Formalizing the use of Rasch item fit statistics