I have a question about GLM in general although the context is SPM, this
may be elementary but I would appreciate some input. Basically, I wonder
if there is a rule of thumb for deciding how orthogonal the columns in a
design matrix need to be for the model to be robust in some sense? Put
in other words, if two covariates are dependent to some degree, how will
this affect the parameter estimates? A "naive" guess is that the effect
will be shared by both covariates, in which case it would only affect the
interpretation of the difference between the two covariates in
question. However, is it possible that the dependence would affect other
parameter estimates in the model? Basically, what I'm looking for is some
way of numerically assessing the reliability/robustness of a given design.
best regards,
Jonas Larsson, MSc
______________________________________________________________________________
Division of Human Brain Research
Department of Neuroscience Phone: +46-8-728 7298/7297/7784
Karolinska Institute Fax: +46-8-30 90 45
Berzelius väg 3, 5th floor
S-171 77 Stockholm
Sweden E-mail: [log in to unmask]
______________________________________________________________________________
|