Dear SPM’ers
I have a question regarding the calculation of correlations.
I have conducted a study in which participants had rCBF measured with
SPECT on two different occasions: after receiving a placebo and after
receiving an active treatment. I first did a “multi-subj: conditions &
covariates” analysis (without introducing any covariate yet) and found
specific regions of increased blood flow associated with the
administration of the active drug.
Subsequently, I looked for correlations between the blood flow increases
(active treatment – placebo) and measures of subjective effects and blood
hormone levels. Each of these measures was introduced separately as a
covariate again in a “multi-subj: conditions & covariates” model together
with the two scan conditions (placebo/active treatment). Correlations
between blood flow increases and subjective or neuroendocrine variable
increases were calculated with a [0,0,1] contrast. However, the
correlations found usually made no sense from a physiological point of
view, usually appearing in areas that had previouly not shown rCBF
increases associated with drug administration.
I have more recently tried a “multi-sub: covariates only” model and the
picture that emerges is now perfectly coherent: correlations appear in
areas which showed significant rCBF increases in the absence of covariates
and the different covariates correlate with different specific brain areas
previouly shown by other researchers to be involved in the specific
subjective effect or neuroendocrine response measured.
I would be greatful if someone could indicate me why such different
results appear in each case and if the “covariates only” approach is
adequeate for this kind of experimental design. Also if I should expect
reviewers to criticize the use of this model instead of the “multi-subj:
conditions and covariates” in my search for correlations.
Jordi
Department of Neuropsychology
Otto-von-Guericke University
Magdeburg
Germany
|