Re: Comparing fractions (or proportions)

On Wed, 24 Jan 2007 17:18:47 EST, Jack Tomsky <jtomsky@xxxxxxxxxxxxx>
wrote:
[snip, a bunch]

The randomized Fisher test which uses the hypergeometric distribution and randomizes on the boundaries is UMP unbiased. That means that it is uniformly more powerful than any other test, unless the competing test is biased. (See Lehmann, pp. 140-143.)

Unless the other test has different assumptions.
Are you only randomizing the acceptance/rejection
"boundary", or are you (and Lehman) randomizing the
Marginal totals, to create a super-FET test?

The Barnard test does not assume two sets of fixed margins.
I think that some people still argue for the FET in any case,
but it is by no means universally accepted.
Don't beat up on Luis where he does not deserve it.

My stats-FAQ has references to articles in the Journal of
the Royal Statistical Society, in the 1980s, where Yates and
others discussed various issues of the 2x2 table.


[snip, bunch more]


--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.