headers
Sarah Wright | 4 Aug 03:55
Picon

Two sample t test

Dear all

This is a question regarding a two-sample t test.  What is the best way to compare the two groups at the second level?  With the third column of mu, one is unable to use a simple F test between the groups.  What do others do?

Many thanks

Sarah

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 0 comments |
headers
Rick Helmich | 7 Aug 09:51
Picon

F-contrasts in Marsbar

Dear Matthew and list,

 

I have a question about the validity and interpretability of F-statistics in MarsBar. I have a 3 x 2 factorial fMRI design with factors GROUP (3 levels) and CONDITION (called “BMC”; 2 levels).

 

When performing an F-test in an anatomical ROI (left Pallidum) in MarsBar, I got a very significant result (subgroup x BMC; F=5.79, 0.003563). However, when I extracted the averaged betas in order to plot them, and performed the same F-contrast in SPSS, I found that the same interaction effect was far from significant (subgroup x BMC, F=1,29; p=0.28). The same difference did not apply to a simple t-test in the same design: (MarsBar: t=1.56; p=0.060622; SPSS: t=1.649; p=0.108). I understand that MarsBar and SPM use one-sided t-tests, explaining the different p-values for the t-test, but how does this influence F-statistics? And how should I interpret a significant F-statistic in MarsBar?

 

Below I have pasted the results from MarsBar.

 

I am curious to hear your opinion, thanks a lot!

 

Yours, Rick

 

=========================================

Rick C.G. Helmich, MD (Ph.D student)

 

Donders Institute for Brain, Cognition and Behaviour, Centre for Cognitive Neuroimaging

RadboudUniversityNijmegen

 

P.O. Box 9101

6500 HB Nijmegen

The Netherlands

 

T  00 31 (0)24 36 65077

F  00 31 (0)24 36 10989

E  rick.helmich <at> fcdonders.ru.nl

web: www.fcdonders.ru.nl

=========================================

 

 

 

F-CONTRAST

 

Contrast name                   ROI name:       Extra SS:    F statistic:  Uncorrected P:    Corrected P

--------------------------------------------------------------------------------------------------------

 

Interaction: subgroup x BMC

------------------------------------------

                                                                                                                                           

                              Pallidum_L:          27.56:           5.79:       0.003563:       0.003563

 

 

 

T-CONTRAST

 

Contrast name                   ROI name: Contrast value:    t statistic:  Uncorrected P:    Corrected P

--------------------------------------------------------------------------------------------------------

 

CON>TDT_BMC

------------------------------------------

                                                                                                                                                           

                              Pallidum_L:           0.52:           1.56:       0.060622:       0.060622

 

 

 

 

 

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 3 comments |
headers
Rick Helmich | 7 Aug 11:11
Picon

Re: Antwort: F-contrasts in Marsbar

Dear Klaus,

 

Thanks for your e-mail. I re-checked, but the dependencies were equally defined in both statistical procedures. In SPM, I defined the factor GROUP as “independent” with “unequal variance” and the factor CONDITION as “dependent” with “equal variance”. In SPSS, I chose repeated measures ANOVA, with factor GROUP as the “between-subjects” factor and CONDITION as the “within-subject” factor.

 

I don’t think this causes the differences, because the t-test I compared across MarsBar and SPSS gave very similar results, and that t-test was also a comparison of the difference between the 2 conditions across 2 of the 3 groups. If the dependencies were the source of the problem, then the t-test should also be different.

 

My hunch is that MarsBar somehow treats the 2 x 3 F-test as a one-sided comparison, but I am still unsure how that works...

 

If you have any further ideas, I’d be happy to hear them.

 

Yours, Rick

 

=========================================

Rick C.G. Helmich, MD (Ph.D student)


Donders Institute for Brain, Cognition and Behaviour, Centre for Cognitive Neuroimaging

Radboud UniversityNijmegen

 

P.O. Box 9101
6500 HB Nijmegen
The Netherlands

 

T  00 31 (0)24 36 65077
F  00 31 (0)24 36 10989
E  rick.helmich <at> fcdonders.ru.nl

web: www.fcdonders.ru.nl
=========================================

From: Klaus.Fliessbach <at> ukb.uni-bonn.de [mailto:Klaus.Fliessbach <at> ukb.uni-bonn.de]
Sent: Friday, August 07, 2009 10:55 AM
To: Rick Helmich
Cc: marsbar-users <at> lists.sourceforge.net; 'Matthew Brett'
Subject: Antwort: [Marsbar-users] F-contrasts in Marsbar

 

Dear Rick,

I guess you have already checked this, but are you sure that the dependencies between you cells are correctly defined either in your SPM design and your SPSS statistics (i.e. probably an independent factor (subgroup) and a dependent, repeated measurement factor (condition)? This might be an obvious reason for completely different results when you calculate an interaction between the factors.

Best,
Klaus


Dr. med. Klaus Fliessbach
Department of Epileptology
University of Bonn
Sigmund Freud-Str. 25
D-53105 Bonn
Germany
0049-(0)228-6885264

Attachment (Rick C. G. Helmich (rick.helmich@fcdonders.ru.nl).vcf): text/x-vcard, 421 bytes
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 0 comments |
headers
Klaus.Fliessbach | 7 Aug 10:55
Picon

Antwort: F-contrasts in Marsbar

Dear Rick,

I guess you have already checked this, but are you sure that the dependencies between you cells are correctly defined either in your SPM design and your SPSS statistics (i.e. probably an independent factor (subgroup) and a dependent, repeated measurement factor (condition)? This might be an obvious reason for completely different results when you calculate an interaction between the factors.

Best,
Klaus


Dr. med. Klaus Fliessbach
Department of Epileptology
University of Bonn
Sigmund Freud-Str. 25
D-53105 Bonn
Germany
0049-(0)228-6885264

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 1 comment |
headers
QQ Liu | 7 Aug 11:42
Picon

Ask something about Marsbar software

Dear Matthew Brett and all,
 
    I'm now studying your marsbar software to do some data analysis. I don't have any background about ROIs or fMRIs, so my questions maybe very superficial and I hope you can understand me clearly.
 
    When we run the ROI analysis repeatedly, we feel that's very fussy and easy to make a mistake. So because we do the routine with the same steps only with different design files to set. I want to ask you the relationship between the event type name like 'cl' we defined, which chooses "cl: session 1" and "cl: session 2". I plan to do this work automatically or give "cl" "cm" "dl" "dm"in the event type window as default. Then we cannot new a event type any more.
 
    When installing marsbar-0.42, I found that it will report error when I just installed spm5 in this path like:"D:\Program Files\MATLAB71\toolbox\spm5". The error message said that "D:\Program Files\MATLAB71\toolbox\spm5\spm5\templates\T1.nii doesn't exist.".
So I made a new file named spm5 and in which there was spm5 software. Another prblem I met is that I cannot run the run_preprocess normally. Once I run this program it popup a window to choose the example data file, but not like the phenomenon in the manual, it cannot show any files like *.img files. What's wrong with it or is there anything wrong about my environment? I use it in Windows XP and Matlab 7.1.
 
    It looks that there is some bugs in marsbar. When I choose some optional work to do and then I want to give it up. At this time there are some error messages in the Matlab command window.
 
    I'm a novice of cognitive neuroscience and learning. I was graduated in Pattern Recognition. I'm looking forward to your letters.
 
Thanks.
Best regards!
 
Qianqian Liu


--
祝好！

刘倩倩

13426232824

qqliu16 <at> gmail.com
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 501 comments |
headers
Matthew Brett | 7 Aug 20:16
Picon
Gravatar

Re: Ask something about Marsbar software

Hi,

>     When we run the ROI analysis repeatedly, we feel that's very fussy and
> easy to make a mistake.

Yes - personally I often do a first analysis with the GUI, and then,
for multiple subjects or any significant analysis, I use the batch
mode.

> So because we do the routine with the same steps
> only with different design files to set. I want to ask you the relationship
> between the event type name like 'cl' we defined, which chooses "cl: session
> 1" and "cl: session 2". I plan to do this work automatically or give "cl"
> "cm" "dl" "dm"in the event type window as default. Then we cannot new a
> event type any more.

I'm afraid I don't understand this - can you explain more?

>     When installing marsbar-0.42, I found that it will report error when I
> just installed spm5 in this path like:"D:\Program
> Files\MATLAB71\toolbox\spm5". The error message said that "D:\Program
> Files\MATLAB71\toolbox\spm5\spm5\templates\T1.nii doesn't exist.".

If you installed SPM5, in that place, then that file should exist.
Are you sure you installed SPM5 correctly?

> So I made a new file named spm5 and in which there was spm5 software.
> Another prblem I met is that I cannot run the run_preprocess normally. Once
> I run this program it popup a window to choose the example data file, but
> not like the phenomenon in the manual, it cannot show any files like *.img
> files.

The dialog box is asking you to choose a _directory_ not a file; chose
the root directory with the example data directories within.

>     It looks that there is some bugs in marsbar. When I choose some optional
> work to do and then I want to give it up. At this time there are some error
> messages in the Matlab command window.

Please do ask on the list if you run into an error, and you can't find
anything about it in the documentation.

Best,

Matthew

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
Permalink | Reply | 501 comments |
headers
Matthew Brett | 7 Aug 20:22
Picon
Gravatar

Re: timecourse value

Hi,

> I'm extracting timecourses from SPM design. I know summary functions
> available are mean, weithed mean, median and 1st eigen value. However, I
> would like to use the maximum value. Is there a way to get that? Of course,
> I tried "max" in the script, but with no result.

No, I'm afraid that's not available.  It would be a bit odd as a
summary value, because it could be giving you a value from a different
voxel in the ROI for each time point.  It will probably also emphasize
noise.   Having said that, you could get the values you need with a
script something like (not tested):

design_obj = mardo('SPM.mat');
vols = images(design_obj);
roi_obj = maroi('my_roi.mat');

data = getdata(roi_obj, vols);
mx_data = max(data, 2);

Best,

Matthew

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
Permalink | Reply | 0 comments |
headers
Matthew Brett | 7 Aug 21:02
Picon
Gravatar

Re: F-contrasts in Marsbar

Hi,

> When performing an F-test in an anatomical ROI (left Pallidum) in MarsBar, I
> got a very significant result (subgroup x BMC; F=5.79, 0.003563). However,
> when I extracted the averaged betas in order to plot them, and performed the
> same F-contrast in SPSS, I found that the same interaction effect was far
> from significant (subgroup x BMC, F=1,29; p=0.28). The same difference did
> not apply to a simple t-test in the same design: (MarsBar: t=1.56;
> p=0.060622; SPSS: t=1.649; p=0.108). I understand that MarsBar and SPM use
> one-sided t-tests, explaining the different p-values for the t-test, but how
> does this influence F-statistics? And how should I interpret a significant
> F-statistic in MarsBar?

I'm afraid - following on from Klaus' reply - I don't understand the
correspondence between SPM and SPSS variance models (MarsBaR uses the
SPM machinery).    It would be surprising if the variance estimation
made such a large difference.   The F value and the t value are the
same whether you are doing one-sided or two-sided tests (only the p
value changes) so I don't think that's it either.

Do the betas correspond exactly?  I mean - how specifically are you
doing this model in marsbar?  How are you getting the betas to put
into SPSS?

Best,

Matthew

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
Permalink | Reply | 0 comments |
headers
Shelli Kesler | 17 Aug 21:32
Picon
Favicon

mac matlab select files problem

I'm using matlab 7.8 on an intel mac and when trying to run marsbar, the gui will not show any of my files - it seems to think they don't exist even though they are definitely there.  I read that this was a problem for linux as well and involved in matlab's uigetfile routine.  I have marsbar version 0.42 and am using spm5.  I tried the setappdata(0,'UseNativeSystemDialogs',false) but still cannot select files. I've written matlab about this but was wondering if anyone else is having this problem.

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
_______________________________________________
Marsbar-users mailing list
Marsbar-users <at> lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/marsbar-users
Permalink | Reply | 0 comments |
headers
吴定诚 | 23 Aug 23:11
Picon

about Mean Parameter Estimate

Hi,
    Recently I read an article from J Cogn Neurosci. When the authors
did the ROI analysis, they used Mean Parameter Estimate to copaire
difference of activation between conditions. I'm wondering what's that
Mean Parameter and how can we do that except for extracting percent of
activated voxels. That article is as follows:
    Quadflieg, S., Turk, D. J., Waiter, G. D., Mitchell, J. P.,
Jenkins, A. C., & Macrae, C. N. (2009). Exploring the neural
correlates of social stereotyping. J Cogn Neurosci, 21(8), 1560-1570.

Thanks.

Dingcheng.

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with 
Crystal Reports now.  http://p.sf.net/sfu/bobj-july
Permalink | Reply | 0 comments |

Gmane