Abstract : Ranked set procedures assume perfect ranking ability when the data is reduced from n*2 observations to n by choosing order statistics from groups of n each. Some procedures are robust to misrankings and some are not. Hence, as a preliminary procedure, one should check whether the rankings can be said to be perfect or not.

This work (to appear in JASA 2007, jointly with Omer Ozturk and Jesse Frey) proposes some tests to verify the assumption of perfect rankings. Considerations of power and ease of use lead us to one of the procedures out of the several proposed.

These techniques are useful preliminary tests (in the same sense as are Goodness-of-fit tests) whenever rank based procedures are to be used.