Dear All
Many thanks to those who gave suggestions as to how I should I should
calculate a sample size for a parallel group study, using information from
data which are highly skew.
For those who expressed an interest in this topic, here are some of the
ideas I received.
you can use my software NP-SIZE. You can download it from the page
http://www.geocities.com/degdemartini/
Now, only a dos version is available. Soon (July) the complete windows
version will be available (hope).
Please, tell me if you find NP-SIZE useful and any comment regarding
NP-SIZE.
An earlier version of NP-SIZE was presented in the international workshop on
statistical modelling in Graz (1999) and in August the final version will be
presented at COMPSTAT2000.
Best regards
Daniele De Martini
I think that the package PASS (http://www.ncss.com/pass.html) does sample
size calculations for Mann-Whitney assuming a variety of underlying
distributions including log-normal (somewhat skewed). There is a link to
the
program author on the web site so you could ask him.
Alternatively, you could always use a simulation approach. I can send some
references on this if you don't know how to go about it.
Good luck,
- Len Thomas
--
Len Thomas ([log in to unmask])
Research Unit for Wildlife Population Assessment
Mathematical Institute, University of St Andrews
Scotland
Trevor,
Most folk use the sample size for comparison of two independent samples from
approximately normal distributions (i.e. sample size for unpaired t) and add
a margin of safety (conservatism) when intending to use Mann-Whitney.
Permutational sample size calculation for Mann-Whitney would be overkill
when sample size calculation itself is inherently imprecise/speculative.
You might find StatsDirect software helpful - see http://www.statsdirect.com
for a free 10 day trial
Kind regards
Iain Buchan
CamCode & University of Cambridge
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|