On Wed, 22 Dec 1999 11:27:21 +0100, Miguel Verdu <[EMAIL PROTECTED]>
wrote:
> Why log-linear models can reflect a significant partial association
> between factors when none of the cells are significant?.
- I'm not 100% sure about your question, but I think the answer is,
"That is the nature of *partial* association."
In a measure or test of partial association, you look at what is
*added* by the variable when it is taken *last*. If two variables
are similar, then neither will add much -- even though they may each
serve excellently, if you look at each separately.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
