Hi David,
if SNQ is a between-subject variable, it's not to be used at 1st level,
but rather as a 2nd level covariate.
For 1st level, you must specify duration for each trial in the 1st
level model specification- possibly with a parametric modulation if each
trial has some kind of varying values assigned to it.
Best, Brian
On 2015-11-16 14:35, David Pollard wrote:
> This may be a basic question but I've run into some trouble and I
> need help completing my first SPM analysis.
>
> We have performed an experiment that investigates appraisals of peer
> relationships using fMRI and a social network questionnaire (SNQ). So
> during the first-level analysis specification I want to include the
> values given by the SNQ in the model as a regressor.
>
> There were 48 trials in the session and 150 scans. There was 1
> regressor value for each trial. SPM says that the length of the
> regressor vector must match the length of the scans vector. Each
> trial
> took 9 seconds on average (range = 8 - 10 seconds).
>
> So I have a problem with creating a regressor vector that I can input
> into SPM.
>
> What I have done to resolve this problem is write a function that
> repeats each regressor value 3 times (TR = 3; 3xTR = ~9 second
> trial).
> I have also input null values in relevant areas between trials.
>
> However, this generally gives me a regressor vector length that does
> not quite match the scan vector length because the trial times are
> not
> consistent or perfectly divisible by TR; the overall length is
> usually
> out by 1-5 values.
>
> So right now to resolve this the function adds or removes 0 values in
> areas of the vector that are not important i.e. outside of trial
> times.
>
> This method is not perfect and will still lead to regressor values
> being slightly out of place but I can't think of an alternative
> method.
>
> QUESTION:
>
> Is this a good way of creating a regressor vector? Is there a better
> way that I am missing? Am I even thinking of the issue in the correct
> way as I am trying to marry scan data and regressor data (are they
> both similar enough data i.e. should I think of the scans as
> sequential like the regressors?).
>
> Please let me know if you have any questions or if my question is
> unintelligible.
>
> Cheers,
>
> Dave
|