Gang:

Thank you very much for the clarification.

My goal was to:
1. identify target regions of activation using the full model,
2. determine the full amount of variance explained by all three predictors together, and
3. then examine each predictor separately (3dttest for significance of the beta weight (fit coefficient; regressor coefficient) and the partial R^2 for the unique variance explained by each predictor).

I am trying to tease out how much each predictor (continuous heart rate, respiration and self-reported emotion intensity measures) relates to the BOLD response in specific brain regions during different emotion conditions.

Question:
1. Assessing the assumption of normality: when we convert the t-values to z-values, what would be the best way to test that the distribution of z-values (or t-values for that matter) are normally distributed? Examine the output of 3dhistog of t or z values in all voxels in the brain or perhaps just voxels in an ROI mask? Would it make sense to look at skew and kurtosis?

2. Asessing independence of observations down the time domain: has someone put together a plugin to calculate auto-regression for the time series data?

philippe