There shouldn't be a problem using SPM to process these images. In an ideal
world, with enough subjects, the two methods should be close to identical.
The main difference (besides the parametric/non-paramtric issue) is the way
the multiple comparison correction is made. This paper provides a nice
A comparison of random field theory and permutation methods for the
statistical analysis of MEG data. Pantazis D, Nichols TE, Baillet S, Leahy
Neuroimage. 2005 Apr 1;25(2):383-94.
Here is a paper that uses SPM on beamformer images :
Dynamic Modulation of Human Motor Activity When Observing Actions The
Journal of Neuroscience, 23 February 2011, 31(8):2792-2800;
In practice it does seem that one often gets different results with the two
methods. Besides the basic statistical assumptions and the multiple
comparison correction the way the variance is pooled is also different. So
there are many differences, but both methods are valid and the larger the
cohort the more similar one would expect the two methods to be.
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
Behalf Of Haris Styliadis
Sent: 07 March 2011 23:54
To: [log in to unmask]
Subject: [SPM] Beamformers / 2nd level analysis SPM / SnPM
I have performed a beamformer analysis (SAM) for an MEG study. Our
experimental design has two independent variables that have 2
levels(high-low). So I want to do a factorial 2x2 design. Originally, I
aimed to use SnPM but I found the plugins offered not enough for a 2x2
factorial design. Maybe it is due to lack of understanding SnPM's potential
in such a design. So I used the 2nd level statistics in SPM and created a
factorial design with two variables as factors. There it was possible to set
the levels of these factors. Finally, I got the main effects for the two
I have the following questions
1. I did not find studies on MEG using 2nd level analysis. Instead I found
MEG studies using SnPM. Does anyone know whether it is methodologically
correct to perform 2nd level analysis in in SPM for SAM images. Any papers?
2. SnPM is offered as an alternative to SPM. How do they correlate? Although
I could not perform a factorial design, should I expect same results for a
one sample test, or a paired t test implemented in both SnPM and SPM?
Any answer is kindly appreciated