Sun -
> Dear Vladimir,
>
> Sorry for troubling you with this stupid question, but I am really not familiar with SPM's algorithm. I am doing ERP sensor space analysis of 2*2 design. By using 2nd level full factorial analysis to each subject, I got the beta and contrast files for each condition, however, I found that the contrast values are not consistent with the amplitude values gotten from averaging. Follows are values I checked at the same spacial location and time point from my four averaged conditions (after converting to images) and the corresponding contrasts:
> type_1: 3.7094 contrast_1: -0.8063
> type_2: 3.7921 contrast_2: -0.2722
> type_3: 2.8490 contrast_3: -0.5204
> type_4: 3.2480 contrast_4: -0.5435
>
> I got the averaged file by doing averaging within each session and then doing grandmean across sessions. The amplitude values from the averaged files made by SPM procedure are almost as same as those from Scan4.3 procedure (my data were collected by Neuroscan system). I am afraid that SPM statistical analysis will show me some results different from my traditional data processing by using Scan4.3.
>
Are the "contrast_*" values above from SPM's con*img's after fitting the
2nd-level model? Then their values will depend on the nature of the
model (e.g, whether you included extra columns to model subject effects;
if you did, then the values you report above will have the mean over
subjects of each condition removed). I remain somewhat confused however
since I would have thought that the ordinal relationship across
conditions would remain - did you additionally use non-sphericity
correction in the 2nd-level model - this might change relative order?
Anyway, it might be worth sending the precise model you used (eg the
SPM.mat file or batch script you used). Another thing you can do is plot
the data for a given voxel from within SPM's Results window, which will
give you a variable "y" in the Matlab workspace, and average this over
subjects to compare with your sensor results above.
If, on the other hand, the "contrast_*" values above are taken directly
from the scalp-time images written by SPM (ie, the data that went into
the 2nd-level model), then they should be more similar. Note that the
values in these images at a location close to a sensor will not be
identical to the original sensor data, because of the spatial
interpolation used, but they should be similar than above, I would have
thought (unless your sensor positions are incorrect). For example, they
are unlikely to differ by an order of magnitude (unless your D.units was
not defined correctly?). In this case, perhaps you could send one of the
original subject SPM averaged files (off-list).
> Another question is that whether I could input the averaged data (after converting to images and smoothing) of each condition per subject into the 2nd level analysis for cross subject analysis. I ask this because I have already made some averaged files by using traditional preprocessing by Scan4.3, and I do not want to repeat the preprocessing by SPM again. I am sorry to say that SPM is really slow, especially when I have to transfer each trial into an image.
>
So the only difference from what you did above is that you now propose
to do the preprocessing up to the stage of averaging within a subject
within a different package, then convert to SPM and write out images? If
so, that is absolutely fine!
R
--
-------------------------------------------------------
Dr Richard Henson
MRC Cognition & Brain Sciences Unit
15 Chaucer Road
Cambridge
CB2 7EF, UK
Office: +44 (0)1223 355 294 x522
Mob: +44 (0)794 1377 345
Fax: +44 (0)1223 359 062
http://www.mrc-cbu.cam.ac.uk/people/rik.henson/personal
-------------------------------------------------------
|