Partitioning of the independent and joint contributions of each variable in a multivariate data set, to a linear regression by hierarchical decomposition of goodness-of-fit measures of regressions using all subsets of predictors in the data set. (i.e., model (1), (2), ..., (N), (1,2), ..., (1,N), ..., (1,2,3,...,N)). A Z-score based estimate of the 'importance' of each predictor is provided by using a randomisation test.
