Recursive partitioning as a tool for pharmcogenetic studies of complex diseases: II. Statistical considerations