[R] Normal tests disagree?

[GlenB]

And now some advice:

I'm of the opinion that one should only in very rare circumstances be doing
hypothesis testing of distributions. For example, the most common reason
people give for doing a hypothesis test of the
distribution of a sample is that they want to check the appropriateness of
the assumptions of 
some other procedure.

The problem there is that a hypothesis test answers entirely the wrong
question.

Doing two hypothesis tests of the same hypothesis just answers the wrong
question twice ...

[There are circumstances where I think it can be appropriate to do
hypothesis tests, but they're not very common. In that situation, you either
have specific kinds of deviations from the assumed distribution that you're
interested in - in which case you use a test statistic that's sensitive to
them - or you don't, in which case you pick a good omnibus test like
Anderson-Darling or Shapiro-Wilk (in the case of normality) ]

-- 
View this message in context: http://n4.nabble.com/Normal-tests-disagree-tp932615p948201.html
Sent from the R help mailing list archive at Nabble.com.