Dear Matthew,
I got it. Thanks!
Best, Daqiang
----- Original Message ----
From: Matthew Webster <[log in to unmask]>
To: [log in to unmask]
Sent: Fri, September 10, 2010 5:28:53 AM
Subject: Re: [FSL] randomise on multiple groups
Hi,
Randomise will orthogonalise the design and data against any column in the
design that is excluded by a zero in the contrast. The simplest way to express a
3-group design is with separate columns ( where a 1 indicates membership of that
column's group ).
Many Regards
Matthew
> Dear FSL experts,
>
> A question about running randomise on 3 groups of FA data, partly following the
>
> "tbss on subgroups" post:
>https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0906&L=FSL&P=R27740&1=FSL&9=A&I=-3&J=on&d=No+Match%3BMatch%3BMatches&z=4
>4
>
>
> If I include all groups in the design matrix and want to test the difference
> between group 1 and 2, would the dummy codes for these two groups also enter
>the
>
> time series block of group 3 in certain randomisation runs so that group
>members
>
> are not exclusive in those runs? Or would randomisation only happen among group
>
> 1 and group, with group 3 screened out?
>
> Also another question about design matrix in this case:
> I found in some posts the design is set up in the way that a full column of '1'
>
> is used for baseline group mean and the other two (1,0) columns are used for
> differences from the baseline group, which is like the treatment contrast in R
> etc. Is this preferred for randomise purpose to the way in which all group
>means
>
> are modeled separately with (1,0) columns? I realize that the contrasts are not
>
> the same.
>
> Thanks in advance for any feedback!
>
> Best, Daqiang
>
>
>
>
>
|