 |
 |
This means that no cluster reached significance using these two methods for multiple comparison correction. Best Regards, Donald McLaren, PhD On Mon, Aug 17, 2015 at 10:03 AM, Joelle Zimmermann < [log in to unmask]> wrote: > Hi Donald, > I sometimes find that FWEc and FDRc are infinity. What does this mean? Im > assuming this is a problem because it means that FWE at the cluster level > (FWEc), as well as false discover rate at cluster level (FDRc) are infinite. > Thanks, > Joelle > > On Wed, Aug 12, 2015 at 12:10 PM, Joelle Zimmermann < > [log in to unmask]> wrote: > >> Wow cool - this is helpful! Thanks Donald. >> >> On Wed, Aug 12, 2015 at 12:02 PM, MCLAREN, Donald < >> [log in to unmask]> wrote: >> >>> See below. >>> >>> On Wed, Aug 12, 2015 at 11:42 AM, Joelle Zimmermann < >>> [log in to unmask]> wrote: >>> >>>> Hi Donald - just to clarify: >>>> >>>> You need to enter the value specified for FWEc or FDRc in the GUI when >>>>> it asks for the cluster extent. >>>>> >>>> Do you mean when it asks for the 'extent threshold' in the results tab? >>>> >>> >>> >> Yes. >>> >>> >>>> And do you mean the specified FWEc or FDRc value I get from the bottom >>>> of the results table? >>>> >>> >>> >> Yes. >>> >>> >>>> I am surprised by this because I can only know which 'extent threshold' >>>> to put in the results tab only AFTER I get the results table (and the >>>> results table comes after the results tab). >>>> >>> >>> >> Yes. It just means you have to open the results twice to view only >>> the corrected values. I have written a non-GUI program to generate the >>> corrected images. It's called peak_nii and is available at >>> http://www.nitrc.org/projects/peak_nii >>> >>> >>>> >>>> >>>>> Technically, only FWEc would lead to controlling the cluster FWE rate; >>>>> however, controlling the cluster FDR is also an acceptable method for >>>>> multiple comparison correction. >>>>> >>>> That's good to know, thanks! >>>> >>>> >>>> >>>> >>>> >>>> >>>> >>>>> >>>>> >>>>> On Wed, Aug 12, 2015 at 11:28 AM, Joelle Zimmermann < >>>>> [log in to unmask]> wrote: >>>>> >>>>>> Hi Donald - I'm still unclear about how to control the FWE at the >>>>>> cluster level in the GUI. Do I need to change the FWEc/FDRc value then? >>>>>> >>>>>> Thanks, >>>>>> Joelle >>>>>> >>>>>> On Wed, Aug 12, 2015 at 10:59 AM, MCLAREN, Donald < >>>>>> [log in to unmask]> wrote: >>>>>> >>>>>>> See inline responses below. >>>>>>> >>>>>>> Best Regards, >>>>>>> Donald McLaren, PhD >>>>>>> >>>>>>> >>>>>>> On Wed, Aug 12, 2015 at 9:31 AM, Joelle Zimmermann < >>>>>>> [log in to unmask]> wrote: >>>>>>> >>>>>>>> Hi Donald - I know this may seem rather obvious, but could you tell >>>>>>>> me where in the bottom left of the table is the minimum cluster size that >>>>>>>> has a cluster corrected p-value less than 0.05? >>>>>>>> >>>>>>> >>>>>>> >> FWEc: # and FDRc:# list the minimum cluster extent for this >>>>>>> contrast. It the last line of the page. >>>>>>> >>>>>>> >>>>>>> >>>>>>>> Is this the same thing as the 'extent threshold k = 0 voxels' that >>>>>>>> I see at the bottom left of the table? I expect this is because I put in 0 >>>>>>>> under the extent threshold in the results tab. >>>>>>>> >>>>>>> >>>>>>> >> No. The extent threshold k = # is the the # you provided SPM. >>>>>>> >>>>>>>> >>>>>>>> Also, how can I cluster FDR from the table? And what is FDR? I only >>>>>>>> see in the results table a FDRp value and FDRc? >>>>>>>> >>>>>>> >>>>>>> >> FDR is the false discovery rate. FDRc is the cluster-corrected >>>>>>> FDR value for the cluster, FDRp is the peak-corrected (e.g. topological) >>>>>>> FDR value. >>>>>>> >>>>>>> >>>>>>>> >>>>>>>> I'm attaching a screenshot of my results table for reference. >>>>>>>> >>>>>>>> Thanks and apologize for my extensive questioning :-P >>>>>>>> Joelle >>>>>>>> >>>>>>>> On Tue, Aug 11, 2015 at 2:58 PM, MCLAREN, Donald < >>>>>>>> [log in to unmask]> wrote: >>>>>>>> >>>>>>>>> In the results table, the bottom left of the table provides the >>>>>>>>> minimum cluster size that has a cluster corrected p-value less than 0.05. >>>>>>>>> You can then use this cluster size when you display the data and the >>>>>>>>> clusters showing will all be significant with the FWE control. You can also >>>>>>>>> use cluster FDR from the table (SPM) or 3dClustSim (afni) to determine the >>>>>>>>> cluster size to use. >>>>>>>>> >>>>>>>>> Best Regards, >>>>>>>>> Donald McLaren, PhD >>>>>>>>> >>>>>>>>> >>>>>>>>> On Tue, Aug 11, 2015 at 12:22 PM, Joelle Zimmermann < >>>>>>>>> [log in to unmask]> wrote: >>>>>>>>> >>>>>>>>>> Thanks Donald. How can I control the FWE at the cluster level in >>>>>>>>>> the GUI? - where can I find this option? >>>>>>>>>> >>>>>>>>>> On Tue, Aug 11, 2015 at 12:14 PM, MCLAREN, Donald < >>>>>>>>>> [log in to unmask]> wrote: >>>>>>>>>> >>>>>>>>>>> If you go with an uncorrected voxel criteria, which I suggest, >>>>>>>>>>> then you should correct for multiple comparisons by controlling the FWE at >>>>>>>>>>> the cluster level for publication. If its an intermediate analysis, >>>>>>>>>>> controlling the cluster-wise alpha is less of an issue. >>>>>>>>>>> >>>>>>>>>>> Best Regards, >>>>>>>>>>> Donald McLaren, PhD >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> On Tue, Aug 11, 2015 at 12:06 PM, Joelle Zimmermann < >>>>>>>>>>> [log in to unmask]> wrote: >>>>>>>>>>> >>>>>>>>>>>> Hi Donald, >>>>>>>>>>>> >>>>>>>>>>>> Good to hear from you again. Right, that's what I suspected >>>>>>>>>>>> about the first-level. I have 28 subjects in my dataset, and not seeing all >>>>>>>>>>>> that much at FWE. As long as there isn't a strict standard SPM people go >>>>>>>>>>>> with, I think .001 is better for me. >>>>>>>>>>>> >>>>>>>>>>>> Later I plan to split my 28 subjects into 2 groups so will have >>>>>>>>>>>> even fewer subjects. >>>>>>>>>>>> >>>>>>>>>>>> Thanks, >>>>>>>>>>>> Joelle >>>>>>>>>>>> >>>>>>>>>>>> On Tue, Aug 11, 2015 at 12:02 PM, MCLAREN, Donald < >>>>>>>>>>>> [log in to unmask]> wrote: >>>>>>>>>>>> >>>>>>>>>>>>> Statistical Thresholding at the first level has no impact on >>>>>>>>>>>>> the second-level data. >>>>>>>>>>>>> >>>>>>>>>>>>> For the second-level, it depends on how large the effect size >>>>>>>>>>>>> you expect to see and how many subjects you have in your dataset. >>>>>>>>>>>>> >>>>>>>>>>>>> Best Regards, >>>>>>>>>>>>> Donald McLaren, PhD >>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>>> On Tue, Aug 11, 2015 at 10:58 AM, Joelle Zimmermann < >>>>>>>>>>>>> [log in to unmask]> wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>> Hi Experts, >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'm running a first-level analysis per subject, and then >>>>>>>>>>>>>> forwarding the .con images into a second-level analysis. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'm wondering if there is a standard way to set the p-value >>>>>>>>>>>>>> adjustment to control in the results tab for the first-level and the >>>>>>>>>>>>>> second-level analysis. For instance, would you recommend FWE for the >>>>>>>>>>>>>> first-level single subject analyses, and .001 for the second-level analysis? >>>>>>>>>>>>>> >>>>>>>>>>>>>> Thanks, >>>>>>>>>>>>>> Joelle >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>> >>>>>>>>> >>>>>>>> >>>>>>> >>>>>> >>>>> >>>> >>> >> >