If non-normal (skewed, kurtotic) data can be used in the analysis, how is PLS-SEM a valid 'regression' technique?