In most neuroimaging, people use p<.01 (sig threshold > 2). It is more important to have a more stringent threshold when using gaussian random fields because the assumptions built into it. Since we use simulations directly, we don't have problems with those assumptions and so I think you can go down to p<.05 (sig>1.3). Other than that, there are not good guidelines.
On 4/13/16 8:36 AM, Clara Kühn wrote: > Hi Doug, > > thanks for your reply. It made things a lot clearer. I totally understand > that you're probably receiving more than one cry for help per day. > > What would you say are the conventions for picking a threshold for analyses > on structural data? > > Cheers, Clara > > ----- Ursprüngliche Mail ----- > Von: "Douglas N Greve" <[email protected]> > An: [email protected] > Gesendet: Mittwoch, 13. April 2016 00:15:41 > Betreff: Re: [Freesurfer] REPOST: Monte Carlo correction in QDEC > > sorry, I could have sworn that I answered this one > > On 04/12/2016 08:57 AM, Clara Kühn wrote: >> Dear FreeSurfer experts, >> >> for the analysis in QDEC I created my own Monte Carlo correction. >> My questions relate to the threshold option. >> >> 1. Would I use neg if I have mostly blue clusters in the QDEC display and >> pos if I have mostly red clusters? > No, you would use neg when you have an apriori assumption that your > effect is going to be negative. Once you look at the results, it is no > long apriori >> 2. When do I use abs? > If you do not have an apriori assumption about the sign of the effect >> 3. I compared the neg option at different thresholds (1.3, 2.0 and 2.3) and >> I get different clusters: >> 1.3 (=.05) >> # ClusterNo Max VtxMax Size(mm^2) MNIX MNIY MNIZ CWP CWPLow >> CWPHi NVtxs Annot >> 1 -2.669 76591 1148.00 -6.7 14.5 62.7 0.00880 >> 0.00760 0.01000 1600 superiorfrontal >> 2 -2.443 91912 2128.78 -10.3 55.3 -23.5 0.00010 >> 0.00000 0.00020 2971 medialorbitofrontal >> >> 2.0 (=.01) >> # ClusterNo Max VtxMax Size(mm^2) MNIX MNIY MNIZ CWP CWPLow >> CWPHi NVtxs Annot >> 1 -3.289 92754 420.03 -36.5 57.0 -21.0 0.02140 >> 0.01960 0.02330 571 parsorbitalis >> 2 -2.669 76591 428.11 -6.7 14.5 62.7 0.01940 >> 0.01760 0.02120 604 superiorfrontal >> >> 2.3 (=.005) >> # ClusterNo Max VtxMax Size(mm^2) MNIX MNIY MNIZ CWP CWPLow >> CWPHi NVtxs Annot >> 1 -3.289 92754 314.80 -36.5 57.0 -21.0 0.01900 >> 0.01730 0.02080 410 parsorbitalis >> >> Technically, if parsorbitalis is significant at .01 and .005, shouldn't it >> also be significant at .05? I've compared the clusters in freeview and the >> superiorfrontal is the same cluster in both 1.3 and 2.0. The parsorbitalis >> and the medialorbitofrontal clusters are completely different, not even >> closely overlapping. > You are confusing the two types of thresholds. The voxel-wise threshold > that you are changing defines what is and is not a cluster. As you > change it clusters will change size. As you make it more liberal, you > make it more likely that you see a cluster of a certain size by chance > (ie, the p-value for the cluster gets worse). So as you make the > threshold more liberal, there are two competing effects: (1) the cluster > gets bigger, and (2) the p-value of a cluster of a given fixed size gets > worse. If the cluster size does not increase enough to overcome the > second effect, then the cluster p-value will get worse. It is just very > complicated. > >> Do you have any idea why that is and what it does differently? >> Thank you! >> Cheers, Clara >> _______________________________________________ >> Freesurfer mailing list >> [email protected] >> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer >> >> _______________________________________________ Freesurfer mailing list [email protected] https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
