Hi Donal,
The 0 entries are fully part of the null-distribution, as a trivial example, if only the unpermuted statistics displayed non-zero sized clusters, and _all_ the permutated stats had zero clusters, then it is very likely the clusters in the unpermuted case are significant.
Kind Regards
Matthew
> Hi Matthew,
>
> I have another question regarding the text file output of maximum cluster size. Very many of the values are zero, which I guess occurs when you permute and find no cluster.
>
> Should I consider the zero values as part of the nut distribution I form, or should I build the null distribution only from the non-zero clusters found? This will obviously change the position of the 95th percentile, so just thought I’d check!
>
> Cheers,
> Donal
>
>
>> On 14 Mar 2016, at 15:56, Matthew Webster <[log in to unmask]> wrote:
>>
>> Hi Donal,
>> Correct - the file is the max stat value for each permutation. Note that if you happen to have saved the standard output from randomise, the critical (95%) output will be output at the end of each contrast reading something like:
>> Critical Value for: output_vox_corrp_fstat1 is: 519.41.
>>
>> Kind Regards
>> Matthew
>>
>>> Hi Matthew,
>>>
>>> Thanks for the info. Can I ask (ahead of rerunning randomise) what the format of the -N text file output is? Do I get the maximum cluster size value from each of my permutations?
>>>
>>> I guess I can then order them in size from smallest to largest, and move to the value appearing 95% of the way through in order to determine the threshold cluster size.
>>>
>>> Cheers,
>>> Donal
>>>
>>>
>>>> On 10 Mar 2016, at 23:16, Matthew Webster <[log in to unmask]> wrote:
>>>>
>>>> Hello Donal,
>>>> Cluster isn't the right tool to use here - you need to enable the '-N' option in randomise to obtain the null-distribution of cluster values for each stat - you can then find the minimum significant cluster size at any percentile.
>>>>
>>>> Hope this helps,
>>>> Kind Regards
>>>> Matthew
>>>>
>>>>> Dear experts,
>>>>>
>>>>> I am interested in knowing what the minimum significant cluster size is in my output from randomise when I use the “-c” option. I have tried using the "--minclustersize” argument, but nothing additional is provided in the output.
>>>>>
>>>>> I would like to compare the size of the cluster I actually observe to the minimum cluster size observable by chance given a FWE corrected p-value of < 0.05.
>>>>>
>>>>> Many thanks for your help,
>>>>> Donal
>>>>>
|