On using linear ordered rank statistics for detecting early differences between two distributions