In article
<a8ba98b0-b184-4943-8e04-335eacd30315@[EMAIL PROTECTED]
>,
<davidjones@[EMAIL PROTECTED]
> wrote:
>I am not sure how to phrase this question, so I shall ask it in 2
>different ways and then describe my problem in more detail.
>How can you define a reference range from sample set when the data is
>not obviously (or even obviously not) normally distributed? We are
>expected to produce a 95% range.
>How can you justify tests for normality, especially after a log
>transform? I can understand most statistical tests, where the
>objective is to reject the null hypothesis, but it seems that tests
>for normality take this process and do it the wrong way round. I
>cannot see how this can be justified.
It is almost impossible for a natural distribution to be
normal, and for most purposes, it is unim****tant whether
it is. The Central Limit Theorem, often quoted as an
argument for normality because of a "sum of a large
number of factors", only states that the distribution is
APPROXIMATELY normal, and the approximation is not that
fast. If, instead, the result is a product of a large
number of factors, the logarithm will be approximately
normal.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@[EMAIL PROTECTED]
Phone: (765)494-6054 FAX: (765)494-0558
|
