So, I actually *am* a professor of Statistics. For reals. AND I've worked in mobile gaming, running stats on purchasing behavior. Mainly I just wanna say that you guys are awesome and funny. Most of you are at least kind of right and some of you are very wrong, but I love that you guys are thinking along these lines.
The bad news is that in the mobile-gaming company I worked for, they were always talking about using what's called "Dark UX" to manipulate players into spending. A horrible example that I do not know to be true (but might be): when you sim for a character shard or a piece of gear, and you need multiple shards to complete the character or set, the RNG is probably whatever it is. However, when you are only ONE shard away (say 49/50) they might indeed reduce the probability for that one, singular roll. Because nothing will make a person buy gems more than being balked that close to success and having to wait for their energy to replenish. I'll bet they literally get more gem purchases from people in that circumstance than almost any other, because it's not a thought-out purchase (like a pack of Chromiums), it's an angry, psychologically-induced spontaneous purchase.
It's all conjecture, though. Some companies are not above this. But although I believe in my angry human heart sometimes that EA is messing with us, I actually believe in my statistical, intellectual head that they are not and that very small probabilities just look like impossibilities to human eyes.
One more thing: the law of large numbers simply states that as your sample size approaches infinite, the mean of your sample will come to more closely approximate the true (population) mean. This may or may not be a good reference for what you're discussing. Maybe the central limit theorem (that as your sample size approaches infinite, your sampling distribution will come to more closely approximate the normal curve) is a better example. ; )
Keep on keeping on!
Samurai UX
How is it that your a professor of statistics.... and worked in mobile gaming.... Yet STILL stink at ships
If the drop rate is set to be 33% then I don’t think it’s possible for drop rates to reach a normal distribution, even with infinite repetitions (i.e. large numbers).
The data would fit something more like a Poisson distribution. If you observes one billion instances of drop rates of a shard when simmed every 5 times, and your drop rate is x/5 with possible values ranging from 0-5.
In this case, you would have an x axis for your probability distribution ranging from 0-5. A normal distribution would place the mean at 2.5 (50% drop rate) and allow for a symmetrical bell shaped curve. However, if we know the true population value is 1.65 shards per 5 sims (33% drop rate), then infinite observations of groups of 5 sims would converge on 1.65, not 2.5.
As such, the probability distribution would be positively skewed and appropriately fit a Poisson distribution. Because the highest probability value is 1.65, values closer to that will also have higher probabilities (i.e. more area under the curve) than values further away. This distribution would also accurately illustrate the fact that the probability of getting 0 or 1 drops out of 5 sims is much greater than 5 (which would be the least probable value).
The means sampled from various (infinite) trials would form a normal distribution. But it doesn't mean it is being sampled from a normal distribution.
I agree with your thinking that the distribution of drops is not normally distributed. The way you phrased stuff was "technically" confusing, though.
If the drop rate is set to be 33% then I don’t think it’s possible for drop rates to reach a normal distribution, even with infinite repetitions (i.e. large numbers). .
Huh? What does drop rate in a binary-outcome event have to do with "normal distribution" (of what exactly)?
Regarding the distribution of shard/gear drops for every N simulations. The law of large numbers suggests that as observations increase, the data will more closely fit a normal distribution. But, I don’t think this would be true if the true population distribution is a Poisson rather than normal.
See the above post. It is the sampled means from various trials that form a normal distribution. It isn't as if having tons of individual data points from the distribution transforms it into a Normal.
Like i said, there's no need for them to lay down a complicated mechanism where it frustrates you for that last piece when it's just as easy to let the RNG do that for them. In fact, it works against them to do so.
I will tell you that someone very close to me had an interview with Blizzard many months ago for some kind of micro-transaction consulting position...anyway, the entire interview was about how to manipulate compulsory behavior to entice spending. This included ways to manipulate perceived RNG and rewards, "shorting" rewards, etc.
Now, this is anecdotal, but I would be absolutely shocked if EA didn't employ similar strategies for the highest grossing Star Wars mobile game...
Like i said, there's no need for them to lay down a complicated mechanism where it frustrates you for that last piece when it's just as easy to let the RNG do that for them. In fact, it works against them to do so.
I will tell you that someone very close to me had an interview with Blizzard many months ago for some kind of micro-transaction consulting position...anyway, the entire interview was about how to manipulate compulsory behavior to entice spending. This included ways to manipulate perceived RNG and rewards, "shorting" rewards, etc.
Now, this is anecdotal, but I would be absolutely shocked if EA didn't employ similar strategies for the highest grossing Star Wars mobile game...
Amazing! It took this post to bury this thread.
I was wondering what happened to this thread...lol. I found a job listing that is very similar (maybe even exactly) the job I was describing:
Replies
Please also avoid implying other forumites don’t understand things. Explain, tell, don’t disparage.
How is it that your a professor of statistics.... and worked in mobile gaming.... Yet STILL stink at ships
The means sampled from various (infinite) trials would form a normal distribution. But it doesn't mean it is being sampled from a normal distribution.
I agree with your thinking that the distribution of drops is not normally distributed. The way you phrased stuff was "technically" confusing, though.
See the above post. It is the sampled means from various trials that form a normal distribution. It isn't as if having tons of individual data points from the distribution transforms it into a Normal.
Amazing! It took this post to bury this thread.
I was wondering what happened to this thread...lol. I found a job listing that is very similar (maybe even exactly) the job I was describing:
https://www.econ-jobs.com/economics-jobs/game-economist-11866
Job Responsibilities:
Anyone in the professional world knows EXACTLY what the above means...