| Recently ran a DBM analysis on two groups (controls and patients) using the
| contrast 1 0; 0 1 with age or sex as the covariates of interest and
| uninteresting covariates. I have one set-up question: is it possible to run
| the analysis without specifying uninteresting groups weights and/or
| covariates?
Yes. You need to enter an empty vector (which ca be done by just typing a
space and hitting return).
| Regarding the results (of which I have a bunch of questions), I understand
| that the mean difference display shows whether or not there is a significant
| difference in the mean vectors of the displacement of the deformations
| between the two groups (right?), but what do the arrows and the chosen
| coordinate indicate in the supplied image? Specifically, does the size of
| the arrow imply a larger displacement between the mean vectors of the two
| groups and does the coordinate given coincide with the region that had the
| largest difference between the two groups? Similarly, can one conclude that
| all of the vectors shown represent significant differences in shape or only
| those vectors with (and please forgive the ambiguity) 'substantial' length?
The <Run Stats> option gives you a P value representing the significance of
the difference.
If you show the results as a canonical variate representation, then what you
get is a caracature of a difference among the groups. No significance can be
attributed to any of the individual arrows, but a longer arrow generally means
that the displacement at that point is a more robustly occurring feature than a
shorter one.
It would require a great deal of work in order to make spm_dbm more
intuitive and easier to use - especially as the results of canonical variates
analysis involves a 3D vector field. There is also a bit of necessary
background reading on multivariate stats before anything can be done.
| Questions regarding the graph: The x-axis lists the covariate of interest
| (e.g. age), but the range seems constrained form ~ -1 to .5 while the ages
| in our groups are much more distributed - and (shockingly) not negative -
| how should this be interpreted?
The covariate of interest is first centred.
| Also the y-axis has a large range of values
| (raised to a power of 3) with no label... what does that range of numbers
| correlate with?
A canonical variates analysis is done on the first few principal components
of the data, after orthogonalising both the design matrix and the effects
of interest with respect to the confounds.
The plot is of the linear combination of the principal components of
the data that has the maximum correlation with another linear combination
of the effects of interest part of the design matrix.
You may find it useful to check out something like:
W. J. Krzanowski "Principles of Multivariate Analysis - A Users Perspective"
Oxford 1988.
What spm_dbm.m does is described in:
J. Ashburner, C. Hutton, R.S.J. Frackowiak, I. Johnsrude, C. Price and
K. J. Friston. " Identifying Global Anatomical Differences: Deformation-Based
Morphometry". Human Brain Mapping 6(5):348-357 (1998)
| The points in the graph that is generated are principally
| aligned into two vertical lines that parallel one another. Since I cannot
| seem to resolve the number of points that comprise either of the "lines",
| what do they represent (e.g. the number principal components, number of
| subjects, etc...)?
The plot is of the canonical variate versus the specified effect of interest.
If there are two vertical lines, I would guess that the effects of interest
divide the data into two groups.
|
| Finally (for now), how does one setup the canonical correlation? The menu
| asks for a selection to 'plot which vector' with 3 choices, followed by a
| choice of 'against which covariate of', also with 3 choices. This is
| confusing to me as I only had 2 covariates (age and sex) and I don't really
| know what the vectors in first option are.
"Interesting group weights" and "Interesting covariates" all enter as
effects of interest. The group weights come first before the covariates.
| I am under the impression that
| this particular analysis will show which of the covariates had the greatest
| effect on the dependent variables (in this case the deformations), in
| contrast to the 'difference representation' which basically stated whether
| or not there was an overall difference. But I am not sure how to set it up
| to answer that question.
The difference is a representation of the magnitude of the effect of interest
after removing confounding effects, whereas the Canonical variate
representation is a characature takes into account the variance-covariance
of the residuals.
Best regards,
-John
|