Hello allstat,
Somebody ask to do multiple comparison for chromosome aberration rate
between 10 groups. The rates are all under 2.0%(the smallest is
0.9%),so i take it as Poission distribution. 'cause the positive
number is not large enough(most of them less than 20), I think the
approximate Z test can't work. For chi-square, it seems the expected
frequency is small too, so i would like to perform Fisher's exact
probability for each fourfold table. If all the positive numbers of
every group are greater than 5, may i use binomial-based approximate
Z test too?( since the Poission distribution is an asympototic for
binomial distribution, shall i use it directly but not an
approximate).
If fisher or Z test can do, then how to decide the nominal level for
each comparison? If all the paired comparison are of interested, then
45 times test will be perform. If i set family or overall level as
0.05, using conservative Bonferroni cutoff value, the nominal alpha is
about 0.0001, then the test is hard to sig.
So, would mind telling me how to do that? Increasing overall level?
Thanks for any suggestion or comment in advance.
Best regards,
wangtong mailto:[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|