Hi,
in a first-level FEAT run the data gets de-meaned, i.e. you only need
to model deviations from rest in your design. In your case this means
that you need 2EVs (context, no context). You can then use 1 -1 and
-1 1 contrasts (in terms of 'original EVs') to ask about
differences between the two conditions. The "context vs. rest"
question then is a question of where there are parameter estimates
>0, i.e. the relevant contrasts are 10 and 0 1 (again in terms of
original EVs).
In FEAT you can choose to specify the contrasts in terms of the
'original EVs' or in terms of 'real EVs', the distinction is as follows:
'original EVs' refers to what you typically conceptualise as an event
type, e.g. the context condition. However, such a single event train
might be modelled by more than one column in the final design matrix,
e.g. you might choose to include the temporal derivative in the
design. In this case, the original set of events get modelled by 2
columns in the design, i.e. the actual on/off block vector and it's
temporal derivative (in order to account for slight miss-timings).
Similarly, if you choose to use basis functions to model an event
train this often results in 3 or more columns in the design matrix
for each original event train of interest. In some cases you might
want to include these additional vectors in a contrast. In this case
you can specify the contrast using the 'real EVs' setting. In your
particular case, unless you've switched off the 'Add temporal
derivative' option, you have 2 original EVs but 4 real columns in the
design, the second and fourth (the temporal derivatives) are of no
primary interest. For this reason you should probably stick to the
'original EVs' setting and use the following contrasts:
1 -1 (context > no-context)
-1 1 (context < no-context)
1 0 (context > 0)
-1 0 (context < 0)
0 1 (no-context >0)
0 -1 (no-context <0)
Given that FEAT de-means the data for a first-level analysis the
final 4 contrasts effectively answer the "... vs rest" questions"
cheers
Christian
On 2 Mar 2007, at 23:34, Ananth Narayanan wrote:
> Suppose I have 2 EVs, context and noContext. Times of stimuli are
> given as
> follows
>
> NoContext 30
> Rest 18
> Context 30
> Rest 18
> NoContext 30
> Rest 18
> Context 30
> Rest 18
> NoContext 30
> Rest 18
> Context 30
> Rest 18
> NoContext 30
> Rest 18
> Context 30
> Rest 18
> NoContext 30
>
> My primary goal is find where Context activation is greater than
> NoContext
> activation.
> Should I do 3 EVs, Context, NoContext and Rest? If so, I selected 3
> EVs in
> the full model setup, and defined three contrasts for "Real EVs" as
> follows
>
> context vs Rest, 1 0 0 0 -1 0
> NoContext vs Rest 0 0 1 0 -1 0
> Context vs NoContext 1 0 -1 0 0 0
>
> In this case, how would real EVs come into play?
>
> Or should I do 2 EVs, Context and NoContext only? If so, I am
> unsure of how
> to define contrasts. How should I define "Context vs Rest" contrast?
> context vs Rest, 1 0 0 0 (Method 1)
> context vs Rest, 1 -1 1 0 (Method 2)
> Or for that matter, the "NoContrast vs Rest" contrast?
> NoContext vs Rest 0 0 1 0 (Method 1)
> NoContext vs Rest 1 0 1 -1 (Method 2)
> Or even more confusing, "Context vs NoContext" contrast?
> Context vs NoContext 1 0 -1 0 (Method 1)
> Context vs NoContext 2 -1 -2 -1 (Method 2)
>
> If I choose (Method 1), how would the software distinguish between
> REst and
> the other EV? And if I choose (Method 2), is my definition of the
> "Context
> vs NoContext" contrast correct? Also, if (Method 2) is correct,
> what would
> the contrast I defined in (Method 1) give me?
>
> I guess my confusion comes partly from my lack of understanding of the
> statistics behind this and partly from not knowing what "Real EVs"
> mean. I
> would be very grateful if someone can please explain this to me.
> Thanks.
--
Christian F. Beckmann
Oxford University Centre for Functional
Magnetic Resonance Imaging of the Brain,
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
Email: [log in to unmask] - http://www.fmrib.ox.ac.uk/~beckmann/
Phone: +44(0)1865 222551 Fax: +44(0)1865 222717
|
