> I'm very puzzled by results of an SPM99 covariate analysis that are
> making me question my basic understanding of what a covariate analysis
...
> that made sense in terms of the relevant literature. The problem is that
> when we did a covariate only analysis using the symptom severity rating,
> the activations and deactivations seemed to reverse with each other. Since
...
Speaking about covariate analysis broadly (I lack expertise in how SPM
implements it): one controversy in the general stat literature is
whether covariates (as in ANCOVA) can legitimately be used when levels
of the independent variable (IV) differ on the covariate. Some say that
that's the most useful situation in which to use a covariate. But the
overwhelming sentiment in the stat literature is that in general an
ANCOVA-type covariate can't be used when levels of the IV differ on the
covariate. One way to understand this is that removing variance shared
between the covariate and the IV can fundamentally alter the (residual)
IV, so that it no longer validly measures what it used to. Applying this
to the above case, it may be that the reversal seen is a due to a highly
correlated covariate making mischief with the IV, so that the vestigial
variance being analyzed as the IV has little to do with the IV as it was
analyzed without the covariate variance removed.
Greg Miller, U. Illinois, Champaign
|