FDR and FWE

Colleagues, I performed a MEMA and received the results (files formats .nii.gz and .RxMat). However, when adjusting the FDR and q< 0.05, the necessary clusters disappear. I would like to apply a more gentle FWE correction in SPM, but for this issue I need a file in the .mat format. Has anyone encountered the task of making a file from .nii.gz and .RxMat to .mat for FWE correction in SPM? Or maybe there is another way to solve my issue of saving clusters for publishing results? Thank you very much in advance!

Hi-

It's a tricky thing. Results kind of should be results, whether they end up significant or not. There are many different multiple comparisons corrections and adjustments available (both FWE and FDR are large families, with different assumptions and implementations), but it is generally not encouraged to keep trying different methods until one leaves clusters. On a sidenote, I have never heard of the *.RxMat format.

A better way to go would be to consider this way of reporting results, where "below threshold" things can be shown, because they are certainly not nothing:

You can clusterize and have no "officially sanctioned" clusters, but still be able to display your results meaningfully. Importantly, the tiny differences between those various FDR and FWE methods become less important, behaving less likely random filters.

--pt

2 Likes

Thanks for the super thorough response and for sharing the article! It's got me really considering switching up how I present my findings!

PS: When conducting group analysis, an additional file is always generated for me and I mistakenly believed that it was similar to .mat in SPM (look at the attached photo).

Ah, I see. If you run 3dMEMA with verbosity on, then that *.RxMat is the X-matrix of the model that is output. I wasn't aware it would be created, because I have never used that option---learned something new today! It should just be a simple table file of parameters.

--pt