Hi VM,
Thanks for letting me know. The bug can be fixed, although it's probably
better to simply drop the -saveparametric option. It's seldom useful and
will likely be removed in a future version.
The -zstat option was added automatically because the correction involves
both t and F contrasts (more specifically, it involves contrasts with
different ranks).
Hope this helps.
All the best,
Anderson
On 8 January 2017 at 00:43, neuroimage analyst <[log in to unmask]
> wrote:
> Thank you, Anderson.
>
> I ran the follwoing line:
> palm -i bh_volume_con_s.15.mgz -i bh_cbf_con_s.15.mgz -i
> bh_thickness_con_s.15.mgz -s bh.white -d design_matrix_con_vs_s.mat -t
> contrast_matrix_con_vs_s.con -f f_contrast_matrix_con_vs_s.fts -o
> /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.results -n 100 -corrcon
> -approx tail -T -tfce2D -logp -fdr -npc -saveparametric -Tnpc
> -nouncorrected -corrmod
>
> After running for 2 days the script showed an error in saving the p-values:
>
> Computing p-values.
> Saving p-values (uncorrected, and corrected within modality and within
> contrast).
> Saving file: /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.
> results_dpv_ztstat_fwep_m1_c1
> Saving file: /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.
> results_dpv_ztstat_fdrp_m1_c1
> Saving file: /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.
> results_tfce_ztstat_m1_c1
> Saving file: /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.
> results_tfce_ztstat_fwep_m1_c1
> Saving file: /media/sf_M_DRIVE/subjects/palm_stats/Con_vs_s/bh.
> results_tfce_ztstat_fdrp_m1_c1
>
> Undefined function or variable 'P'.
>
> Error in palm_saveall (line 323)
> palm_quicksave(fastfdr(P),1,opts,plm,y,m,c, ...
>
> Error in palm_core (line 2474)
> palm_saveall(plm,opts);
>
> Error in palm (line 81)
> palm_core(varargin{:});
>
>
> It appears the error happened during computing fdr p-values. Would you
> know why would that error occur?
>
>
> Second, as you can see I didnt invoke zstat but the results are
> "*_ztstat*'. Would you know why?
>
> Thanks
>
> Regards
>
> -VM
>
> On Sat, Jan 7, 2017 at 4:04 AM, Anderson M. Winkler <
> [log in to unmask]> wrote:
>
>> Hi VM,
>>
>> So there is some overlap... group1 + group2 = subtype1 + subtype 2. The
>> controls aren't part of this set. To get around the rank deficiency, leave
>> EV10 without any splitting (i.e., coded as 0 for controls and +1/-1 for the
>> two subtypes).
>>
>> All the best,
>>
>> Anderson
>>
>>
>> On 6 January 2017 at 14:42, neuroimage analyst <
>> [log in to unmask]> wrote:
>>
>>> Hi Anderson.
>>>
>>> Thanks.
>>>
>>> There are total of 155 patients which are classified into 60 as group 1
>>> and 95 as group 2. Out of these total 155 patients, 43 are subtype1 and
>>> remaining as subtype2. There isn't a perfect overlap among the patient
>>> subgroups.
>>>
>>> The EV for the design matrix has 170 rows. The way I code is EVx = 1 for
>>> subtype1 ( total 43) 0 for all othrs and 1 for subtype2 ( total 112) 0 for
>>> all others. I was thinking to create just 1 EV with 0 for controls, 1 for
>>> subtype1 and -1 for subtype2. Is this coding not right?
>>>
>>> Thanks
>>> Regards
>>> -VM
>>>
>>>
>>> On Jan 6, 2017 4:01 AM, "Anderson M. Winkler" <[log in to unmask]>
>>> wrote:
>>>
>>>> Hi VM,
>>>>
>>>> Who are the patients? Are these the same as Group1 or Group2, or both?
>>>> How does Patient Subgroups 1 and 2 differ from Groups 1 and 2? There may be
>>>> some perfect overlap between some of these groups/subgroups...
>>>>
>>>> All the best,
>>>>
>>>> Anderson
>>>>
>>>>
>>>> On 6 January 2017 at 05:12, neuroimage analyst <
>>>> [log in to unmask]> wrote:
>>>>
>>>>> Hi Anderson.
>>>>>
>>>>> I forgot to ask this question. If I split EV 10 below into 2 EVs as
>>>>>
>>>>> EV 10: patient subgroup 1; controls and patient subgroup 2 =0
>>>>> EV11: patient subgroup 2 =1; controls and patient subgroup 1 =0
>>>>>
>>>>> Then it will make the matrix rank deficient. What is the best way to
>>>>> code them without making the matrix rank deficient?
>>>>>
>>>>> Thanks
>>>>>
>>>>> Regards
>>>>>
>>>>> -VM
>>>>>
>>>>> On Jan 4, 2017 3:34 AM, "AndersoM. Winkler" <[log in to unmask]>
>>>>> wrote:
>>>>>
>>>>> Hi VM,
>>>>>
>>>>> Please see below:
>>>>>
>>>>> On 30 December 2016 at 23:01, neuroimage analyst <
>>>>> [log in to unmask]> wrote:
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I wish to compare cortical thickness across 3 groups (Control, and 2
>>>>>> patient groups). There are six different type of races that 3 groups belong
>>>>>> to + gender. The 2 patient groups themselves have 2 subdivisions between
>>>>>> them (SubGroup1 and SubGroup2). There are additional covariates of age,
>>>>>> education, MoCA, disease score 1, disease score2, disease score 3, and ICV.
>>>>>>
>>>>>> I have constructed the following design matrix and contrast matrix
>>>>>> and will appreciate if somebody could comment if this is right.
>>>>>>
>>>>>> *Design Matrix:*
>>>>>> EV1: Intercept
>>>>>> EV2: Controls/Group1 = -1/+1; Group2=0;
>>>>>> EV3: Controls/Group2 = -1/+1; Group1 = 0
>>>>>> EV4: Male/Female = 1/-1
>>>>>> EV5: RaceA/RaceB = -1/+1; Race C through F = 0;
>>>>>> EV6: RaceA/RaceC = -1/+1; All other races 0
>>>>>> EV7: RaceA/RaceD = -1/+1; All other races 0
>>>>>> EV8: RaceA/RaceE = -1/+1; All other races 0
>>>>>> EV9: RaceA/RaceF = -1/+1; All other races 0
>>>>>> EV10: Patient Subgroup1/Patient SubGroup 2 = -1/+1; Control =0
>>>>>> EV11: EV2*EV5
>>>>>> EV12: EV2*EV6
>>>>>> EV12: EV2*EV7
>>>>>> EV13: EV2*EV8
>>>>>> EV14: EV2*EV9
>>>>>> EV15: EV3*EV5
>>>>>> EV16: EV3*EV6
>>>>>> EV17: EV4*EV7
>>>>>> EV18: EV3*EV8
>>>>>> EV19: EV3*EV9
>>>>>> EV20: Age
>>>>>> EV21: education
>>>>>> EV22: EV2*EV23
>>>>>> EV23: EV3*EV23
>>>>>> EV24: MoCA for Controls; 0 for other group
>>>>>> EV25: MoCA for Grp1; 0 for other group
>>>>>> EV26: MoCA for Grp2; 0 for other group
>>>>>> EV27: disease score1 for Controls; 0 for other group
>>>>>> EV28: disease score1 for Grp1; 0 for other group
>>>>>> EV29: disease score1 for Grp2; 0 for other group
>>>>>> EV30: disease score2 for Controls; 0 for other group
>>>>>> EV31: disease score2 for Grp1; 0 for other group
>>>>>> EV32: disease score2 for Grp2; 0 for other group
>>>>>> EV33: disease score3 for Controls; 0 for other group
>>>>>> EV34: disease score3 for Grp1; 0 for other group
>>>>>> EV35: disease score3 for Grp2; 0 for other group
>>>>>> EV36: ICV
>>>>>>
>>>>>>
>>>>> On a quick skim this seems right. However, you may want to split the
>>>>> patient subgroups (EV10) into two EVs, one for each patient group.
>>>>>
>>>>> Another thing is that, in seeing the whole thing, it seems more
>>>>> logical and easier to construct the contrasts if the intercept were dropped
>>>>> and the 3 groups (controls, patients 1, patients 2) had each its own EV
>>>>> (coded as 0 and 1). The contrasts for the group comparisons can then be
>>>>> simple comparisons as [1 -1 0 0 ...] and so on.
>>>>>
>>>>>
>>>>>> *Contrast Matrix:*
>>>>>>
>>>>>> a) Test if there are any differences between the groups after
>>>>>> regressing out other effects
>>>>>>
>>>>>> Grp1>Con : 0,2,1,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> Grp1<Con : 0,-2,-1,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> Grp2>Con : 0,1,2,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> Grp2<Con : 0,-1,-2,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> Grp1>Grp2 : ,0,1,-1,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> Grp1<Grp2 : 0,-1,1,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> ,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>>
>>>>>
>>>>> Looks right.
>>>>>
>>>>>
>>>>>>
>>>>>> b) Test if there are any differences due to education between the
>>>>>> groups after regressing out other effects
>>>>>>
>>>>>> education Grp1>Con : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,2,1,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> education Grp1<Con : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,-2,-1,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> education Grp2>Con : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,1,2,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> education Grp2<Con : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,-1,-2,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> education Grp1>Grp2 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,1,-1,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> education Grp1>Grp2 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,-1,1,0,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>>
>>>>>
>>>>> Looks right, but education by group interaction could have been coded
>>>>> in the same manner as MoCA and the disease scores (as below).
>>>>>
>>>>>
>>>>>>
>>>>>>
>>>>>> c) Test if there are is any association between MoCA within group
>>>>>>
>>>>>> MoCA Con >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> MoCA Con<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0,0,0,0,0,0
>>>>>> MoCA Grp1 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0
>>>>>> MoCA Grp1 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0,0,0,0,0
>>>>>> MoCA Grp2 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0
>>>>>> MoCA Grp2 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0,0,0,0
>>>>>>
>>>>>> d) Test if there are is any association between disease score 1
>>>>>> within group
>>>>>>
>>>>>> Disease Score1 Con : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0
>>>>>> Disease Score1 Con<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0,0,0
>>>>>> Disease Score1 Grp1 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0
>>>>>> Disease Score1 Grp1<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0,0
>>>>>> Disease Score1 Grp2 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0,0
>>>>>> Disease Score1 Grp2<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0,0
>>>>>>
>>>>>> e) Test if there are is any association between disease score2
>>>>>> within group
>>>>>> Disease Score2 Con >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0,0
>>>>>> Disease Score2 Con<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0,0
>>>>>> Disease Score2 Grp1 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0,0
>>>>>> Disease Score2 Grp1<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0,0
>>>>>> Disease Score2 Grp2 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0,0
>>>>>> Disease Score2 Grp2<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0,0
>>>>>>
>>>>>> e) Test if there are is any association between disease score2
>>>>>> within group
>>>>>> Disease Score3 Con >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0,0
>>>>>> Disease Score3 Con<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0,0
>>>>>> Disease Score3 Grp1 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0,0
>>>>>> Disease Score3 Grp1<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0,0
>>>>>> Disease Score3 Grp2 >0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,2,0
>>>>>> Disease Score3 Grp2<0 : 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>>>>>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,-2,0
>>>>>>
>>>>>> f) How to test if there are any differences due to race between the
>>>>>> groups after regressing out other effects
>>>>>>
>>>>>
>>>>> Create one t-test for each of the EVs that test the race pairs (so,
>>>>> one contrast for each of EV5 through EV9), then an F-test that encompasses
>>>>> all these. This will test for any effect on any direction.
>>>>>
>>>>>
>>>>>>
>>>>>>
>>>>>> *Questions:*
>>>>>>
>>>>>> 1) Does the design matrix and contrast matrix makes sense? There are
>>>>>> in total 16 controls, 40 grp1, and 140 grp2 subjects?
>>>>>>
>>>>>
>>>>> Yes, although, as indicated above, it could be made simpler (or less
>>>>> error prone) if in this case the intercept is dropped and the 3 groups
>>>>> coded separately.
>>>>>
>>>>>
>>>>>>
>>>>>> 2) This is my call to PALM. Is the call correct?
>>>>>>
>>>>>> palm -i bh.thickness.mgz -i bh.area.mgz -i bh.volume.mgz -i
>>>>>> bh.cbf.mgz -d design.mat -t design.con -o bh.results -n 1000 -corrcon
>>>>>> -corrmod -approx tail -nouncorrected -s bh.white -T -tfce2D -logp
>>>>>>
>>>>>
>>>>> Yes, looks right.
>>>>>
>>>>>
>>>>>>
>>>>>> 3) How to test if there are any differences due to race between the
>>>>>> groups after regressing out other effects? -- Contrast matrix? Can I use an
>>>>>> f test here? If yes, how should I pass the argument into PALM call above?
>>>>>>
>>>>>
>>>>> Yes, as indicated above. To pass the F-test use the option -f, just as
>>>>> you would in randomise.
>>>>>
>>>>>
>>>>>>
>>>>>> 4) With so many contrasts and using -corcon option, I am going to
>>>>>> lose a lot of power? Can the contrast matrix be simplified to test what I
>>>>>> want to test?
>>>>>>
>>>>>
>>>>> Yes, you can remove beforehand hypotheses or tests that are not
>>>>> relevant or meaningful, although in this case it seems all of them are
>>>>> plausible and presumably could be investigated.
>>>>>
>>>>>
>>>>>>
>>>>>> 5) How do I get FDR corrected p-value for the call to PALM above?
>>>>>>
>>>>>
>>>>> Add the option -fdr.
>>>>>
>>>>>
>>>>>>
>>>>>> 6) Since the intercept is present in the model, my understanding is I
>>>>>> don't have to demean the regressors in the design matrix, Is that right?
>>>>>>
>>>>>
>>>>> Yes. And if you drop the intercept as indicated above, replacing it
>>>>> (and the two other group EVs) for a new set of 3 EVs, the intercept will
>>>>> still be present, albeit split across the 3 regressors, such that demeaning
>>>>> is still not necessary.
>>>>>
>>>>> Hope this helps.
>>>>>
>>>>> All the best,
>>>>>
>>>>> Anderson
>>>>>
>>>>>
>>>>>
>>>>>>
>>>>>> Thank you for any response. I will greatly appreciate it.
>>>>>>
>>>>>> Regards and a Happy new year!
>>>>>>
>>>>>> -- VM
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>
>>
>
