Dear Donald,

Thank you very much for your help.

Indeed, I have 1 parametric modulator and 3 basis functions (i.e. A1, A2, A3, A1xPmod, A2xPmod, A3xPmod; sorry for the confusion). I still have some questions regarding the interpretation of the contrasts at the 2nd level, whether they are computed out of a flexible factorial design or two-sampled t-tests.

In a given significant volume resulting from a test contrasting e.g. Group 1 positively (+1) and Group 2 negatively (-1) :
1) A1xPmod : the canonical HRF in response to A is more modulated by Pmod in Group 1 than in Group 2?
2) A2xPmod : the HRF time derivative in response to A is more modulated by Pmod in Group 1 than in Group 2?
3) A3xPmod : the HRF dispersion derivative in response to A is more modulated by Pmod in Group 1 than in Group 2?
Does SPM allows to be any more specific in the interpretation of these results?

Moreover, does e.g. a positive contrast (+1) tests for a positive correlation between Pmod and betas? Does it mean higher values of Pmod are associated with :
1) A1xPmod : a higher amplitude HR?
2) A2xPmod : an earlier peak HR?
3) A3xPmod : a less dispersed HR?

Many thanks in advance!

Nicolas Martin


-----Original Message-----
From: MCLAREN, Donald [mailto:[log in to unmask]]
Sent: Thu 2013-06-13 03:20
To: Martin Nicolas
Cc: [log in to unmask]
Subject: Re: [SPM] Parametric modulator, HRF informed basis set, & independant groups
 
Please see inline responses below.


On Wed, Jun 12, 2013 at 8:31 AM, Martin Nicolas <
[log in to unmask]> wrote:

> Dear SPM list,
>
> I wish to analyse the effects of one parametric modulator (Pmod) on the
> HRF informed basis set (i.e. three basis functions) in two independant
> subject groups. My problem regards 1) the contrasts best suited to answer
> my research questions and 2) the interpretation of such contrasts. Any
> guiding advice will be much appreciated!
>
> At the 1st level, individual design matrices included condition regressors
> (A) and parametric modulator regressors (AxPmod), both convolved with the
> three basis functions (i.e. A1, A2, A3, AxPmod1, AxPmod2, AxPmod3). Basic
> t-contrasts on the AxPmod regressors resulted in three con images for each
> subject.
>

Do you have 1 Parametric modulator and 3 basis functions (HRF, first
deriv., second deriv.) or do you have 3 parametric modulators?


>
> At the 2nd level, I used a 3x2 Full factorial design with Basis set as
> factor #1 and Group as factor #2. Now, here is where I need guidance :
>
> The following F-contrast :
> 1 -1 0 0 0 0
> 0 0 1 -1 0 0
> 0 0 0 0 1 -1
> would answer the question : "Is there a group difference in the modulation
> of the shape of the HRF by Pmod?", but would not inform the nature of the
> difference; e.g. in a given significant volume, the correlation between
> Pmod and betas is
> a) positive in Group 1 and negative in Group 2?
> b) more positive in Group 1 than in Group 2?
> c) null in Group 1 and positive in Group 2? etc...
> Am I correct? How can I assess the directivity of the slopes to interpret
> the results?
>

This is not a statistically valid test, irrespective of the answer above.
Each row is testing a group difference (between-subject effect), which
isn't valid in a repeated measures design. Also, for repeated measures, you
should use the flexible factorial with a subject term. I would use 3
two-sample t-tests to tes the effects of the 3 con images in your case as
you are interested in the group effect of each of them.


>
> Thank you very much for addressing this issue!!!
>
> ==
> Nicolas Martin, Ph.D. candidate
> Université de Montréal
> Center for advanced research in sleep medicine
> Montreal (QC), Canada
>