Thanks again Tim!
I'm now trying to set up the analysis as you suggested, and a couple of
more questions came up... I know this topic has been discussed already
on the FSL list, but I couldn't find the answer to my particular questions.
1. Is it sufficient to choose the "orthogonalize" option under model
setup within FEAT, or do I also have to demean the performance scores
before I enter them as weights in the third column of EV2?
2. If the "orthogonalize" option is sufficient, is it a problem that
some of the weights of EV2 will be 0 (raw performance scores range from
0 to 6)?
3. How do I setup a contrast that will test the effect of EV2? I have
three regressors: EV1 (condition A), EV2 (modulation of A by
performance) and EV3 (condition B). I also have some short periods of
fixation in my design, which I don't model explicitly. I am mainly
interested in the effect of A>B (EV1=1, EV3=-1). But how do I find out
whether this effect is influenced by performance?
Hope you can help me clarify these things.
Kind regards,
Hanne
Tim Behrens wrote:
> Exactly -
> If you want PE1 to represent the mean activity in condition A, then
> you should orthogonalise EV2 wrt EV1 (this is equivalent to demeaning
> the vlaues in the 3rd col of EV2).
>
> If you do not demean EV2, PE1 will represent the Y-intercept (i.e. the
> activation pattern present when the performance score was 0).
>
> T
>
>
> On 8 Jan 2008, at 16:59, Hanne Lehn wrote:
>
>> Hi again,
>>
>> Thanks for the quick reply.
>>
>> Just to make sure I understand you correctly: If I go for option C, I
>> define EV1 (condition A) and EV2 (weighting). The first two columns
>> in their onset files (onset and duration) should be identical. The
>> third column should be all 1's for EV1 and different values (0-6) for
>> EV2. Ex:
>>
>> EV1:
>> 0.0 28.4 1
>> 64.2 28.3 1
>> 128.3 28.4 1
>> 194.3 28.4 1
>> ...
>>
>> EV2:
>> 0.0 28.4 3
>> 64.2 28.3 1
>> 128.3 28.4 5
>> 194.3 28.4 2
>> ....
>>
>> Is that correct?
>>
>> Kind regards
>> Hanne
>>
>>
>>
>>
>> Tim Behrens wrote:
>>> Hi
>>>
>>> A) would generally work fine.
>>>
>>> The only reason you might not want to use (A) is if you don't have
>>> many blocks per condition
>>> in this case, you can use :
>>> C) Have two EVs - one defines the block timings, and the other is
>>> the weighting EV
>>>
>>> If you just have the weighting EV, you cannot tell whether an effect
>>> is really correlated with weighting, or if it is just condition A
>>> itself that causes the effect
>>>
>>> T
>>>
>>>
>>> On 8 Jan 2008, at 15:31, Hanne Lehn wrote:
>>>
>>>> Hi,
>>>>
>>>> I have data from a block design experiment with two conditions, A
>>>> and B.
>>>> I'm interested in whether activation in condition A is modulated by
>>>> task
>>>> performance. Performance was measured throughout the experiment and
>>>> has a
>>>> separate value (0-6) assigned to each block. How do I set up the
>>>> appropriate first-level analysis? Do I
>>>>
>>>> (A) create 7 EVs for condition A, one for each level of
>>>> performance, and
>>>> compare their effects with the appropriate contrasts?
>>>>
>>>> or
>>>>
>>>> (B) create 1 EV for condition A which includes all blocks, then add
>>>> the
>>>> performance variable as a weighting factor (use performance scores
>>>> instead
>>>> of "1"s in the third column of my custom onset file)?
>>>>
>>>> Hope my question is clear.
>>>> Many thanks in advance!
>>>>
>>>> Hanne
>>>>
>>
>> --
>> Hanne Lehn
>> PhD student Neuroscience
>>
>> MR-Centre, St. Olav's Hospital
>> 7006 Trondheim, Norway
>> Phone: (+47) 73 59 88 04
>> Fax: (+47) 73 86 77 08
>>
--
Hanne Lehn
PhD student Neuroscience
MR-Centre, St. Olav's Hospital
7006 Trondheim, Norway
Phone: (+47) 73 59 88 04
Fax: (+47) 73 86 77 08
|