Dear Alexandre,
This only holds for regressors specified as parametric modulators in single-subject (first-level) models, also see "The SPM software package automatically performs orthogonalization for parametrically modulated regressors in versions 8 and older". In contrast, first-level regressors specified via conditions or regressors are NOT orthogonalized, regressors in group (second-level) models are NOT orthogonalized automatically.
Whether to go with "manually" orthogonalized regressors is up to you, in any case it should be consistent throughout the paper. E.g. if you conduct a regression analysis on behavioral data with no orthogonalization (this is more or less the default), then probably you would not do for the (f)MRI part. It might also be useful to consider why these regressors correlate in the first place if the correlation is very high, and only take one of them as a predictor or some composite score (e.g. they might both measure the same, e.g. scores on two attentional tests). Possibly also turn to a mediation or moderation analysis depending on hypotheses.
Best
Helmut
|