On Mon, 5 Feb 2001 19:26:46 +0900, "rjkim" <[EMAIL PROTECTED]>
wrote:
> hi, all.
>
> Isn't a chi-squre test inherently a 'one-sided sig. test'?
The chi-square is to F as the normal z is to t.
t^2 (xx degrees of freedom) equals F (1, xx degrees of freedom).
z^2 equals chi-squared.
F is "two-tailed" and t can be "one-tailed."
Similarly, for the other two.
>
> I just read a paper claiming that it uses a 'one-sided' rather than
> 'two-sided' test. So it regards a chi-square value of 3.3 (df=1) is
> significant at the level of 0.05. (As you all know, the critical value
> of chi-square (df=1) at the alpha of 0.05 is 3.84.)
> [ ... ]
Okay, this is like computing an F-test but then interpreting as if
you had done the t-test. Technically, it is legitimate (assuming that
you could have done the t-test). I may have done the same thing in
the past, but it is not something to admire.
The excuse, in my opinion, is that tossing in one interpretation
(this) places less strain on the discussion than introducing one more
variety of test into a report. That is, the author might mention
this, offhandedly, in a particular context; this should not be the
major test of hypothesis, nor anything else major.
--
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/
=================================================================
