July 2nd, 2014, 9:28 am
There are probably several ways to do this, but I would run PCA on the residuals. That way, not only will the end result be orthogonal, but you have a natural ordering of them in terms of their "importance", i.e. the amount of variance they explain.