Talk:Drop rate/Trick-or-Treat Bag

Is there an automated way to update the percentages? Jimbowe 04:46, 27 October 2007 (UTC)
 * Nope, it's all manual work. [[Image:Entropy Sig.jpg]] (T/C) 05:43, 27 October 2007 (UTC)
 * and it's painful :) --Gwain 15:24, 30 October 2007 (UTC)

Percentages
Just by looking at the percentages and my own tests, it looks like the percentages are roughly 20% for the Apples, Corns, and Cookies, 10% for the Serums, Absinthes, and Brews, and 5% for the Ghosts and Tonics.


 * And a lot of that was contributed by Unindal, his data alone pretty much agrees with yours. 1500 bags... jeez --Gimmethegepgun 22:24, 30 October 2007 (UTC)
 * Speaking of numbers, in future it may be better for contributors to hold onto their bags until a reasonable number can be opened and recorded. While variances are put back in line due by adding up the data, opening 30 or 40 bags will produce more useful data than opening 10. Since we are measuring probability, higher numbers of bags opened at one time will produce more consistent results. We don't need to open 1500 at once (thanks Unindal :) ), but 20 or more seems like a reasonable measure. --[[Image:Rapid Fire.jpg|19px]]  Scottie_theNerd  (argue)  09:17, 31 October 2007 (UTC)
 * Statistically, there's no difference in opening 1000 bags at once or 10 bags 100 times. Would the chance of getting '2' on a dice be different if you roll it two times in a rapid succession, or one right now and one after 1 hour?--Gwain 16:03, 31 October 2007 (UTC)
 * I have another 550 I need to put in the table, will try to do when I get home. They follow suit as far as the percentages are concerned too.  And actually, the percentage post that started this was by me, forgot to sign. >< Unindal 16:42, 31 October 2007 (UTC)

(edit conflict)
 * The die example is invalid because we already know the probability is 1/6 for any value. Here, we are trying to establish the drop probabilities using raw data. If we use the die example, this is comparable to people rolling the die several times compared to several hundred times. One person who rolls the die twice might record 100% for one value, or 50% for two values. In contrast, the person rolling a thousand times is likely to more accurately reflect the 1/6 probability.
 * Adding all the test samples together will alleviate variances due to irregular sample numbers, so it's not a real problem as far as our record-keeping is concerned. I am simply pointing out the weakness in experimenting with low numbers of bags. There are 8 possible outcomes (i.e. items), so opening only 10 bags will not likely result in a mathematically useful number of results. For example, Bob the Hick's results from 11 samples go against the data obtained from Unindal. In fact, Unindal's extraordinary effort alone can establish reasonably accurate drop rates.
 * The point here is not to challenge the laws of probability; I am addressing consistency in the research method. True, 100 users opening 10 bags each will most likely show the same rate as one person opening 1000 bags. However, we don't have 100 users contributing their data to the table. So far, it is the large sample testers that are providing the bulk of the statistics, and for the most part they are canceling out the results obtained from the small-scale testers. This is not to discourage contributors, but a small point to use a reasonably proportionate amount of samples in relation to the number of possible outcomes. --[[Image:Rapid Fire.jpg|19px]]  Scottie_theNerd  (argue)  16:53, 31 October 2007 (UTC)
 * Props to Scottie for the above post. Coudn't have said it better myself.  I understood this concept prior to opening the bags and that is why I open them in bulk.  Once again, excellent post Scottie (you nerd :-) ). Unindal 19:39, 31 October 2007 (UTC)
 * However, your point is largely invalid, assuming that people (the ones who posted here) record every single bag they open. If they do that, then it will, added up, make it all even. However, the large numbers added will balance out those that do NOT record every bag they open, so they're still useful --Gimmethegepgun 20:29, 31 October 2007 (UTC)
 * All the same arguments, different drop rate article. :-)  Unindal 20:58, 31 October 2007 (UTC)
 * [The die example is invalid because we already know the probability is 1/6 for any value. Here, we are trying to establish the drop probabilities using raw data. If we use the die example, this is comparable to people rolling the die several times compared to several hundred times. One person who rolls the die twice might record 100% for one value, or 50% for two values. In contrast, the person rolling a thousand times is likely to more accurately reflect the 1/6 probability. <-- i think (i could be wrong) that this is plain wrong: because you don't care about the single testers when you try to guess the % of every object, you get the WHOLE community and sum up the probabilities of the whole bunch of items. so, if i try 1 bag 100 times i can say that, summed with the other gift got from every other players, my prizes will fall into a Normal Distribution for every istance of every single object (i really hate to explain something so technical in a language that isn't my own, i hope you guys undertand :) )--Gwain 04:57, 1 November 2007 (UTC)
 * after all, what's the difference in your example, statistically speaking, of one guy rolling a dice one thousand times, from 1000 guys rolling the same dice once and summing up their results? none.--Gwain 05:00, 1 November 2007 (UTC)

Again, I'm not arguing that they differ statistically. I am addressing a possible weakness in the experimentation method. One person opening 1000 bags and recording the distribution of outcomes is a single, controlled experiment. If we had 1000 users open one bag each and we collated the results, that too is a single, controlled experiment. The results we obtain would be accurate either way because we can replicate the results if the same guy opened another 1000 bags, or if we got another 1000 people to open one bag each again.

However, the drop rate projects are based off players contributing their own complete data. Each player records their distribution of outcomes, and together we calculate the probability rates based off these figures. Several users have compiled results from 100 or more samples; Unindal has his 1000 and 500 sets, and we've recently seen more large contributions. We can visually see that the results are generally consistent with each other, which means they are accurate results using the identical method.

The point of my digression on this page is simply this: if you are going to submit your experiment data, please be reasonable in the number of samples you use considering the number of possible outcomes. Opening 7 bags when there are 8 possible items won't give a useful representation of the item distribution. Additionally, because of the small sample number, the results of others will practically cancel out the impact it might have made on calculating the probabilities. Because we are compiling all the results into a grand total, inconsistencies are ironed out and we can establish the probability with reasonable accuracy.

This was only intended as a minor observation. Mathematically, the results will be the same. Experimentally, they will not -- and the data table we have here shows that. I do not intend to debate the mathematics behind it because, for our purposes, what we are doing works -- mainly due to the large-sample experiments "making up" the large total sample number (5000 as of this comment -- good job). However, it doesn't hurt to follow a reasonable experiment process, and that's really all I had to say. --  Scottie_theNerd  (argue)  08:23, 1 November 2007 (UTC)