Evaluation of mixed effects in event-related fMRI studies: impact of first-level design and filtering