ops, df2 is the number of observations minus the rank of the design, not
just the rank of the design.
Am 05.02.14 11:24, schrieb Anderson M. Winkler:
> Hi Stefanie,
>
> There isn't a strict rule for this. For the t-statistic, many people
> like using 2.3, because that roughthly corresponds to a p-value of
> 0.01 if the sample isn't too small.
>
> If you decide to follow the same logic, you can just refer to the
> F-distribution and see what statistic gives a p-value of 0.01 for your
> design.
>
> In Matlab, it'd be F = finv(1-0.01,df1,df2), where df1 is the rank of
> the contrast and df2 is the rank of the design matrix. I think
> LibreOffice and Excel have similar functions.
>
> All the best,
>
> Anderson
>
>
>
> Am 05.02.14 09:32, schrieb Stefanie K:
>> Hi all,
>> Can someone recommend a suitable threshold for cluster mass
>> thresholding in randomise (using the -S option with a F statistic,
>> analysing data from FIRST)?
>> Many thanks,
>> Stefanie
>
|