Thanks a lot Mark!
Cheers from San Diego,
John Keltner
________________________________________
From: FSL - FMRIB's Software Library [[log in to unmask]] on behalf of Stephen Smith [[log in to unmask]]
Sent: Saturday, June 16, 2012 8:24 PM
To: [log in to unmask]
Subject: Re: [FSL] Is FSL VBM mixed effects modeling possible?
Hi - if you want to treat the difference scanners as categorical variables
and not assume the response across them, then just have one EV per
scanner, with zeros and ones in.
Cheers.
> Hello,
>
> This is a follow up question(s), thanks very much to Mark for his previous
> reply!
>
> I have 241 subjects who have had T1 images acquired on 7 different
> scanners at 5 different locations in the United States.
>
> When I analyze data for these subjects using the statistical tool JMP I
> control for scanners by labeling the scanner variable as a numeric nominal
> variable which implicitly assumes that there is not a linear effect from
> the scanners.
>
> I understand from your helpful previous email response that in my FSL VBM
> GLM design matrix that I may model these different scanners using a
> categorical variable with values such as -3 -2 -1 0 1 2 3.
>
> My limited understanding is that if I model the scanner variable as a
> categorical variable in this manner (-3 -2 -1 0 1 2 3) it is implicit that
> I think there will be a linear effect from the different scanner sites.
>
> In fact, I have no idea what the effect will be from these different
> scanner sites and I strongly suspect the effect is not linear.
>
> Up front, I am a permanent newbie to statistical modeling. Having said
> that it seems to me that the continuous variable and categorical variables
> in the GLM design matrix model fixed effects for VBM analysis.
>
> In my FSL VBM analysis I would like to model variables such as scanner (7
> scanners) or race (4 categories of race) as random effects so that
> ultimately I want to be doing a mixed effects model: random effects (i.e.
> scanner AND/OR race) + fixed effects(i.e. age[continuous variable] AND/OR
> #signs of neuropathy[categorical variable]).
>
> Mark previously confirmed that my use of the randomise command in my FSL
> VBM analysis is correct - i.e. randomise -i GM_mod_merg_s3 -o
> GM_mod_merg_s3 -m mask.nii.gz -d design.mat -t design.con -n 5000 -T
> –V.
>
> Is it possible to model random effects as well as fixed effects using the
> randomise command?
>
> If not, is there some other method for introducing random effects modeling
> along with fixed effects modeling using FSL VBM analysis?
>
> One final question about modeling signs of neuropathy as a categorical
> variable in my GLM design matrix. In my JMP analysis I have a categorical
> variable '# of signs of neuropathy' which is either 0, 1, or 2. When I
> enter this variable in my GLM design matrix should I enter the variable
> values as 0, 1, or 2 OR should I enter the values as -1, 0, or -1?
>
> Thank you very much in advance!
> John Keltner
>
>
|