File spoon-archives/bhaskar.archive/bhaskar_1998/bhaskar.9803, message 10

Date: Fri, 6 Mar 1998 00:51:41 -0700
Subject: BHA: Realist Methods

One caveat about regression is that only the dependent
variable has an error term associated with it.  The
explanatory variables are assumed to be effective as
observed, i.e., here the empirical and the real is
conflated.  Once you associate error terms also with the
explanatory variables, you get so-called latent variable
models.  But I am hesitant whether a depth realist can
endorse those in their present form.  Their theory is rather
murky, because basically you have lost identification, and
instead of accepting the fact that the empirical surface
will usually not fully reveal what is really going on, these
models try to use tricks or magic to supply the missing
information.  The alternative, which does exist but only
marginally, is to use the surface as a starting point
for a systematic survey of all the underlying relationships
which can give rise to the observed data.

Hans Ehrbar.

     --- from list ---


Driftline Main Page


Display software: ArchTracker © Malgosia Askanas, 2000-2005