Hi K,
the models are equivalent (they are linear combinations of each other). The preferred way (at least for the Matlab LME stuff) is the second one.
Note that usually people want to control for age (or test for an age effect). To do so they would have
intercept + time_from_baseline + age_at_baseline + group + sex + groupXtime_from_baseline
Best, Martin
On Jul 9, 2016, at 12:02 AM, K kam.ntnu@gmail.com wrote:
Hello! We are trying to run a longitudinal morphometry analysis with a control group and clinical group scanned twice. We'd like to look at group x time interactions and would like help figuring out the best way to do so. As the time variable we use centered age -- the deviation from the mean age across all subjects and timepoints. To do so, we use the following lme model:
intercept + centeredage.controls + centeredage.clinical + group(0,1) + sex(0,1)
Contrast vector: [0 1 -1 0 0]
where the centeredage column contains only the centered age for that group and zeros for the opposite group. The thinking behind this is to investigate which cortical regions show opposite effects of time between the groups.
However, the FreeSurfer lme software requests a specified "time column". Obviously, we cannot add a centered age for all subjects, nor does it seem to make sense to use one of the two group-specific centeredage columns as the time column.
Our alternative model for investigating interatction is:
intercept + centeredage + group(0,1) + sex(0,1) + groupxcenteredage
where centeredage is the time columns and is a common regressor for both groups and thus, groupxcenteredage becomes 0 for control subjects. Contrast vector: [0 0 0 0 1] This would test whether there is an added effect of centered age in the clinical group.
Is there a better way to do an group x time interaction analysis here, or is there a solution for the time column? Thanks!
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer