Thanks, Mark! That makes sense.
Interestingly, via simple literature review, it seems that 2.3 is a commonly accepted min. value in the papers using FSL. Thus, I guess although it is arbitrary, one possibly should set it at least 2.3.
Thanks again!
Daniel
> On May 6, 2014, at 2:39 AM, "Mark Jenkinson" <[log in to unmask]> wrote:
>
> Hi,
>
> This Z value is the arbitrary cluster-forming threshold and the corresponding p-values are uncorrected, voxelwise p-values and are not related to the cluster-based multiple-comparison-corrected p-values derived later. The choice of this value is completely arbitrary (up to some limits). See the FSL Course lectures, wiki, and GRF papers for more information.
>
> All the best,
> Mark
>
>
>> On 6 May 2014, at 05:54, "Yang, Daniel" <[log in to unmask]> wrote:
>>
>> Dear FSLers,
>>
>> When specifying a Z value for cluster thresholding, the default value is 2.3 (which equals to p<.01, one-sided). I am wondering what the theoretical basis is for this default value. How is it compared to using Z=1.96 (which equals to p<.05, two-sided)? Or is it the case that two-sided does not make sense in this context of finding contiguous clusters?
>>
>> Thanks!
>> Daniel
|