On Sun, 30 Mar 2008 15:24:42 GMT, Paul Rubin <rubin@[EMAIL PROTECTED]
> wrote:
[snip, preceding]
>
> I believe there's a "metatheorem" that with a big enough sample you can
> reject just about any simply null hypothesis. In this case your null is
[snip, rest]
No, that statement of what pertains is nonsensical.
It would never apply to the process of randomization.
The relevant "meta-theorem" might say that in an
observational study, a large enough N will result in
apparent statistical significance for some simple tests,
because other underlying sources of dependency or bias
were not accounted for.
The problem is weak design combined with large N.
We can get away with a lot of approximations to
"meeting all the assumptions" when the N is small,
- that is, not thousands - because the usual biases are
small.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
|
