 |
 |
Hi Nancy Li, In the manual, the upper part of the examples has the "design". The lower part (where C1, C2, etc, are indicated) are the contrasts. You must not mix these. I think I've just answered these other questions in the other thread. All the best, Anderson On 25 January 2017 at 17:20, Nancy Li <[log in to unmask]> wrote: > Hi Anderson, > Good morning. Thank you so much! It seemed more close to my puzzle. So in > my understanding, the design look like1 -1 > > > -1 1 > > > 1 0 > > > -1 0 > > > 0 1 > > > 0 > -1 > > Is it right? Then bring the design.con and design.mat to the randomise > command of two unpaired group. Through this way you could get the > comparision results of two groups and mean activation of seperate > group. Looking forward to your more instructions.Thanks a lot! But in the > high level ananlysiis of Feat, the design only 1 -1 > > > > > -1 1 > > > > > 1 0 > > > > > 0 1 > > > > > > It only induced mean activation Z value>0 ? > > > > > Best > > Nancy > > 在 2017-01-25 16:46:58，"Anderson M. Winkler" <[log in to unmask]> 写道： > > Hi Nancy, > > As shown in the manual, the contrasts for group comparisons are [1 -1] and > [-1 1]. For single group means, you can still do in the way as I commented: > you can test these contrasts but you have to add to randomise the option > "-1", such that it does sign-flippings. > > All the best, > > Anderson > > > On 23 January 2017 at 15:59, Nancy Li <[log in to unmask]> wrote: > >> Hi Anderson, >> >> Good morning. Thank you very much first. I am a little puzzled. If I want >> to use the randomise command of two group to gain the mean activation of >> seperate group by adding (1 0 0 1) to design.cn. It is also reasonable, >> is it? Thank you for confirmation. I appreciate your help greatly. >> >> Best >> >> Nancy >> >> >> >> >> >> At 2017-01-21 21:54:18, "Anderson M. Winkler" <[log in to unmask]> >> wrote: >> >> Hi Nancy, >> >> For the simple group mean the contrast [1 0] is fine, but you'll need to >> use sign-flippings then (add the option -1 to randomise). >> >> All the best, >> >> Anderson >> >> >> On 20 January 2017 at 13:43, Nancy Li <[log in to unmask]> wrote: >> >>> Hi Anderson, >>> >>> Good morning. What I want to do is to compare two group to find is there >>> any difference between two groups, at same time I want to know the mean >>> activtion of seperate group(eg: the mean activation of patient group and >>> control group ), not the whole mean activation. At this >>> condition, design.con: 1 0 >>> >>> >>> >>> 0 1 >>> >>> >>> >>> 1 -1 >>> >>> >>> >>> -1 1 >>> Is it right? just same to the model in the highlevel analysis of Feat. >>> Looking forward to your help. Thank you so much! >>> >>> Have a great weekend >>> >>> Nancy >>> >>> >>> >>> >>> >>> At 2017-01-20 16:52:51, "Anderson M. Winkler" <[log in to unmask]> >>> wrote: >>> >>> Hi Nancy Li, >>> >>> This isn't correct I'm afraid, but I don't understand what you have >>> there or what you are trying to do. Try to follow the manual. >>> >>> All the best, >>> >>> Anderson >>> >>> >>> On 19 January 2017 at 14:11, Nancy Li <[log in to unmask]> wrote: >>> >>>> Hi Anderson, >>>> >>>> Good morning. I appreciate you greatly for sincere and patient >>>> response. In most cases, one subject could not only have one volume. So in >>>> my opinion, if we want to get the mean activation, we usually do the two >>>> unpaired group randomise, just add (1 0) and (0 1) to the design.con. Am I >>>> right? If one subject only have one volume, if we want to get the mean >>>> activation, the command may be this way:randomise -i OneSamp4D -o >>>> OneSampT -1 1 1 -1 -1 -v 5 -T , is it? It is so helpful to talk to >>>> you. Thank you so much! >>>> >>>> Have a great day >>>> >>>> Nancy >>>> >>>> >>>> >>>> >>>> 在 2017-01-19 18:15:11，"Anderson M. Winkler" <[log in to unmask]> >>>> 写道： >>>> >>>> Hi Nancy, >>>> >>>> Please see below: >>>> >>>> >>>> On 18 January 2017 at 14:06, Nancy Li <[log in to unmask]> wrote: >>>> >>>>> Hi Anderson, >>>>> >>>>> Great, It is what I want to know. So in the randomise command It >>>>> seemed like this way:" >>>>> >>>>> randomise -i 4D image -o rando -d design.mat -t design.con -n >>>>> 5000 -T , design.con may be: 1 0 >>>>> >>>>> >>>>> >>>>> 0 1 >>>>> >>>>> >>>>> >>>>> 1 -1 >>>>> >>>>> >>>>> -1 1 >>>>> >>>>> >>>>> Is it right? >>>>> >>>> >>>> >>>> Yes, contrasts are fine. For the overall mean activation, however, you >>>> need to add two contrasts more: [1 1] and [-1 -1]. In randomise, you'd test >>>> these latter two by adding the option -1, that does sign-flippings. >>>> >>>> >>>> >>>>> And How about one sample ttest, it also the way to get the mean >>>>> activation, but how to understand "There should be no repeated measures, >>>>> i.e., there should only be one image per subjec ", It equals one subject >>>>> only have one volume? Need your help to confirmation. Thank you so so much! >>>>> >>>> >>>> Yes, one volume per subject. >>>> >>>> All the best, >>>> >>>> Anderson >>>> >>>> >>>> >>>>> >>>>> Best >>>>> >>>>> Nancy >>>>> >>>>> >>>>> At 2017-01-18 17:26:21, "Anderson M. Winkler" <[log in to unmask]> >>>>> wrote: >>>>> >>>>> Hi Nancy, >>>>> >>>>> For the mean activation, using the same design shown in the Wiki, >>>>> create a contrast that is [1 1]. >>>>> >>>>> All the best, >>>>> >>>>> Anderson >>>>> >>>>> >>>>> On 17 January 2017 at 14:45, Nancy Li <[log in to unmask]> wrote: >>>>> >>>>>> Hi Anderson, >>>>>> >>>>>> Thank you so much for sincere reply. Could you help me make sure if I >>>>>> want to get the mean activation of two different group, It could be done >>>>>> through one sample ttest(seperate group) or unpaired two group ttest, the >>>>>> results are same, is it right? Looking forward to more instruction. Thanks >>>>>> a lot! >>>>>> >>>>>> Best >>>>>> >>>>>> Nancy >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> At 2017-01-16 18:26:55, "Anderson M. Winkler" <[log in to unmask]> >>>>>> wrote: >>>>>> >>>>>> Hi Nancy Li, >>>>>> >>>>>> An example of a two-sample t-test is in the manual, please see here: >>>>>> https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_D >>>>>> ifference_.28Two-Sample_Unpaired_T-Test.29 >>>>>> >>>>>> All the best, >>>>>> >>>>>> Anderson >>>>>> >>>>>> >>>>>> On 13 January 2017 at 18:29, Nancy Li <[log in to unmask]> wrote: >>>>>> >>>>>>> Hi Colin Hawco, >>>>>>> Great thanks. It is more clear. We could also get the result of mean >>>>>>> activation in two group contrast by design matris(1 0; 0 1; 1 -1; -1 1). >>>>>>> In my opinion it is same to the result of seperate one sample test, is it >>>>>>> right? Looking forward to your confirmation. Thanks a lot! >>>>>>> Best >>>>>>> Nancy >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> At 2017-01-14 01:43:04, "Colin Hawco" <[log in to unmask]> >>>>>>> wrote: >>>>>>> >>>>>>> Ahh I see. The means of permuting is different. Instead of >>>>>>> randomizing group assignments, half the data is sign flipped (i.e. >>>>>>> multiplied by -1). In this case, the data should follow a distribution >>>>>>> centered around zero. >>>>>>> >>>>>>> >>>>>>> >>>>>>> Everything in that wiki vis a vis the estimatablility of the true >>>>>>> p-value remains true. >>>>>>> >>>>>>> >>>>>>> >>>>>>> If you can possibly run 5000 permutations you should do so. if for >>>>>>> no other reason than to please reviewer. I would recommend a minimum of >>>>>>> 1000 or 2000 permutations. >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> Colin Hawco, PhD >>>>>>> >>>>>>> Neuranalysis Consulting >>>>>>> >>>>>>> Neuroimaging analysis and consultation >>>>>>> >>>>>>> www.neuranalysis.com >>>>>>> >>>>>>> [log in to unmask] >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> *From:* FSL - FMRIB's Software Library [mailto:[log in to unmask]] *On >>>>>>> Behalf Of *Nancy Li >>>>>>> *Sent:* January-13-17 11:31 AM >>>>>>> *To:* [log in to unmask] >>>>>>> *Subject:* Re: [FSL] Question about randomise of one sample ttest >>>>>>> >>>>>>> >>>>>>> >>>>>>> Hi Colin Hawco, >>>>>>> >>>>>>> >>>>>>> >>>>>>> Thank you so much. It may be understood easily if there is two group >>>>>>> test. But in one sample test we want to get the mean activation, there is >>>>>>> no permutation. How to undrestand the impact of different number of >>>>>>> permutation. looking forward to your further help. >>>>>>> >>>>>>> Best >>>>>>> >>>>>>> >>>>>>> >>>>>>> Nancy >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> At 2017-01-14 00:19:51, "Colin Hawco" <[log in to unmask]> >>>>>>> wrote: >>>>>>> >>>>>>> more permutations give you a more accurate estimation of true >>>>>>> significance. It also defines the possible increments and confidence limits >>>>>>> of the calculated p-value, with more permutations providing greater >>>>>>> confidence. >>>>>>> >>>>>>> >>>>>>> >>>>>>> see: >>>>>>> >>>>>>> https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/Theory >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> Good luck, >>>>>>> >>>>>>> >>>>>>> >>>>>>> Colin Hawco, PhD >>>>>>> >>>>>>> Neuranalysis Consulting >>>>>>> >>>>>>> Neuroimaging analysis and consultation >>>>>>> >>>>>>> www.neuranalysis.com >>>>>>> >>>>>>> [log in to unmask] >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> *From:* FSL - FMRIB's Software Library [mailto:[log in to unmask]] *On >>>>>>> Behalf Of *Nancy Li >>>>>>> *Sent:* January-13-17 11:03 AM >>>>>>> *To:* [log in to unmask] >>>>>>> *Subject:* [FSL] Question about randomise of one sample ttest >>>>>>> >>>>>>> >>>>>>> >>>>>>> Helo experts and folks, >>>>>>> >>>>>>> >>>>>>> >>>>>>> I want to get results of mean group activation using randomise of >>>>>>> one sample. The default permutation is 5000 in the command. My question is >>>>>>> "is there difference between 500 and 5000 permuttaion of one sample test >>>>>>> and why?" Looking forward to your help or may suggest some paper about this >>>>>>> question. Thanks a lot in advance. >>>>>>> >>>>>>> >>>>>>> >>>>>>> Best >>>>>>> >>>>>>> >>>>>>> >>>>>>> Nancy >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> >>>>> >>>>> >>>> >>>> >>>> >>>> >>>> >>> >>> >>> >>> >>> >> >> >> >> >> > > > > > > > >