Hi Anderson,
many thanks for answering so quickly. The patch did it, erverything runs
fine.
I've one more question regarding your final remark:
As far as I understood the variance groups option tells palm that
variances in these groups may differ and, therefore, the G statistic has
to be used instead of F. If I treat then my whole sample as one
exchangeability block, I expected that palm permutes the cases within this
one large block and calculates G for each permutation, building the
permutation distribution in this way. This is in my eyes a reasonable
procedure. Do I really need three exchangeability blocks when I assume
different variances? With other words: Are data not exchangeable if they
come from populations with different variances?
All the best,
Bertram
On Mon, 30 Jan 2017 11:37:10 +0100, Anderson M. Winkler
<[log in to unmask]> wrote:
> Hi Bertram,
>
> Thanks for letting me know and sorry for the bug. To fix, in palm_core.m,
> lines 2946-2948, where it reads:
>
> *bsum = bsum/plm.rC{m}(c)/(plm.rC{m}(c)+2);*
> *df2 = 1/3./bsum;*
> *G = G./(1 + 2*(plm.rC{m}(c)-1).*bsum);*
>
>
> please replace for:
>
> *bsum = bsum/plm.rC0{m}(c)/(plm.rC0{m}(c)+2);*
> *df2 = 1/3./bsum;*
> *G = G./(1 + 2*(plm.rC0{m}(c)-1).*bsum);*
>
>
> Also, lines 2973-2975, where it reads:
>
> *bsum = bsum/plm.rC{m}(c)/(plm.rC{m}(c)+2);*
> *df2 = 1/3./bsum;*
> *G = G./(1 + 2*(plm.rC{m}(c)-1).*bsum);*
>
>
> please replace for:
>
> *bsum = bsum/plm.rC0{m}(c)/(plm.rC0{m}(c)+2);*
> *df2 = 1/3./bsum;*
> *G = G./(1 + 2*(plm.rC0{m}(c)-1).*bsum);*
>
>
> That is, the instances of "plm.rC" should be "plm.rC0". This has just
> been
> fixed in version alpha103. You can instead simply download the newest
> version from the webpage:
> https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/PALM/UserGuide
>
> Also, unrelated to the error, if the subjects belong to different
> variance
> groups, then you also need exchangeability blocks, but with this design,
> using EBs leave nothing to be permuted. Then you need sign-flippings
> (option -ise).
>
> So, you can either: (1) Drop the option "-vg", such that all subjects
> will
> be treated as with same variance (the usual assumption); or (2) add the
> option -ise, to do sign flippings instead of permutations (this assumes
> symmetric distribution around zero).
>
> All the best,
>
> Anderson
>
>
> On 30 January 2017 at 08:36, Bertram Walter <
> [log in to unmask]> wrote:
>
>> Dear PALM experts,
>>
>> I used palm for an ANOVA comparing three groups. It was running fine
>> with
>> the unpermuted data, but with the first permutation I got a 'division by
>> zero' error. Where is my mistake?
>>
>> I have attached the input csv/vest files and the screen output.
>>
>> All the best,
>>
>> Bertram
>>
>> --
>> Dr. Bertram Walter
>> Bender Institute of Neuroimaging
>> University of Giessen
>> Otto-Behaghel-Str. 10H
>> 35394 Giessen
>> Germany
>> Phone +49 (641) 99-26307
>> or +49 (641) 99-26331 (Secretary)
>> Fax +49 (641) 99-26309
>> www.bion.de
--
Dr. Bertram Walter
Bender Institute of Neuroimaging
University of Giessen
Otto-Behaghel-Str. 10H
35394 Giessen
Germany
Phone +49 (641) 99-26307
or +49 (641) 99-26331 (Secretary)
Fax +49 (641) 99-26309
www.bion.de
|