On 30 Apr 2004 06:17:12 -0700, [EMAIL PROTECTED] (Katie)
wrote:

> Hi, 
> 
> I think I've mastered the chi square test, but not too sure about its
> outcome and if I have used it appropriately.
> 
> I have 5 age groups with yes/no response data, is a 5x2 chi square
> test the test to choose to check for differences between age groups
> for the question?
> 
> I have found no statistical significant difference  by completeing
> this table(i.e. the chi value was not high enough for the p=0.5
> threshold).  Can I say the yes/no responses between age groups are not
> statistically different?
> 

You can say that "they are not different, by the conservative
test that compares 5 age groups" if the test fails to reject at
the .05  (not .5) level.  But that is a test with 4 degrees of 
freedom.

The 5x2  contingency test does not take the ordering into
account, and someone might easily argue that you should
have tested for a *trend* with age.  If the overall chisquared
is greater than 3.84, it might be true that there is a linear
trend across groups that would describe the sample, and
which would give a test with p< 0.05.   Can you also say
about the sample, "There is no apparent trend"?

SPSS (for one) provides Mantel's test for linear trend in the
contingency table.  It might be just as common to see the
test of trend provided as a t-test between Yes and No responses, 
performed with AgeGroup as the criterion.  Those two tests
will be practically identical.

-- 
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