Hi Donald and Chao and all SPMers,

Thank you very much for your reply. I guess I found what went wrong, and it seems to be SPM's fault but not mine.

I rechecked my batch to see if I had paired the con*.img files wrong, and they were all paired correctedly as con_0001.img & con_0002.img for each subject. I then estimated it again and reviewed the design matrix, which was like
1 0   1 0 0 0 0 0 0 0 0 0
0 1   1 0 0 0 0 0 0 0 0 0
1 0   0 1 0 0 0 0 0 0 0 0
0 1   0 1 0 0 0 0 0 0 0 0
1 0   0 0 1 0 0 0 0 0 0 0
0 1   0 0 1 0 0 0 0 0 0 0
1 0   0 0 0 1 0 0 0 0 0 0
0 1   0 0 0 1 0 0 0 0 0 0
...
such that the first two columns correspond to con_0001 and con_0002, and the rest ones are subjects. However, when I checked "Files and factors", it was like
#img   sF3   sF2   filename tails
001     01     01     Sub01_con_0001.img,1
002     02     01     Sub02_con_0001.img,1
003     01     02     Sub03_con_0001.img,1
004     02     02     Sub04_con_0001.img,1
005     01     03     Sub05_con_0001.img,1
006     02     03     Sub06_con_0001.img,1
007     01     04     Sub07_con_0001.img,1
008     02     04     Sub08_con_0001.img,1
...
020     02     10     Sub20_con_0001.img,1
021     01     11     Sub01_con_0002.img,1
022     02     11     Sub02_con_0002.img,1
023     01     12     Sub03_con_0002.img,1
024     02     12     Sub04_con_0002.img,1
...
I think #img corresponds to the line number of the design matrix, sF3 corresponds to con_0001/con_0002, and sF2 is the subject number. If I'm right about this, then what SPM did is totally wrong, because it made me compare the wrong con*.img's using [1 -1].

As Donald suggested, I opened SPM.mat and looked into SPM.xY.P, which is a 20x2 cell, the first column containing con_0001.img files and the second containing con_0002.img files. I then looked into SPM.xY.VY.fname, where SPM.xY.VY(1:20).fname are con_0001.img's, SPM.xY.VY(21:40).fname are con_0002.img's, which are also incorrect.

I haven't updated SPM for a while. Hoping this bug has disappeared in later versions and not affected many people's analysis. Or, if I did some steps wrong to cause this glitch, please help me out.

Thanks a lot!

Best,
Zhenhao



-----

Zhenhao SHI 石振昊
Culture and Social Cognitive Neuroscience Lab
Department of Psychology
Peking University
5 Yiheyuan Road
Beijing 100871, P.R.China
Phone: 86 134 6655 0474
Email: [log in to unmask]
http://www.psy.pku.edu.cn/LABS/CSCN_lab


On Fri, Nov 9, 2012 at 2:09 PM, Chao Suo <[log in to unmask]> wrote:

Dear Zhenhao,

I asked myself the similar question before.

I think it is wrong by saying “the paired t-test on A>Neutral and B>Neutral should be identical  to the one-sample t-test on A>B, right? “. I guess you assume that One way t-test is “calculate the average”; and paired t-test is not “minus each other”, which is not that simple.

 

The contrast you put in is calculating the beta values which is means how well this activation(intensity) of this voxel fits a particular factor in your GLM design. It means your beta values will be different if you use two different second level design.

 

BTW, “MriCroN, I see that con_0003.img and con_1vs2.img have the same voxel values” is correct, as they are in the same GLM design. So you can get this by calculating the contrast (suppose your order is A B N) : [1 0 -1]-[0 1 -1]=[1 -1 0] which is A>B.

 

Hope this help.

 

Cheers

 

Chao Suo

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

PhD Candidate, School of Psychiatry, UNSW, Australia

Brain & Mind Research Institute THE UNIVERSITY OF SYDNEY

Room 401, Level 4, M02K | 100 Mallett St Camperdown | NSW | 2050

T +61 2 9351 0728  | F +61 2 9351 0551 | Web: rng.org.au/chao-suo-2/

 

From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Zhenhao Shi
Sent: Friday, 9 November 2012 12:49 AM
To: [log in to unmask]
Subject: [SPM] One-sample t-test versus paired t-test

 

Dear SPMers,

 

I have a question regarding the difference between one-sample t-test and paired t-test in SPM8.

 

I got one single group of 20 subjects with Condition A, Condition B, and a Neutral baseline, and I would like to compare A and B. In the first level, I did three T-contrasts for each subject: "A > Neutral", "B > Neutral", and "A > B", which resulted in three contrast images: con_0001.img, con_0002.img, con_0003.img. In the second level, I did a one-sample t-test using the files of con_0003.img from all subjects, and entered  [1] for a t-contrast, which gave me reasonable brain activations for A > B. After that, just out of curiosity, I tried paired t-test and entered con_0001.img and con_0002.img in pairs each subject at a time. The design matrix looks like that the first two columns are of my interest and the next 20 columns are for each subject. I then entered [1 -1] for a t-contrast, but obtained totally different results from those in the previous one-sample t-test. From my understanding, the paired t-test on A>Neutral and B>Neutral should be identical  to the one-sample t-test on A>B, right?

 

To figure out what went wrong, I tried using spm_imcalc_ui to manually substract con_0002.img from con_0001.img for each subject, and got another 20 new image files I named as con_1vs2.img. Using MriCroN, I see that con_0003.img and con_1vs2.img have the same voxel values. So, it looks as if my first level analysis was right. Then could the different results of one-sample t-test and paired t-test be caused by their different algorithms? Or there must be some mistakes I managed not to notice?

 

Looking forward to you experts' reply. Thanks in advance!

 

Best,

Z

 


-----

Zhenhao SHI
Culture and Social Cognitive Neuroscience Lab
Department of Psychology
Peking University
5 Yiheyuan Road
Beijing 100871, P.R.China
Email: [log in to unmask]
http://www.psy.pku.edu.cn/LABS/CSCN_lab