Hi Marco, I'm copying the list on this because this is a good question about an aspect that is not well documented.
The answer to the question has to do with the sign of the test. By default, mri_glmfit produces p-values that are for unsigned tests. The p-value for a signed test if half that of the signed test (ie, it is more significant). When the p-value is expressed as -log10(p), then the signed test will be 0.301 higher than the unsigned test. Eg, if you have a p-value of .01, the value in the sig.mgh file will be 2.0, but for a signed test it would be .005, making the value in the sig.mgh file 2.3.
When you run mri_glmfit-sim and specify a signed test (ie, --sim-sign pos or --sim-sign neg), it lowers the threshold to account for this. So below you specified 1.3 as the threshold and --sim-sign pos, so it lowered the threshold to 1.0.
hope this is clear
doug
Marco Loggia, PhD wrote:
Sorry Doug,
Below are directory and command line. Also attached is a screenshot showing a vertex located on a significant cluster, but below threshold in the sig.mgh.
Thanks! Marco
dir=/autofs/cluster/randy/2/lbp/freesurfer/patients+controls/analyses/lh.lbp +ctrl.session1.asl2-asl1_30sept10_sm7_03mm_masked_projfrac-max_P=0.05.glmdir /lbp_only_24sept10
pbsubmit -c "mri_glmfit-sim --glmdir lh.lbp+ctrl.session1.asl2-asl1_30sept10_sm7_03mm_masked_projfrac-max_P=0.05. glmdir --sim perm 10000 1.30103 perm.positive --sim-sign pos --overwrite"
-----Original Message----- From: Douglas N Greve [mailto:greve@nmr.mgh.harvard.edu] Sent: Monday, October 18, 2010 10:34 AM To: Marco Loggia, PhD Subject: Re: montecarlo
Hi Marco, can you send the command-line and directory that you ran it from? In general, it is always a good idea to send this ...
doug
Marco Loggia, PhD wrote:
Hello Doug,
I have one question about the montecarlo simulation.
When I compare the mc-z.positive.sig.cluster.mgh and the mc-z.positive.sig.masked.mgh, I notice that the significant clusters include also some vertices that are below the cluster forming threshold. Is this normal? Why is that?
Thanks,
Marco
Marco L. Loggia, PhD
Postdoctoral Fellow, Mass General Hospital & Brigham and Women's Hospital
Harvard Medical School
CNY Building 120, suite 101E
Charlestown, Massachusetts 02129
Phone: (617) 643-7267
Fax: (617) 643-7340 marco@nmr.mgh.harvard.edu