This [putting a bound on the likely error] is true
only if you know the distribution function.

One of course never =knows= the distribution function. However, one can make reasonable estimates ("guesses", if you will) based on ones sampling methods and experience with previous polls (comparing past polls with elections for example). True,
your guess of the distribution function might be off, but if you do things right, it's probably close. How close? Well, there's a distribution function to describe that too.

One doesn't even know the sun will come up the next day, but as a working theory it seems to be more than satisfactory. Statistics is not mumbo jumbo, although it is true that real mumbo jumbo can be disguised as statistics.

Jose
--
Freedom. It seemed like a good idea at the time.
for Email, make the obvious change in the address.
(note to r.a.student: I only follow r.a.piloting and r.a.ifr, to which this is also posted)