• silasmariner
    link
    fedilink
    arrow-up
    6
    ·
    edit-2
    7 days ago

    It’s not P(E) x 1825 that makes no sense because the probabilities are independent. It’s 1 - (1 - P(E))^1825 – your maths would imply that a 1/1000 chance had a 100% probability of happening after 1000 attempts, which is not how independent probabilities work.

    Anyway I’d probably go for it if it were, like, 1/100,000, giving a probability of the bad thing happening once over 5 years at just under 2%

    • silasmariner
      link
      fedilink
      arrow-up
      2
      ·
      edit-2
      7 days ago

      I just remembered I hadn’t tried to write anything in J for like a year so I had a go at seeing how far the approximation was from the real values and got:

      (-.@^&1825@-. % *&1825)@(0.1&^)"0 i. 10
      0.000547945 0.00547945 0.0547945 0.459687 0.914098 0.990935 0.999089 0.999909 0.999991 0.999999
      

      So it looks like the ratio probably converges as you increase n (these results are for 1, 1/10, 1/100 etc) but that x 1825 is always higher and starts off wildly off. But these numbers are probably mad squiffy because floating point yadda yadda.

      Edit again: oh yeah. Binomial expansion. Once the negation of your probability gets very close to 1 the initial coefficient will absolutely dominate. Proof left for reader.