Re: Unsolved problems

On 10 Aug 2006 06:26:51 -0700, "Reef Fish"
<large_nassua_grouper@xxxxxxxxx> wrote:


valter.sundh@xxxxxxxxxxxxxxx wrote:
Richard Ulrich wrote:

Comparing r_s to r, somewhat loosely, is a fairly robust
check on the normality assumption.

That statement is categorically and unconditionally FALSE.


When the correlation is between two continuous variables in linear
scales, it makes comparably more sense to consider r_s a suitable
alternative to r on principal ground, and consider r_s a measure of
monotonic association.

That is correct. r_s is a measure of monotonic association rather
than linear association.


If r_s and r differ considerably it is probably
because one or both of them contain outliers,

The presence or absence of outliers is IRRELEVANT also.

Thus spouts a man whose experience does not include
much cross-sectional data analysis, and whose theory excludes
worrying about normality of variables.

Those of us who want to get meaning from relationships
are regularly concerned about outliers.

[snip]

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.