I am not sure that I understand what is intended here. So I may be
off track.
On 24 Aug 2000 13:30:02 -0700, [EMAIL PROTECTED] (dennis roberts) wrote:
> in some cases ... chi square test statistics require using ONLY 1 tail ...
> of the relevant chi square distribution ... but some cases require using a
> two tailed approach ...
I have had a few instances where I looked at the left-tail (near zero)
rather than the right-tail of chi square. There were very special
circumstances, requiring careful thought as to whether the test was
fair and justified. Off hand, I don't recall *ever* wanting to
combine those left and right tails for a "two tailed approach." They
address separate questions, don't they?
>
> same can be said of F test statistics ...
"The same can be said for F," that I don't see the point of
discussing 'two tails', with one exception for the F: The F-ratio
statistic that used to given with the t-test in SPSS. That was the
two-sample test for heterogeneity of variance that is conventionally
formed by putting the *larger* over the smaller variance. You may
treat (var) over (var) as two-tailed if you arbitrarily take (A) over
(B), or as a doubled p-level (over the tabled value) when you form the
test with (Larger) over (Smaller).
>
> can we say that about t test statistics? ( i am not talking the case where
> the researcher makes a one tail test out of the situation that is usually a
> 2 tailed test ... but where the natural test statistic calls for only using
> 1 end of the t distribution)
>
- Yes, the one-tailed test is the natural way to consider the t-test.
That is, in the way that I was taught statistics, it takes an extra
step to justify using the t-test as a two-tailed test. Almost always,
the weight is split evenly between the two ends, but that is a
convention.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
