Yes, I understand that you were inquiring about controlling for gender
across subjects (gender presumably isn't changing within a subject
unless you're conducting a study involving sex change operations!). But
in a repeated measures design, in which you have an EV with a column of
1's at your 4 timepoints for each subject (to model the mean of that
specific subject), and you have such an EV for each subject, then your
gender EV can always be expressed as linear combination of the subject
specific mean EVs, and thus you would have a rank deficient design
matrix. The same reasoning applies to age if you were to use the same
age value at all 4 timepoints for a given subject.
cheers,
-MH
On Thu, 2011-05-12 at 20:14 +0200, Alexander Olsen wrote:
> Ok. I see. However, I don't thing this was what I wanted to do. What I
> want to do is to control for age and gender differences between
> subjects, not within subjects (e.g. control for the fact that I have
> more younger than older participants, look at gender*time effects etc.).
> Sorry if I'm not being very clear.
>
> Alexander
>
> Michael Harms wrote:
> > You can't covary for gender in your repeated measure model, for the
> > reason that I explained. Try including a gender covariate and you'll
> > find that your design matrix will be rank deficient. As for age, it
> > would only be an option to consider if you want to use a different value
> > for age for each repeated visit of a subject, and even then I think
> > you'll get a poorly conditioned design matrix unless your age at the
> > repeated visits for a subject differed in an appreciable manner (where
> > "appreciable" would be relative to the variation in ages across
> > subjects).
> >
> > cheers,
> > -MH
> >
> > On Thu, 2011-05-12 at 19:44 +0200, Alexander Olsen wrote:
> >
> >> Hi Michael,
> >> Thank you so much for helping me. So that means that I should set up the
> >> design matrices without any covariates for age and gender?
> >>
> >> cheers,
> >> Alexander
> >>
> >>
> >> Michael Harms wrote:
> >>
> >>> Hi Alexander,
> >>> Given that you are modeling the mean of each subject using a repeated
> >>> measures design, that mean already incorporates that subject's age and
> >>> gender as part of its estimate. That is, if you added EVs for age
> >>> and/or gender, those EVs can be represented as a linear combination of
> >>> the columns modeling each subject's overall mean, and thus you would get
> >>> a degenerate design matrix. (This wouldn't strictly be true for age IF
> >>> the value you used for age differed across the repeated scans of a
> >>> subject, but unless your scans were spaced far apart temporally, you
> >>> still might get a poorly conditioned design matrix).
> >>>
> >>> cheers,
> >>> -MH
> >>>
> >>> On Thu, 2011-05-12 at 01:12 +0100, Alexander Olsen wrote:
> >>>
> >>>
> >>>> Dear FLS experts,
> >>>>
> >>>> I'm struggling a bit with setting up my analysis. What I would like to do is a repeated measures design with four timepoints, as in this example:
> >>>>
> >>>> http://www.fmrib.ox.ac.uk/fsl/feat5/detail.html#ANOVA1factor4levelsRepeatedMeasures
> >>>>
> >>>> However, I would also like to add age and gender as covariates in order to "remove" these effects. I hope anyone can help me, or direct me to an example which describes this.
> >>>>
> >>>> Thank you so much in advance.
> >>>>
> >>>> Alexander
> >>>>
> >>>>
|