Item Screening in Graphical Loglinear Rasch Models

摘要: In behavioural sciences, local dependence and DIF are common, purification procedures that eliminate items with these weaknesses often result in short scales poor reliability. Graphical loglinear Rasch models (Kreiner & Christensen, Statistical Methods for Quality of Life Studies, ed. by M. Mesbah, F.C. Cole M.T. Lee, Kluwer Academic, pp. 187–203, 2002) where uniform permitted solve this dilemma modelling the DIF. Identifying a stepwise model search is very time consuming, since initial item analysis may disclose great deal spurious misleading evidence has to disposed during procedure. Like graphical models, possess Markov properties useful statistical if they used methodically. This paper describes how. It contains systematic study way can be distinguish from genuine proposes strategy screening will reduce needed identify fits responses. The last part illustrates procedure on simulated data PF subscale measuring physical functioning SF36 Health Survey inventory.