Originally posted by uzless
I added the two together precisely because the question is non-linear. As you get more accuracy, the amount of boxes does not initially increase in a linear fashion so you need to build upon your previous answers.
What i find interesting is the correlation between the following:
50% equals 7 boxes
90% equals 13 boxes
99.99% equals 37 boxes
...[text shortened]... .9999% you just add nine more boxes etc etc
Anyone of you non-linear boys care to explain?
I'll have a go at that. The asymptotic behaviour is like a power law.
Using my notation from above, for a large enough number of boxes:
P(n, 1) << P(n, 2) << P(n, 3) << P(n, 4)
So as an approximation we can write:
P(n, 1) = P(n, 2) = 0
P(n, 3) = a (where a << 1)
P(n, 4) = 1 - a
Then
P(n + 1, 4) = (1 - a) + a/4 = 1 - 3a/4
P(n + 2, 4) = 1 - (3/4)^2.a
etc.
And (3/4)^8 is very close to 0.1. So roughly every 8 boxes you move up an order of magnitude in your confidence.
99.9 -> 29
99.99 -> 37
99.999 -> 45
99.9999 -> 53
99.99999 -> 61
99.999999 -> 69 etc