Hi Matthieu,

Do you mean which range is significant? You'd take the image with the p-values, threshold it at the significance level you want (e.g., fwep at 0.05, or at 0.95 if 1-p, or at 1.3 if log(p)), then use it to mask the corresponding statistical map (tstat, vstat, zstat).

In any case, the significant is always in the positive side, i.e, positive z, or t, or v. The negative contrasts flip a negative effect, so that it becomes positive.

All the best,

Anderson


On 14 October 2015 at 08:53, Matthieu Vanhoutte <[log in to unmask]> wrote:
Hi Anderson,

I will convert into z-score. But e.g. for a contrast of [-1], how could I determine in z-stat which range of values (positive or negative) I should use ?

Best regards,

Matthieu


2015-10-14 8:48 GMT+02:00 Anderson M. Winkler <[log in to unmask]>:
Hi Matthieu,

It's difficult to tell as it depends on the degrees of freedom. However, you can use the option "-zstat" so that the output statistics will be converted to z. Then things should be in a more familiar range.

All the best,

Anderson


On 13 October 2015 at 10:47, Matthieu Vanhoutte <[log in to unmask]> wrote:
Dear Anderson,

Thank you !

If i used unthresholded t-statistic (or v-statistics) maps, what range have I to consider ? E.g., in one t-statistic map, T varies from -10 to 10 so have I to consider negative, positive or both values according to my contrast ?

Best regards,

Matthieu


2015-10-11 10:17 GMT+02:00 Anderson M. Winkler <[log in to unmask]>:
Hi Matthieu,

The -ise is needed still.

All the best,

Anderson


On 10 October 2015 at 21:09, Matthieu Vanhoutte <[log in to unmask]> wrote:

Hi Anderson,

So I will have to use the -eb and -vg options and not the -ise any more ?
Isn't -ise option needed in m'y case ?

Best regards,

Matthieu

Le 10 oct. 2015 12:26, "Anderson M. Winkler" <[log in to unmask]> a écrit :
Hi Matthieu,

Hmm, I think may have meant to make a change but didn't. Well, it won't matter in this case as the -ise is transparent to the block definitions with within-block shuffling, so try repeating but using the options "-eb VarianceGroup.csv -vg auto", that is, it will use the VG file as EB, and from these, define automatically the actual VG. This is the same as "-eb VarianceGroup.csv -vg VarianceGroup.csv".

All the best,

Anderson


On 10 October 2015 at 10:02, Matthieu Vanhoutte <[log in to unmask]> wrote:

Dear Anderson,

Many thanks for all these detailed explanations !

However I used the -vg option un my command line with variance group differences between subjects in the .csv file, but still got t-stat and not v-stat outputed of PALM...

How could it be possible ?

Best regards,

Matthieu

Le 10 oct. 2015 10:42, "Anderson M. Winkler" <[log in to unmask]> a écrit :
There are two things going on:

1) The two contrasts can go in the same .csv file, i.e., one contrast per row. This single file is supplied with -t. Or, as you did, the -d can be supplied twice, although for this case doing this seems a bit cumbersome (multiple designs can be considered, but usually when these are different). It's fine though.

2) In the t-stat (produced when -vg is omitted, or if all subjects are in the same VG), the results aren't necessarily valid, as the assumptions of the test are violated twice: first the interpretation of "there is a difference in means" only stands if the variances are the same, and second, the unrestricted permutation test requires that the variances are the same so as to preserve exchangeability. The v-stat (produced when -vg is supplied) is valid.

I'll reply the last one next.

 

I have tried to launch PALM with one sample t-test on the 4D COPEs image (merges of all subjects' COPE), with the following command line :

palm -i /SUBJECTDIR/4D_COPE.nii
-d /SUBJECTDIR/design.csv
-t /SUBJECTDIR/ContrastPos.csv
-t /SUBJECTDIR/ContrastNeg.csv
-vg /SUBJECTDIR/VarianceGroup.csv
-ise
-save1-p
-o /SUBJECTDIR/Prefix


with design.csv a unique column vector of 1 (number of subjects = 7), ContrastPos.csv = 1 and ContrastNeg.csv = -1 (because I wanna test two t-contrast)  and

VarianceGroupe.csv =[
2
3
2
4
4
3
3
];

and got this error message :

Error using palm_takeargs (line 1217)
More t-contrast files (2) than valid design files (1) were supplied.

Error in palm_core (line 32)
[opts,plm] = palm_takeargs(varargin{:});

Error in palm (line 80)
palm_core(varargin{:});

Error in PALM_LEMP (line 3)


So in order to test sequentially contrast 1 then -1, I tried (is it not possible otherway ?) :

palm -i /SUBJECTDIR/4D_COPE.nii
-d /SUBJECTDIR/design.csv
-t /SUBJECTDIR/ContrastPos.csv
-d /SUBJECTDIR/design.csv
-t /SUBJECTDIR/ContrastNeg.csv
-vg /SUBJECTDIR/VarianceGroup.csv
-ise
-save1-p
-o /SUBJECTDIR/Prefix


and got no significant results on corrected (fwep) t-stat and not vstat (as you mentioned above with the Aspin–Welch's), is it normal ?

Is my design, contrasts and command line correct to test positive or negative longitudinal effect on my 7-subjects group ?