 'Expected drops / 100 keycodes'
Again, I wouldn't mind giving you guys access to modify the spreadsheet if you want to help, though I'll admit I don't really understand sure what the disagreement with calculations is about.
I opted to have the option to input more than 4 items of a type in one go as a compromise between saving time for actual users submitting data for multiple lockboxes, and preventing trolls from entering large amounts of false data easily. I'm open to suggestions for ways to do this better.
Sadly, as i said in a previous post the last time i checked this thread, you will get people adding in troll data regardless of what you do. the only way to ensure it 100% is to validate, most people will not do this, but i'd gladly submit proof (videos) of my data because i want a truely accurate data, as the jerkish dev's refuse to give us probabilities for the lockboxes...
Okay, first off, I apologize for the prickly tone. Second, I apologize for being wrong, lol. I read your most recent example in a rushed state and misinterpreted something.
So, permit me to grovel for a moment and savor the taste of shoe leather. Okay, that was supposed to be funny but just sounded weird, lol (Jon Stewart face)
Okay, that being done now, you are correct. If the percentages were different from their current observed values, then indeed in the "over time scenario" the cumulative probabilities for 12 T4 versus 24 T3 versus 48 T2 would be a more accurate comparison.
I salute your patience. I will also revise the numbers in my sig block link below to reflect this more accurate comparison.
That all said, in this particular case with these particular numbers, with an observed 3% drop rate on T2 boxes, with a 2000+ sample size, it still holds that even in the 12 - 24 - 48 comparison, T2 is still the clear winner. BTW in that short hand table a few posts above with the 12 - 24 - 48 numbers? The three columns are (left to right): AT LEAST 1......AT LEAST 2.....AT LEAST 3. Similar to the more verbose layout in my sig block link below. Make sense?
Yes, it can be arguable whether to trust the sample collection method in Noupoi's spreadsheet, but let's assume people were mostly honest and didn't try to grief the collection data). Perhaps it _would_ be beneficial to all to wipe the spreadsheet, strengthen the form input against possible (or accidental) impossible results input, and start over. I think our current sample size in the current run of the spreadsheet is high confidence, so wiping and sampling again wouldn't hurt anything.
In fact, as I suggested before, it might be worth wiping and restarting every month just in case the Trion devs decide to tweak the real drop rate numbers and not tell us they've done so.
I'm glad there are other people on here who know how to use statistics. It looks like to me that the efficiency of keycodes to orange will eventually become closer and closer since the statistics of tier 2 lock boxes is more common and more accessible. I think that it doesn't matter what box you buy you will get the same amount of orange over a long enough period. But because it is not basing the orange directly off of each person, I would suggest you buy tier 2 boxes if you think you are lucky enough to catch those rare oranges, but if not go for the tier 4 you will be let down a lot less. My two cents. If you guys need any more help with the statistics on any other part of the game hit me up.
What about bits?
Costs for lockboxes in bits are different from rescources.
T2 120 bits
T3 240 bits
T4 400 bits
So what would be the most effective way to get orange using bits?
I added a rough cost in dollar terms to your quote. Personally, if you're spending bits, don't buy lockboxes. Instead buy inventory or buy Loot/Scrip Boost. Keys and Salvage are too easy to obtain.
I'm 606 EGO with 120k Salvage, 39k Scrip and spent my Keys but I do have 272 inventory. I haven't salvaged any items since 500.
Posting Permissions
Reply With Quote
