Dear AFNI,

I am doing some dry runs through 3dDeconvolve with an experimental design using different durations of jitter. When looking at the output of correlations between the regressors in the design matrix, I was wondering what levels of correlation are considered "high", "medium", "low", and so on. I know that it can depend on what regressors you are especially interested in, and that sometimes high correlations are inherent to the design (such as correlations between polynomials that model baseline, for example). I have heard that correlations above 0.4 are considered high, but I would like to get more perspective on how to interpret correlations within the context of an experimental design (e.g., whether it matters how many trials you have per condition, whether the severity of a correlation can depend on what basis function that is used, etc). I would greatly appreciate any guidance to additional educational material about this issue.


Thanks!

-Andrew