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?ThanksRegards-VMOn 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,AndersonOn 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?ThanksRegards-VMOn 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,AndersonOn 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 asEV 10: patient subgroup 1; controls and patient subgroup 2 =0EV11: patient subgroup 2 =1; controls and patient subgroup 1 =0Then it will make the matrix rank deficient. What is the best way to code them without making the matrix rank deficient?ThanksRegards-VMOn 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: InterceptEV2: Controls/Group1 = -1/+1; Group2=0;EV3: Controls/Group2 = -1/+1; Group1 = 0EV4: Male/Female = 1/-1EV5: RaceA/RaceB = -1/+1; Race C through F = 0;EV6: RaceA/RaceC = -1/+1; All other races 0EV7: RaceA/RaceD = -1/+1; All other races 0EV8: RaceA/RaceE = -1/+1; All other races 0EV9: RaceA/RaceF = -1/+1; All other races 0EV10: Patient Subgroup1/Patient SubGroup 2 = -1/+1; Control =0EV11: EV2*EV5EV12: EV2*EV6EV12: EV2*EV7EV13: EV2*EV8EV14: EV2*EV9EV15: EV3*EV5EV16: EV3*EV6EV17: EV4*EV7EV18: EV3*EV8EV19: EV3*EV9EV20: AgeEV21: educationEV22: EV2*EV23EV23: EV3*EV23EV24: MoCA for Controls; 0 for other groupEV25: MoCA for Grp1; 0 for other groupEV26: MoCA for Grp2; 0 for other groupEV27: disease score1 for Controls; 0 for other groupEV28: disease score1 for Grp1; 0 for other groupEV29: disease score1 for Grp2; 0 for other groupEV30: disease score2 for Controls; 0 for other groupEV31: disease score2 for Grp1; 0 for other groupEV32: disease score2 for Grp2; 0 for other groupEV33: disease score3 for Controls; 0 for other groupEV34: disease score3 for Grp1; 0 for other groupEV35: disease score3 for Grp2; 0 for other groupEV36: ICVOn 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 effectsGrp1>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 effectseducation 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 groupMoCA 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 groupDisease 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 groupDisease 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 groupDisease 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 effectsCreate 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 -logpYes, 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,AndersonThank you for any response. I will greatly appreciate it.Regards and a Happy new year!-- VM