Comments on the rank product method for analyzing replicated experiments