question regarding Fishers exact test

[bgreen]

I am hoping for some advice regarding interpretation of  Fishers exact test
output.

I have some 3 x 2 contingency tables where the cell count is too small for
a Chi-square to be computed. In the usual instance where the cell counts
are large enough the residuals can be used to identify which associations
are statistically significant. In the example that follows, Fishers exact
test is employed, what criteria are used in in interpreting this reported
significant association - e.g is the association just between group 3 and
command.to.harm?


command.to.harm
group  n  y
     1    97  7
     2    28  5
     3    16  7


p-value = 0.006


Any assistance with this enquiry is appreciated,

regards

Bob Green

[Bob Schacht-3]

At 10:19 AM 10/11/2006, Bob Green wrote:

>I am hoping for some advice regarding interpretation of  Fishers exact test
>output.
>
>I have some 3 x 2 contingency tables where the cell count is too small for
>a Chi-square to be computed. In the usual instance where the cell counts
>are large enough the residuals can be used to identify which associations
>are statistically significant. In the example that follows, Fishers exact
>test is employed, what criteria are used in in interpreting this reported
>significant association - e.g is the association just between group 3 and
>command.to.harm?
>
>
>command.to.harm
>group  n  y
>     1    97  7
>     2    28  5
>     3    16  7
>
>
>p-value = 0.006
>
>
>Any assistance with this enquiry is appreciated,

... [show rest of quote]

Was your p-value obtained with the special Fisher's Exact procedure that
SPSS sells separately from the Basic module?
If not, do groups 2 & 3 have enough in common that distinguishes them from
group 1 to merit combination of groups 2 & 3? If so, you can combine them
which will retain the significant contrast with Group 1 and at the same
time eliminate the small cell expected values that you get with a 2x3. This
will collapse the table to a 2x2 so that the Basic Module version of the
Chi-square becomes available.

Also, I'm a bit bothered by the wording of your question, "what criteria
are used in in interpreting this reported
significant association - e.g is the association just between group 3 and
command.to.harm?"
I don't think this is the correct way to frame the question. Remember that
the flip side of the high percentage of "yes's" in Group 3 is the high
percentage of "No's" in Group 1. They are two sides of the same coin.

Bob Schacht

Robert M. Schacht, Ph.D. <[hidden email]>
Pacific Basin Rehabilitation Research & Training Center
1268 Young Street, Suite #204
Research Center, University of Hawaii
Honolulu, HI 96814