Abstract
This paper discusses a number of methodological issues related to item analysis by Rasch models including the relationships between objectivity as defined by Rasch and the different ways construct validity is defined in psychometrics. To reconcile the different definitions of validity, the IRT models have to be extended to include exogenous variables in addition to items. One such family of extended models referred to as graphical IRT models is described. Graphical Rasch models are special cases of a graphical IRT model characterized by both objectivity and construct validity. Finally, a family of models called graphical loglinear Rasch models is described. Measurement by items from such models is neither valid nor objective according to formal psychometric definitions. The paper argues, however, that measurement is essentially valid and objective and that it is preferable to retain all items to preserve reliability rather than eliminate items to obtain measurements satisfying all the requirements of validity and objectivity.