Hi James,
Figure I'd take a shot at your questions- in hopes that any error in my response will be corrected by the experts.
First, if your end goal is a regression at what is called the 2nd-level (group comparison)- meaning inquiring whether BOLD is related to an independent variable, like age, for each subject, you would enter the independent variable as a vector (one number/subject) in the covariates section of the multiple regression.
At the 1st-level, where you specify the model for each subject, the parametric modulations section is used to include, for ex., motion regressors from realignment. Here, you have a number for each scan in your fMRI session (maybe 200-300).
The contrast depends on your task/model and question. A simple fMRI may have 2 task conditions (A, B) separated by rest. If you were interested in task A minus task B, and you entered the conditions into the model in the order task A, task B, rest (take a look at the design matrix for order of conditions), then the contrast weights would be:
1 -1
Specifically, in the contrast manager window, select t-contrasts (see below), Define new contrast..., enter the name for the contrast (doesn't matter- "A-B") in name box, and enter contrast weights (above) in contrast box- click submit- should say contrast valid; click OK. Click on contrast name in contrast manager and Done to continue to results section (here just choose defaults to generate *con file which you will use in 2nd-level analysis).
As regards difference between t and F-tests, wikipedia probably has a good explanation or any beginner stats textbook. Basically, at the 1st-level, majority of the time you be interested in t-test as F-test has no directionality, but there may be circumstances when F-tests are used at 1st-level. At 2nd-level F-tests are used in factorial (analysis of variance/ANOVA), but need to be followed with t-tests to determine which groups differ and/or directionality.
As regards SPM manual, take a look at examples of data analysis for more info.
To do a regression at the 2nd level, within Multiple regression, enter *con files for each subject under Scans and your independent variable (vector with 1 number/subject) under Covariates. You will need to specify contrasts again; a contrast weight of "1" will give you positive relationship between BOLD and covariate; "-1" the opposite.
Hope this is helpful,
Julie
Julie E. McEntee, M.A., C.C.R.P.
Senior Research Program Coordinator
Department of Psychiatry- Neuroimaging
Johns Hopkins University School of Medicine
600 N. Wolfe St./ Phipps 300
Baltimore, MD 21287
Phone: 410-502-0468
Fax: 410-614-3676
----- Original Message -----
From: James <[log in to unmask]>
Date: Thursday, February 11, 2010 9:08 am
Subject: [SPM] SOS (3rd time): How to construct contrast for parametric modulation?
To: [log in to unmask]
> Dear SPMERS,
> Could you help me with the following questions. I have inquired them
> twice in
> this forum but got no response at all. This is the 3rd time; wish to
> get some
> response this time.
> Looking forward to your reply.
> Thank you very much!
> James
>
> I need to do regression analysis (the independent variables in the
> regression
> are independent variables in the experiment design). I heard that it
> should use
> the "parametric modulations", as highlighted in the attached picture,
> right?
> (pls check the highlight in the attached picture, thanks!)
>
> After implementing the parametric analysis, there will be no contrast
> in the
> contrast window. how to define the contrast (pls very specially)? in
> the define
> contrast window, there are t-test and F-test and we have to choose
> one -
> which one to choose and what to do after that?
>
> could you also tell me which file(s) on internet (preferably on
> www.fil.ion.ucl.ac.uk/spm) or which book(s) do address exactly these
>
> questions (I cannot find it in SPM8 manual.)
>
|