Hi,

In my private blog I wrote a discussion based on very simple simulations about “voodoo correlation” raised by Ed Vul’s controversial paper. If you are interested, please see it and give me comments.

Major points are:

1. Less participants may cause “voodoo correlations” regardless whether with double-step analysis or not.
2. In order to avoid “voodoo correlations”, authors should love averaged values across all voxels in single ROIs as ROI data, not single values from single voxels with “best values/statistics” in the ROIs.

Please note my blog is only accessible from any other countries than Japan for some reasons.

(Note: the entry shown above was once lost because of a serious server crash. Today I re-wrote and edited: Aug 3, 2009)