Well, what did the Exact Test (don't call it Fisher's) say? Was it significant, too,
and to a similar p-level? - In that case, you might merely add, "Although some
Expected cell frequencies were small, the overall p-level was also < ...?> by
an Exact Test."
Personally, I always tried to avoid having as many as 11 categories,
because it is hard to discuss that many, and there are hardly ever so
many that are actually interesting, especially when you consider some
small observed Ns for the lesser categories. - You can lump small
numbers into "Other (too small for testing)" if that description fits the
case. Or, as Bruce W. suggests already, sometimes it is possible to
test a more precise hypothesis, with fewer degrees of freedom (and
thus, more power) by considering whether the categories are ordered,
or if there is some other principle for a better test.
By the way, the particular *problem* with the X^2 approximation when
Expected-frequencies are small is that a tiny value, as a divisor, may inflate the
contribution of cell unreasonably. You might look at the contributions cell-by-
cell to confirm that this has not happened to your data.
--
Rich Ulrich
> Date: Wed, 21 Sep 2011 01:42:26 -0700
> From:
[hidden email]> Subject: Chi -square test
> To:
[hidden email]>
> Dear all,
>
> I have 800 sample size and the contingency table forms 3 * 11. I am doing
> SPSS Chi square test statistics . Pearson chi-square test shows the results
> is significant . But at the bottom it shows 33.3% have expected count less
> than 5 and minimum expected count is .63. The table does have results of
> fisher exact test ? what should i do ?
>
Locked
