1) Is there a better and faster method than 3dRegAna to create such maps.

As far as I know, currently there is no better solution than 3dRegAna. The reason for the slow speed of running 3dRegAna is probably due to the huge temporary files involved and also heavy-duty matrix operations.

2) How should I set the -rsmin option for these measures?

It just takes too much time to pick up a right threshold value with trial-and-error, but other than the suggestion given in the manual (a value much smaller than the natural measurement error), I don't know any alternatives which would not bear the risk of screening out too many voxels than necessary.

3) Would it be preferable to convert the reaction times in z-scores based on mean reaction times and its standard deviations?

I assume that the behavioral measures are raw data, and I don't see any need for z-score conversion. Are the behavioral measures some statistic values?

4) I used the '-bucket 0 name' option, because I found the other ones to difficult to understand. How should I look at my data. I have b0 and b1 subbriks and corresponding t-statistics. There is also a R2 and F-stat subbrik. It says in the manual that b0 and b1 would not so much differ. Only the colorizing would change.

First, the estimate of b0 and that of b1 are usually different, as you have noticed, because one is the constant and other is the slope of the linear dependence relationship. As people are mostly not interested in the inference of intercept b0, no statistic values are provided for b0 in AFNI output; Instead the statistic values of the linear regression, same thing as the ones for slope estimate b1, are given in the dataset.

Has anybody tried to put in orthogonalized and mean centered polynomial expansions of behavioral data in 3dRegAna (see papers of Christian Buechel)? And how would one set up the -model option, if I have two columns (linear + quadratic measure) in order to test the fit for both?

For quadratic fitting, usually you set the option as '-model 2 1 : 0' -- a full model with both linear and quadratic terms versus a reduced one with only a constant. See example two in the manual.

Gang