A new discussion of important statistical practice

A discussion of how and when to use two-sided and one-sided tests:
These are some pretty fundamental aspects of statistical tests. Please read and spread!

A few highlights:

"Here, we first aim to clarify between cases where one-sided or two-sided testing is appropriate. The general rule can actually
be stated succinctly: If the investigator can justify focusing on one side (direction) based on prior knowledge, then they should
explicitly state their reasoning and decision, and perform their individual one-sided test; otherwise, two-sided testing should be
conducted.  This can be applied straightforwardly to voxelwise, ROI-based or other study approaches, as described below."

"...performing a pair of one-sided tests at the significance level of A is equivalent to performing two-sided testing at the
significance level of 2A, and therefore the expected FPR is actually twice the stated rate."  

"This sidedness issue is not a matter of interpretation about effect directionality but simply a black-and-white, quantitative
fact about FPR controllability."  

Oh, and there’s some code posted online in association with this draft that demonstrates AFNI tools for group analysis + clustering, including the new 3dClusterize (which replaces the older 3dclust in a hopefully more syntactically friendly way). But ya gotta read the whoooole article to find it!


The published version now: