When I do group analysis based on six individual subject’s results using 3dDeconvolve, the uncorrected p value should always set over 0.05 so we can see the activation. The q value is always close to 1 like 0.98, 0.99… Is uncorrected p value 0.05 enough to say our results significant? Do we need to care about the q value? Is there anyway to lower the q value to make our results more sense?
You might look at some of the educational resources. There are video recordings of recent bootcamps held at NIH.
The short answers are:
The p-value uncorrected isn’t enough to get published (and for good reason). There are a few AFNI publications about responsible use of statistics that are relevant (1[/url], [url=https://www.biorxiv.org/content/early/2018/04/28/308643]2).
Your options are to use FDR (q-values) or Cluster Correction. If you do cluster correction, you don’t “care” about q-values.
To lower your q-values you need to activate more voxels. If you tell us more about your design we can hypothesize some reasons that it’s high… But often people use one of the cluster correction approaches. You can read about the ways of doing so in AFNI in this paper[/url]. And also by looking at the help of [url=https://afni.nimh.nih.gov/pub/dist/doc/program_help/3dttest++.html]3dttest++.
When you mention Cluster Correction are you referring to the “clusterize” button on the GUI, or something else?
The “Clusterize” button can show you the significant clusters on the group maps. It’s a combination of setting the correct p-value and looking for clusters of a particular size. The papers I mentioned above have more details.