Hi All,
If you have the time I would appreciate your thoughts on this before I
go 'stomping' off to 'argue' with someone; if you would be kind enough
to support or refute my logic I would appreciate it.
A collegue has undertaken a hierarchical regression analysis. Using a
collection (I just mean a group, or selection of) of independent
variables (lets call them collection A), then adding a second collection
of variables (collection B). B does not improve the model fit above
that obtained from A alone - his interpretation - that any effect that B
variables may have is mediated through variables in the A collection
(there is modest degree of correlation between all these variables).
My thoughts are, if he had put the B variables into the model first - it
is possible that adding the A variables would have not produced any
improvement (I suppose that it might, but it might not have) in which
case by his logic he would have to conclude that any effect the A
variables have would be mediated through the B variables.
I guess that I also feel that A and B variables - given their degree of
correlation may in fact be indicitive of other 'latent' variables anyway.
I just feel uncomfortable with his initial interpretation that since
adding the B variables doesn't improve the fit, they have their impact
'via' the A variables.
If you could share your thoughts on this with me, I'd appreciate it.
Best Regards
Graham
--
Dr G.S.Clarke
Lecturer in Physiology & Biometry
Faculty of Health Studies
University of Wales, Bangor
Fron Heulog
Ffriddoedd Road
Bangor
Gwynedd LL57 2EF
Tel: 01248 383157
e-mail: [log in to unmask]
|