Hi - yes that looks correct.
Steve.
---------------------------------------------------------------------------
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
---------------------------------------------------------------------------
Stop the cultural destruction of Tibet
> Dear Mark,
> I want to know whether it is right if I want to get the beta value in randomise.
> Here is my design matrix:
> Group EV1 EV2 EV3 EV4
> input1 1 score_demean1 ... ... ...
> input2 1 score_demean2 ... ... ...
> input3 1 score_demean3 ... ... ...
> ... ... ... ... ... ... ...
> input48 1 score_demean48 ... ... ...
> input49 1 score_demean49 ... ... ...
> input50 1 score_demean50 ... ... ...
> Here is my contrast:
> EV1 EV2 EV3 EV4
> C1 age 1 0 0 0
> C2 edu 0 1 0 0
> C3 sex 0 0 1 0
> C4 reactiontime 0 0 0 1
> Here is my script:
> randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d design.mat -t design.con -T -n 5000
> --glm_output=GLM
> Did I run randomise properly if I want to get the beta value from the --glm_output? Then I can see GLM file for next analysis.
> Thanks in advance.
> All the best.
> Rujing Zha
> 发送时间:2013-11-27 10:49
> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
> 抄送:
>
> Dear professor Mark,
> I see. Thanks in advance.
> All the best.
> 2013-11-27
> Rujing Zha
> 发送时间:2013-11-27 07:57
> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
> 抄送:
>
> Hi,
>
> The -D option will effectively turn EV1 into a zero EV, which is then a problem. So it is better to avoid the -D or avoid having EV1.
>
> All the best,
> Mark
>
>
>
>> Dear professor Mark,
>> Thanks for your precious prompt.
>> I can understand your reply about -D option and group mean(EV1). I want to understand in advance. What will happen, if I put -D option and group mean contrast(EV1) simultaneously? As I know, -D option will demean the data. I guess that beta value of EV1 will be zero if I put them simultaneously. Am I right?
>> It will not change beta value and p value of other EVs if it is true. Am I right?
>> All the best.
>>
>> 2013-11-26
>> Rujing Zha
>> 发送时间:2013-11-26 16:03
>> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>> 抄送:
>>
>> Hello,
>>
>>> Thanks you in advance.I also have 3 problems.
>>> 1,Doesnot -t design.con option represent contrasts that you just told me? Did you mean I should design design.con,which should be revise in "Contrasts & F-tests" of Glm? If I get your prompt, here is my next thoughts.
>> Yes, that is right.
>>
>>> I need 5 EVs as I have four regressors of interests(age,edu,sex,reaction times).Then set the contrast:
>>> title
>>> EV1 EV2 EV3 EV4 EV5
>>> C1 group mean 1 0 0 0 0
>>> C2 age 0 1 0 0 0
>>> C3 edu 0 0 1 0 0
>>> C4 sex 0 0 0 1 0
>>> C5 reactiontime 0 0 0 0 1
>> That looks fine ...
>>> As I have used -D option, so C1 isnot needed,right?
>> … but not if you use -D. You should drop the -D option, or remove EV1 and the column in the contrasts associated with it. Either choice is valid.
>>
>>> Is that right? I want to know what option I should add to make up this pitfall, if it is wrong.
>>
>> I would just drop the -D option if you already have your design matrix and contrasts set up with EV1 being a column of ones.
>>
>>> 2,As I have used -D option in FA,so it is not necessory to add a contrast for group mean contrast. Can -D option be needed in gray matter regression analysis?
>> As I said above, you have two options with the analyses: (1) use -D and do not include EV1 or the associated column in the contrasts; (2) do not use -D but keep EV1 (a column of ones) and the associated column in the contrasts. Both are valid. If you choose option 2, then you can either have a contrast for the group mean (your first contrast - i.e. the first row) or not have this, either is fine.
>>
>>> 3,I cannot get correlation values in randomise. Can I get the beta value in randomise?
>> Yes, use the --glm_output option.
>>
>> All the best,
>> Mark
>>
>>
>>
>>> All the best.
>>>
>>> 2013-11-21
>>> Rujing Zha
>>> 发件人:Mark Jenkinson <
[log in to unmask]>发送时间:2013-11-21 01:09主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm收件人:"FSL"<
[log in to unmask]>
>>> 抄送:
>>>
>>> Hi,
>>>
>>> Your design matrices look fine, but you also need contrasts. If you are not sure about these then you should go through the FSL course material (especially the practicals) as well as look at material in text books related to GLM analysis in neuroimaging.
>>>
>>> Note that you can use different calls to randomise to do the same, or similar, analyses. The choice of multiple-comparison correction is not fixed, but is up to the user (although some choices, such as T2, are tuned specifically for some data, such as skeletonised TBSS data for the T2 case). The number of permutations is also a user choice. If you are unsure about this then just use the larger value (5000) as it is more conservative/accurate.
>>>
>>> Regarding correlation versus regression analysis, I think you are confused about what is necessary for statistical testing. We implement everything via regression analysis, which gives exactly the same statistical test results as a correlation analysis, but does not calculate correlation values. You can calculate your own correlation values if you really want them, but they are not necessary for our GLM analyses and any statistical test on them would give exactly the same statistical results.
>>>
>>> All the best,
>>> Mark
>>>
>>>
>>>
>>>> Dear Joe,
>>>> Thank you for your suggestion for a long time.
>>>> Please forgive me that I cannot find apparent description about correlation and multiple regression analysis in your link. Maybe it does, I cannot find.
>>>> I have thought about my analysis, and there are some ideas to deal with the correlation and regression analysis.But I have never done that by FSL,so I donot know wether is is right.
>>>> Dear Joe,can you help me look over my idea? Can you help me correct if the idea has some pitfalls?I just want to correlate FA(whiter matter) and score(obtain from questionare),and regress FA and subjects' demographic data(such as age,education,IQ et.al.).
>>>> Here are my idea as below:(I have 30 controls and 20 patients)
>>>> +++++++++++++++++++++++++++
>>>> First:correlation between FA and score:
>>>> 1,demean original score
>>>> score1 >>> score_demean1
>>>> score2 >>> score_demean2
>>>> score3 >>> score_demean3
>>>> ...
>>>> score48 >>> score_demean48
>>>> score49 >>> score_demean49
>>>> score50 >>> score_demean50
>>>> 2,create design.mat and design.con by Glm
>>>> Group EV1 score
>>>> input1 1 1 score_demean1
>>>> input2 1 1 score_demean2
>>>> input3 1 1 score_demean3
>>>> ... ... ... ...
>>>> input48 1 1 score_demean48
>>>> input49 1 1 score_demean49
>>>> input50 1 1 score_demean50
>>>> 3,randomise
>>>> FA: randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -n 500 -D --T2
>>>> or
>>>> gray matter: randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d design.mat -t design.con -T -n 5000
>>>> Second:multiple regression between FA and behavior measurements(age,reaction time,sex,education, other interested behavior measurements ),FA are considered as dependent variable.
>>>> 1,demean all behavior measurements
>>>> 2,create design.mat and design.con by Glm
>>>> Group EV1 age reaction time sex education ...
>>>> input1 1 1 age_demean1 RT_demean1 sex_demean1 edu_demean1 ...
>>>> input2 1 1 age_demean2 RT_demean2 sex_demean2 edu_demean2 ...
>>>> input3 1 1 age_demean3 RT_demean3 sex_demean3 edu_demean3 ...
>>>> ... ... ... ... ... ... ... ...
>>>> 3,randomise
>>>> FA: randomise -i all_FA_skeletonised -o tbss -m mean_FA_skeleton_mask -d design.mat -t design.con -n 500 -D --T2
>>>> or
>>>> gray matter: randomise -i GM_mod_merg_s3 -m GM_mask -o fslvbm -d design.mat -t design.con -D -T -n 5000
>>>> +++++++++++++++++++++++++++++
>>>> Dear joe,is it appropiate? How can I get the beta and p value of each behavior measurement in regression and correlation coefficient in correlation analysis if what I wrote was right?If it is wrong,can you help me design a correct one?
>>>> Any reply will be highly appreciated.
>>>> All the best.
>>>>
>>>> 2013-11-20
>>>> Rujing Zha
>>>> 发送时间:2013-11-20 10:31
>>>> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>> 抄送:
>>>>
>>>> Dearly Joe,
>>>> Thank you in advance.I will try.
>>>> All the best.
>>>>
>>>> 2013-11-20
>>>> Rujing Zha
>>>> 发送时间:2013-11-20 01:07
>>>> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>> 抄送:
>>>>
>>>> What makes that example test for controls and patients are the design.mat and design.con files. You should have these files already if you ran a GLM previous, and randomise will run whatever contrasts you have set. If you need help with setting up the design matrix, the Feat help page should have almost every comparison you might want to make, and has advice for applying these examples to randomise:http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FEAT/UserGuide#Group_Statistics
>>>>
>>>> This is also a helpful page for setting up design matrices: http://mumford.fmripower.org/mean_centering/
>>>>
>>>>
>>>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Rujing Zha
>>>> Sent: Monday, November 18, 2013 10:56 PM
>>>> Subject: Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>>
>>>> Dearly Joe,
>>>> Thank you in advance.I have read the link before e-mail to FSL list,but it is testing the difference between controls and patients. I alse have read the randomise --help, but I could not script regression analysis code, so am I.
>>>> Could you please give me a detail information?
>>>> Thanks very much.
>>>> All the best.
>>>> 2013-11-19
>>>> Rujing Zha
>>>> 发送时间:2013-11-19 13:47
>>>> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>> 抄送:
>>>> Hi, there is an example on the page I linked to. Look for the code in the gray box that refers to randomise. You can probably copy and paste that exact code if you are in the stats directory of where you ran tbss. There is also a link on that page to the randomise wiki page that further explains its usage. You could also type 'randomise --help' on the command line.
>>>>
>>>>
>>>> Dearly Joe,
>>>> Thanks for your precious prompt. However I can not script this code.
>>>> Can you give me a example or a help archive? Then I can study it.
>>>> Any reply will be highly repreciated.
>>>> All the best.
>>>> Rujing Zha
>>>> 2013-11-19
>>>> Rujing Zha
>>>> 发送时间:2013-11-18 23:43
>>>> 主题:Re: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>> 抄送:
>>>> If you want to run statistics on skeletonized FA data, you should use randomise, as described on the page for Tract-Based Spatial Statistics:http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/TBSS/UserGuide#voxelwise_statistics_on_the_skeletonised_FA_data
>>>>
>>>> Randomise will apply the appropriate corrections.
>>>>
>>>> Joe
>>>>
>>>> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Rujing Zha
>>>> Sent: Monday, November 18, 2013 7:03 AM
>>>> Subject: [FSL] correction about regression result of whiter matter and behavior data by fsl_glm
>>>> Dear fsl users,
>>>> I have done the multiple regression analysis by fsl_glm.
>>>> fsl_glm -i all_FA_skeletonised.nii -d design -o my_result --demean --out_p=pvalue
>>>> Do I need to correct the pvalue?
>>>> Thanks!
>>>> All the best.
>>>> Rujing Zha
>