----- Email forwarded from Ulrich Moeller <[log in to unmask]> -----
Dear SPMers,
recently (5 March) we asked two questions concerning the problem
of how to assess the importance of a covariate.
Steven Grant posted a question (2 Apr and, again, 4 Apr) that touches
the same general topic (how to include a nuisance covariate).
Although covariate modelling seems to be common, there was
apparently no reply in the list. Therefore, we try it once more.
It would also help us to know if the answer is either trivial or a simple
answer cannot be given.
1. What are the appropriate (or commonly accepted) criteria in order
to decide whether a particular covariate should be included in a model
(single subject condition and covariates on the second level in this case)?
We think that some guidance can obtained by "assessing the relative
sizes of the parameter estimates for a given voxel", as discussed in a
mail (20 June 2000) from Stefan Kiebel.
For example, the covariate would be included if its parameter estimate
is of comparable size to the parameter estimate of the condition.
2. If this way of decision-making is appropriate, we face another problem.
The scaling of the covariate (e.g. seconds or milliseconds for reaction
times) will affect the size of the parameter estimate (as discussed by Karl
Friston on 26 Apr 2001).
How to preprocess the covariate so that its parameter estimate can be
directly compared to the parameter estimate of the condition (using SPM's
plot > contrast of parameter estimates > effects of interest)?
Marc Ligges
Ulrich Moeller
-----------------------------------------------------------------------------
---- Dr.-Ing. Ulrich Moeller Friedrich Schiller University Jena
Institute of Medical Statistics, Computer Sciences and Documentation,
Clinic for Child and Adolescents Psychiatry
Phone: +49-3641-936458 Philosophenweg 3/5, 07740 Jena, Germany
Fax: +49-3641-936581 Email: [log in to unmask]
-----------------------------------------------------------------------------
|