>>> Wensui Liu <liuwensui@GMAIL.COM> 03/28/07 9:27 AM >>>
Contrary to you, I really have hard time to understand what Peter
said. In a mediation model, collinearity is almost unavoidable, which
will cause estimate bias of other regressors.
Sorry what I wrote was hard to understand.
In a mediation model, once you discover the mediation, you do NOT want
to include the mediating variable
Let's take David's example
> *MY* favorite mediation example is different. :-) It is an old data
> set. It reveals that there is a really high Pearson
> correlation between number of babies born in Netherlands towns
> and their stork populations. The mediating factor is, of course,
> the *size* of the town.
Now, suppose you actually wanted to model the number of babies born.
Once you KNOW that the stork population variable is being mediated by
the much more sensible town size variable, you would no longer include
stork population in the model, and this would get rid of the
||3/28/2007 4:38:42 PM