On Mon, 30 Dec 2002 23:47:34 +0800, "K�J"
<[EMAIL PROTECTED]> wrote:

> I hear people saying that t-tests could be used to test:
> 
> 1. difference between two proportions

You can use a t-test to test a 0/1  variable (proportions).
Use the usual test, with the pooled variance estimate, 
if you are in the midst of reporting dozens of other results
using a t.    If you are trying to report a single result, folks
will be happier if you give them the usual 2x2  contingency
table, and the chi square value.

> 2. difference between two mean differences (i.e. between (M12 - M11)
> and (M22 - M21).

If that is change-score(1)  and change-score(2), where
you can actually compute the scores, then you could
make that the explicit case of comparing two scores:
and that is where you would use a t-test automatically,
based on the degrees of freedom for sample (1) + (2).

In the more general ANOVA context, that might be a
"special contrast".  It can be reported as t, using the 
DF  of the error term.  Some folks prefer to report tests 
on contrasts as  F-tests -- Once-upon-a-time, I ran 
into an innumerate reviewer who had a textbook with an 
example, and therefore wanted to insist that only F was
acceptable.  
 

> Could somebody please advice me on this or give me a reference to read
> about these tests?

I think this went unanswered when posted in sci.stat.consult
because  the <above> are two Unusual Cases,  and you give
no hint about whether you recognize that...   or just why
it is that you are interested in testing in general, or t-tests
in specific;  or where you have looked  already.

For information, or for references:
You could browse my stats-FAQ; also, the home pages
of other people who post answers to these stats groups.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to