Best Of

There's a story in one of Danny Kahneman's books (The Undoing Project I think) about an experiment he used to run with his econ classes. He would tell the students to write a string 100 H and T on the chalkboard for 100 coin flips that they think represents a fair coin. Then actually flip a coin 100 times and record the results above or below the first 100 "flips". He would then leave the classroom, and then come back when they were done and tell them which one is the real coin and which is their heuristic version.

The way he did it was by looking for long strings of H and T. The one that students thought looked fair would always fail to account for the fact that in 100 flips you were overwhelmingly likely to see long strings of H or T in the true random sample. Randomness is much more clumpy than human heuristics think it should be.

Anyways, the Titan crystal is random.

Won't hurt you this time to post the stats like miike did last time. We can thow this alligation in the bin for next 2 years atleast.

There's a problem with this, and it is that people will start to assume whenever Kabam doesn't release drop reports they have something to hide. But on the other hand, if Kabam sends out enough reports, sufficient navel-gazing will start to turn up statistical "anomalies." As I mentioned earlier and Crashed also mentions here, the big problem is most humans don't understand randomness enough to properly evaluate it. That is precisely why I believe there's no there, there, with regard to the GCP mentioned above. They are doing what the crystal conspiracy players are doing, just on a larger scale. They are looking for non-random behavior in random sequences, because they think there is such a thing as specific non-random behavior.

Which one is random, which one is not:

HTHHHTHTHH
HHHHHHHHHH

Trick question: both show up equally often in random sequences. By definition, a genuinely random sequence will produce every possible sequence of ten Hs and Ts about equally often. You need context to determine if either one of those sequences is "non-random."

Show people enough random numbers, and they'll eventually find the non-randomness they are looking for. Randomness is a statistical property, not a localized one.

For years I have been keeping track of my arena crystal drops. Actual, individual drops. Here's four years of TB crystals, showing how many units I got per batch of TB crystals (20 crystals = one million BC).



Is that random or not? Suppose I wanted to prove it wasn't. Hmm, let's see what I can do here. Ah:



Here's the trend line for those drops. It drops from a bit higher than 2000 units per batch to about 1690 per batch. Does that prove crystals are dropping fewer units over time? An unscrupuloous person might try to make that case. But I happen to know the statistical average for arena crystals is 1875 per million BC (regardless of crystal type) so my very first recorded numbers just happened to be significantly above average. Meanwhile my most recent opening was significantly below average. Had I generated this chart just a couple weeks ago the difference would have been much lower, and without that initial lucky opening at the start it would have been even closer. Without those two data points, it looks like:



The difference is now is the average trendline goes from 1931 to about 1821. But even that is potentially cherry picking: I shouldn't be throwing any data points away, even if they seem out of the ordinary to me. Rather, I should analyze the data to see what impact precisely when I do the analysis affects the overall averages, using moving averages:



Now I can see that the ten batch moving average seems to randomly move up and down, roughly around an average just under 2000 units per batch. In fact, highlighting the 1875 unit average:



The moving average clearly moves above and below it, spending about as much time above as below. So when I do the analysis matters. If I'm careless or unscrupulous, I can just wait for the data to say what I want it to say, because being random it will eventually say anything I want it to say, if I focus on what I want and ignore the rest.

Data is great. But data tends not to convince the conspiratorial. They are the very people convinced the data will prove them right if only they look at it in just the right way, and eventually they find a way.