At 12:15 28/10/99 -0400, Hernando Ombao wrote:
| We have a PET study where we have 2 conditions (Awake and NREM) and
| 1 covariate PSQI. There is only one scan per condition per subject.
| We want to test for areas where:
| (1.) Increased activation from Awake to NREM is associated with
| higher PSQI scores;
| (2.) Increased activation from Awake to NREM is associated with
| lower PSQI scores;
| (3.) Decreased activation from Awake to NREM is associated with
| higher PSQI scores;
| (4.) Decreased activation from Awake to NREM is associated with
| lower PSQI scores.
(1) & (4) are the same: If activation is larger for subjects with higher
PSQI scores, then it must be smaller for subjects with lower PSQI scores.
Similarly, (2) & (3) are the same.
An SPM analysis here is basically looking at how the NREM - Awake
for each subject correlates with PSQI score. (1)&(4) correspond to a
positive correlation, (2)&(3) to a negative one.
Watch out for overall differences in global between the two conditions, and
be careful in interpretation if you use global normalisation when there is
| We looked at the SPM mailbase and found a suggestion given by Andrew
| Holmes (about August 4, 1998). That suggestion given was to (i.)
| center the covariates so that the "new scores" all add up to zero,
| and then (ii.) enter the covariates as [-a/2, a/2, -b/2, b/2, .....]
| where a is the "new score" for the first person, b is the new score
| for the second person, etc.
| Suppose that our scans are ordered (per subject) so that the first
| is the Awake and the second is NREM. Does the manner in which the
| covariates are entered above mean that the contrast [0 0 1] is
| testing for areas where **increased** activation from Awake to NREM
| is associated with higher PSQI scores and that the contrast [0 0 -1]
| is testing for areas where **increased activation** from Awake to
| NREM is associated with lower PSQI scores?
Yes: [0 0 1] tests (1)/(4), [0 0 -1] tests (2)/(3).
| Suppose that we want to test for areas that have decreased
| activation from Awake to NREM being associated with higher PSQI
| scores. Do we simply
| (i.) change the signs of the covariates (i.e., use [a/2, -a/2, ...])
| (ii.) form the contrast [0 0 1]?
| Is this equivalent to
| (i.) keeping the covariates the way they were: [-a/2, a/2,...]
| (ii.) but change the order of the scans -- this time NREM first
| and then Awake second and
| (iii.) form the contrast [0 0 1]?
| If my understanding is incorrect, please advise on how I should form
| the contrasts?
The two suggestions above are both equivalent to the keeping the scores and
scans the same and using contrast [0 0 -1].
Hope this helps,
+- Dr Andrew Holmes [log in to unmask]
| ___ __ __ Wellcome Department of Cognitive Neurology |
| ( _)( )( ) Functional Imaging Laboratory, Stats & |
| ) _) )( )(__ 12 Queen Square, Systems |
| (_) (__)(____) London. WC1N 3BG. England, UK |
+------------------------------------- http://www.fil.ion.ucl.ac.uk/ -+