 |
 |
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