Hello AFNI experts, I’m trying to decide whether I need to run a sphericity correction (using the -SC flag) on my data in 3dMVM, but I would like to know whether it is necessary by first seeing what the p-values are for Mauchly’s test. According to a recent publication by Chen, Adleman, Saad, Leibenluft, and Cox (2014), the Mauchly test for sphericity correction can be readily incorporated and established into multivariate modeling. My question is, how can I incorporate the Mauchly test in AFNI?
Full citation: Chen, G., Adleman, N. E., Saad, Z. S., Leibenluft, E., & Cox, R. W. (2014). Applications of multivariate modeling to neuroimaging group analysis: a comprehensive alternative to univariate general linear model. Neuroimage, 99, 571-588.
Hi Gang, thank you for the reply. I have run 3dMVM both with -SC and without. For some more information on my study, it is an event-related design and I am analyzing it using CSPLIN to estimate the BOLD response. My within-subject variable is time, of which there are 11 total timepoints. When I look at the cluster table generated by the output from -SC and without -SC, I see fewer clusters that survive threshold with the -SC flag than those clusters that survived threshold when I did not use the -SC flag. However, I can only see the location of the cluster (x, y, z coordinates) and how many voxels constitute each cluster. It is my understanding that you only need to run sphericity correction if Mauchly’s test is statistically significant, indicating that the sphericity assumption has been violated. Is there a way for me to see the results from Mauchly’s test? I’m worried that using -SC will greatly reduce power, especially because I have a small sample size (51 participants across two groups, 25 in one group and 26 in another group). What do you suggest I do?
The statistical evidence is intrinsically continuous, so why would we have to stick to a specific cutoff value? The sphericity assumption could be violated, leading to inflated F-statistics. On the other hand, the false assumption of no common information shared across the brain results in inefficient modeling with the mass univariate approach, and excessively penalizes the statistical evidence. In addition, I would not consider a cluster threshold as an imprimatur for statistical rigor, but something loosely suggestive. Instead, I would feel much more convincing with the following than the meticulous focus on the threshold. Show the hard evidence of HDR: regardless of the cluster size, demonstrate the estimated response curve at the region with an HDR signature shape.
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.