Iroise,
Just out of curiosity, how are you accounting for cortical changes due to
development? Perhaps I didn't clearly understand what the DV is in your
design, but given that you want to do a second-level group analysis, it
seems to me that 'developmental level' not just time should also be in the
model somewhere. I'd be most interested to learn how you (or others) deal
with this issue, as I have similar sets of data.
Cheers
Jason
On 9/17/10 8:12 AM, "Iroise Dumontheil" <[log in to unmask]> wrote:
> Dear SPM community,
>
> We have a longitudinal study of development in a group of 6 to 20 year olds
> For 2/3 of participants we have two usable scans at 2 years interval, for the
> rest with have either the first or the second scan only.
>
> Instead of looking at the results cross-sectionally at time 1 and time 2 and
> longitudinally for those participants we have two scans for, we would like to
> combine all participants in a single 2nd level analysis, similarly to what we
> can do in SPSS using mixed linear models.
>
> We think we have found a way to do this but we would be very grateful if
> somebody could confirm this is the correct way. Somebody else asked a similar
> question on the SPM mailing list in 2004 but they don't seem to have ever
> received an answer:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;6f70bea1.04
>
> The model we are thinking of using is as follow:
> - Flexible factorial design
> - 3 factors: a dummy factor, independent (coded as 1 throughout)
> time, not independent (coded as 1 or 2)
> subject, independent (no level is specified in the model as
> this is one of SPM's reserved factor name)
> - Main effect for the dummy factor only (as we do not want to explicitly test
> for differences between the two testing times or for subject differences)
> - Covariates: age in months
> gender (coded as 1 or 2)
>
> Is this correct?
>
>
> As a follow up question, we would like to test for genotype and genotype x age
> interactions. What would be the best way to code for these covariates? If we
> code genotype as 0,1,2 then the correlation between genotype and genotype x
> age is very high (0.88). Should we instead use the multiplication of the
> Zscore of genotype and Zscore of age (in which case the correlation
> coefficient is <0.15) as the interaction term, or would this completely change
> the meaning of this covariate?
>
> Help regarding this would be really appreciated.
>
> Regards,
>
> Iroise Dumontheil
> Torkel Klingberg's group, Developmental Cognitive Neuroscience
> Karolinska Institute, Stockholm, Sweden
|