Hi Patrizia,

Usually you would allow the intercept to be a random effect (1st column). Basically random effect means, you allow each subject to have their own intercept, which usually makes sense. That could be sufficient for your model, you can also test if having two random effects is better (including the time slope, so the second column). This allows also the individual slopes to be different from each other. It is usually a good idea to keep the number of random effects small.

Best, Martin

On 04/20/2016 06:49 AM, Patrizia Dall'Acqua wrote:

Dear FS-Group

I am attempting to use the lme matlab tools for my longitudinal analysis.

My dataset consists of 2 time-points and 2 groups (patients and controls). I will not use covariates since the groups are well matched for age, gender and education.

I am trying to adapt the wiki instructions of the mass-univariate model to our data.

Our design matrix X contains 4 columns:

1. intercept

2. time

3. group

4. interaction group*time

Our Cmd history:

[Y,mri] = fs_read_Y('lh.thickness_sm15.mgh');

lhsphere = fs_read_surf('/subjects/average_LTHV_long_all_49/surf/lh.sphere');

lhcortex = fs_read_label('/subjects/average_LTHV_long_all_49/label/lh.cortex.label');

Qdec = fReadQdec('qdec.long.table_4columns.txt');

Qdec = rmQdecCol(Qdec,1);

sID = Qdec(2:end,1);

Qdec = rmQdecCol(Qdec,1);

M = Qdec2num(Qdec);

[M,Y,ni] = sortData(M,1,Y,sID);

X = [ones(length(M),1) M M(:,1).*M(:,2)];

lhstats = lme_mass_fit_vw(X, [4], Y, ni, lhcortex)

I am mainly interested in comparing longitudinal changes between groups (group*time interaction) but also interested in the main effect of group and time.

I have two questions:

1. I am not sure about my „interpretation“ of random and fixed

effect.Since I am interested in the group*time interaction (4. column), does it make sense to use a single random effect for this interaction or should I take a second random effect for time (2. column)?

2. If a model with two random effects make sense (2. and 4. column),

should I compare the model including two random effects with a model with a singlerandom effect to test which one is significantly better?

Thank you for your help!

Patrizia

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer