Permutation testing should be considered for smaller data sets (e.g., less than a dozen scans), or for data where the assumption of normality is questionable (e.g., you've taken the maximum absolute value across a series of Z maps, but keeping the original sign). Under these circumstances, it may offer superior power to the parametric statistics which are otherwise routinely used.

(1) Download the data set from here (the link entitled "rfx.tar.gz") or here.

(2) Place the file rfx.tar.gz in some directory where you have write permissions, then run this command:

This will create a new directory named rfx-data, which will contain con files numbered from 0006 to 0017.

(3) In a tcsh shell, type the following command to add the SnPM directory to the front of your MATLAB search path:

(You can add this command to your .cshrc file so that it is automatically run whenever you get into a tcsh shell.)

(4) In the rfx-data directory, invoke MATLAB:

(5) At the MATLAB prompt type:

(6) Go to page 77 of this lab book, and follow the tutorial from there. At one point, it will create a file named SnPMcfg.mat, and then in the next step ask you to select the file. This file will be written out to whatever directory you were in when you started MATLAB.

(7) Further information. The official SnPM page is here. And the paper by Nichols and Holmes is here.