Dear Ariadna,
you should model all details you know about the experiment, even if you are not interested in testing some of them. Unmodelled effects will contribute to residual variance, which is used to estimate temporal autocorrelation of your data. SPM uses a weighted least squares estimation which takes this autocorrelation into account. A wrong autocorrelation estimate might therefore affect the effect size estimates on first level, even if you do not plan to assess single-subject results.
Hope this helps
Volkmar
|