Several methods have been proposed in the literature to predict item parameters
based on item features. In item cloning, families of clones are derived from a parent
item, by varying those attributes that are assumed not to be related to the item
difficulty (BEJAR, 1993; GLAS; VAN DER LINDEN, 2003). Luecht (2009) proposed the
assessment engineering approach, where items are generated based on construct
maps that describe performance expectations at various levels of the scale. Sheehan
(1997) introduced the application of Classification and Regression Trees (CART)
(BREIMAN et al., 1984), to model the relationship between skills needed to solve
the items and item difficulty. All of these methods have in common that they result in an initial prediction of the item parameters with a certain level of uncertainty.
Incorporating these predicted values in the parameter estimation process, might
offer serious reductions in the costs of item bank development.