Hi,
Thank you very much for looking at my design matrices. I have just one
additional question:
>>1) How BEH correlates with FA in the whole sample?
>>
>> Group EV1(beh, demeaned values)
>> 1 -1
>> 1 -3
>> 1 -5
>> 1 3
>> 1 1
>> 1 5
>>
>> Contrast : -1 (I expect negative correlation)
>>
>> Randomise with -D option
>
> Sure - looks fine.
>
>>
>>
>> Or if the two subgroups have different means for BEH, should I split
>> EV1
>> into two Evs:
>> EV1 Ev2
>> -1 0
>> -3 0
>> -5 0
>> 0 3
>> 0 1
>> 0 5
>> And then skip -D?
>
> Not quite - you need to demean each groups values before padding with
> zeros. And you DO need the -D option.
Let's assume I demeaned the values for each group separately and the design
matrix is correct. What does it really give me to split the two subgroups
into 2 separate EVs? Should this be always done if the means of these
subgroups for these EV differ?
Best regards,
Aga
On 10/29/08 2:00 PM, "Steve Smith" <[log in to unmask]> wrote:
> Hi - the fslroi, fslmerge and fslsplit programs are very simple, so it
> seems unlikely that there is a bug in them. Are you absolutely sure
> about exactly what commands you're typing in this?
>
> More comments below....
>
>
> On 28 Oct 2008, at 18:09, Agnieszka Burzynska wrote:
>
>> Dear all,
>>
>> I would like to correlate a behavioral measure with FA values on tbss
>> skeleton. I wanted to do it for 2 subgroups of my sample separately,
>> but
>> I am not able to split the all_FA_skeletonized file (here I am
>> referring to
>> my email "Fslroi on all_FA_skeletonised - output does not match the
>> input").
>>
>> In case you have an idea how to overcome the problem, I will be very
>> grateful.
>>
>> In meantime I wanted to run some analyses with a design matrix for all
>> subjects together.
>>
>> Here are some of my designs and I wanted to ask if they make sense
>> the way i
>> describe them (I read previous posts but I am still not sure what is
>> the
>> best to answer my questions). Unfortunately I have no experience
>> with FEAT
>> design matrices, hence these basic questions.
>>
>> As I mentioned before, I have 2 subgroups (younger and older), in
>> designs
>> below first 3 entries belong to 1 group, the remaining 3 to the other.
>>
>> 1) How BEH correlates with FA in the whole sample?
>>
>> Group EV1(beh, demeaned values)
>> 1 -1
>> 1 -3
>> 1 -5
>> 1 3
>> 1 1
>> 1 5
>>
>> Contrast : -1 (I expect negative correlation)
>>
>> Randomise with -D option
>
> Sure - looks fine.
>
>>
>>
>> Or if the two subgroups have different means for BEH, should I split
>> EV1
>> into two Evs:
>> EV1 Ev2
>> -1 0
>> -3 0
>> -5 0
>> 0 3
>> 0 1
>> 0 5
>> And then skip -D?
>
> Not quite - you need to demean each groups values before padding with
> zeros. And you DO need the -D option.
>
>
>>
>>
>> 2) How BEH correlates with FA after regressing out for age?
>>
>> Group EV1(beh, demeaned values) EV2 (AGE, demeaned)
>> 1 -1 -12
>> 1 -3 -10
>> 1 -5 -7
>> 1 3 12
>> 1 1 7
>> 1 5 10
>>
>> contrasts: -1 0
>> Randomise with -D (as there is no EV that models the mean)
>
> Fine.
>
>>
>>
>> Or would it make more sense to orthogonalize EV1 with respect to EV2
>> by
>> clicking it in Glm_gui? (if I know that BEH and age are not highly
>> correlated?)
>
> You can also do that - depends on what you want the model to do.
>
>>
>>
>> 3) How FA correlates with BEH, after regressing out for agegroup to
>> which a
>> subject belongs?
>>
>> Group EV1 (young) EV2 (old) EV1(beh, demeaned values)
>> 1 1 0 -1
>> 1 1 0 -3
>> 1 1 0 -5
>> 2 0 1 3
>> 2 0 1 1
>> 2 0 1 5
>>
>> Contrasts 0 0 -1
>> 1 -1 0 (where younger have higher FA than older, after regressing
>> out for
>> BEH?)
>> Having EV1 and EV2 that model the mean, I dont need to add -D
>> option. Is it
>> correct?
>
> Correct
>
>
>>
>>
>> 4)
>> Group EV1(age, demeaned values) EV2 (BEH young, demeaned) EV3(BEH
>> old)
>> 1 -1 -12 0
>> 1 -3 -10 0
>> 1 -5 -7 0
>> 1 3 0 12
>> 1 1 0 7
>> 1 5 0 10
>>
>> Contrast 0 1 -1 (would it tell me where BEH correlates with FA
>> higher in
>> young than old adults, after regressing out for age?)
>
> Yes but as above you need to demean the sub-group values before
> padding with zeros
>
>>
>>
>> 5) or would it be more meaningful to make an interaction analysis
>> (Beh and
>> age interaction) using F-tests? If yes, how should it look like?
>
> No - I would use the above examples.
>
> Cheers.
>
>
>>
>>
>>
>> Thank you very much for help in advance,
>> Best,
>> Aga
>>
>
>
> ---------------------------------------------------------------------------
> Stephen M. Smith, Professor of Biomedical Engineering
> Associate Director, Oxford University FMRIB Centre
>
> FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
> ---------------------------------------------------------------------------
|