Re: Hypothesis Testing: the TEST STATISTIC


Kevin E. Thorpe wrote:
This past week I was heavily involved in a workshop we
gave, so used the few minutes I had to try to post useful
replys to posters, rather than waste my words on one
who refuses to learn.

I wonder who might be referring to. That keeps them
guessing, wouldn't it? :-)

Now, just to bring this back to statistics. I did not read
your lecture in great detail, but from my quick skimming
of it, it appeared to follow the Neyman-Pearson theory.

That's the only theory I know, directly from textbooks.

I would be curious to here your thoughts on the
Fisherian approach.

Fishy. :)

As I understand it, Fisher's
approach involved testing a null hypothesis with no
explicit alternative (I know you'll correct me if I've
misunderstood).

I am talking your word for it about his approach. Unless
he always does a two-tailed test (which would not make
sense) , it also would not make sense for a hypothesis
tester NOT to know which tail of a one-tail test is the
alternative. In the problem of two proportions from two
indepedent populations, lets say p1 is the proportion of
students in non-honor courses exceeding the national
average in a national test after the course; and p2 is the
proportion of students having taken an honor course in
calculus and exceeded the national average when
taking the same national test.

The natural null hypothesis is p1=p2. A skeptic of the
efficacy of the honors course may use a two-tailed
alternative, but what would Fisher do if he wants to
test the hypothesis that the honor-course students have
a have a HIGHER proportion of above-average scores
in the national test? (Assuming your understanding of
Fisher's test of a null hypothesis is correct).

I may be old, but not old enough to be have been
burdened by Fisher's old-fishioned views (especially that
of fiducial intervals). :-)

-- Reef Fish Bob.

.