Dear Mike,
At 15:18 02/08/99 +0100, Dr Mike Glabus wrote:
| In the example I gave, (two conditions (scans) plus one covariate
| measure (verbal fluency) for each) the program requests the covariate
| as a single matrix, which I've entered in the same order as the scans
| [VF_cond1_s1, VF_cond2_s1...etc], but the program creates two
| covariate matrices, one for each condition called VF@cond1, [log in to unmask]
| My questions are:
|
| 1. What numeric value is the program deriving for these covariate
| matrices where the measure has no equivalent scan? Is it simply the
| value in the adjacent column with the sign changed? It's certainly
| non-zero from the figure of the GLM, but I'd like to check the exact
| value.
Should be zero - it is when I try it.
In (most of the) Design Matrix depictions in SPM, you can "surf" the design
matrix to find out the actual values in the design matrix. (By using the
middle & right mouse buttons you can also "surf" the corresponding image
filenames and parameter names, respectively. See the help for spm_DesRep.m
for details.)
In the results section, all the design information is available to you in
the base workspace. xX is the design structure, and you can get the design
matrix by typing:
xX.xKXs.X
(fMRI people may also want to look at the design matrix before temporal
filtering - this is in xX.X. In the absence of temporal filtering these two
matrices are the same.)
Type "spm_help spm_getSPM.m" for details of the data structures produced by
the results section.
| 2. Am I right with my interpretataion of the interactions [-1 1] and
| [1 -1]? Does this mean that the first [-1 1] is equivalent to a
| positive correlation, where the t-map will display voxels which
| increase in intensity between condition 1 and 2 in parallel with an
| increase in the covariate (VF) value from condition 1 to condition 2,
| and vice versa for the [1 -1] contrast?
Yes - if I'm understanding you correctly!
A contrast with weights [-1 1] for the two condition specific covariate
effects (& zero elsewhere) looks for voxels where rCBF response increases
more per unit increase in the covariate in condition 2 than condition 1.
| 3. > | other. Notice that this model implicitly also models the main
| > | effect of the covariate (tested for with 1 1). I'm not entirely
| clear how to interpret this (or it's mirror, [-1 -1]), mainly because
| I'm not 100% sure of what the program is inserting in the matrices.
| Presumably, this is not an interaction effect, but can you help me out
| here?
No, this is the main effect: What you have in this scenario is basically
two regression lines (in the absence of other terms in the model), one for
each condition, each with it's own intercept and slope. (Only the residual
variance is assumed the same.) The intercepts are the condition effects,
the slopes are the parameters corresponding to the two columns containing
the covariate.
[1 1] weights for the covariate effects is asking whether the average of
the slopes of the two regression lines is greater than zero, i.e., a main
effect.
Previously you effected the covariate by condition interaction manually, by
entering the covariate in two columns: First, you'd centre the covariate so
that it sums to zero within condition. The first column is simple the
centred covariate - the main effect. The second column was for the
interaction effect, and contained the covariate with the sign changed on
all the entries under the second condition. As Karl noted previously, this
model is exactly the same as the SPM99 model referred to above. Equivalent
contrasts for these two models are:
[ 1 0] & [ 1 1] +ve main effect
[-1 0] & [-1 -1] -ve main effect
[ 0 1] & [+1 -1] interaction: cond1 slope > cond2 slope
[ 0 -1] & [-1 +1] interaction: cond2 slope > cond1 slope
| Thank you for your continuing patience and assistance.
Hope this helps,
-andrew
+- Dr Andrew Holmes ------------------ mailto:[log in to unmask] -+
| ___ __ ___ Robertson Centre for Biostatistics |
| ( ,) / _)( ,) Boyd Orr Building, |
| ) \( (_ ) ,\ University Avenue, |
| (_)\_)\__)(___/ Glasgow. G12 8QQ Scotland, UK. |
+----------------------------------------- http://www.rcb.gla.ac.uk/ -+
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|