Modeling typical performance measures

In the educational, employment, and clinical context, attitude and personality inventories are used to measure typical performance traits. Statistical models are applied to obtain latent trait estimates. Often the same statistical models as the models used in maximum performance measurement are applied to typical performance measures. However, different models might be better applicable to describe these typical performance measures. In this dissertation the modeling of two systematic features in the typical performance domain is discussed; 1. the factor structure of typical performance measures, and 2.response processes to typical performance measures. In the first part of the dissertation complex multidimensional models (e.g. bifactor model, non-hierarchical multidimensional model, second-order model) are investigated to describe the factor structure of both a personality inventory and an attitude inventory. In the second part of this dissertation the applicability of different response models, dominance IRT models and unfolding IRT models, to describe the response processes on two personality inventories is compared. In the next chapters an already existing unfolding IRT model, the generalized graded unfolding model (GGUM) is compared to three newly developed alternatives, the collapsed generalized partial credit model (CGPCM), the collapsed graded response model (CGRM) and the quadratic logistic regression model (QLOG) and the statistical fit of the models is investigated. Both person fit (constancy of theta statistic and tendency to agree statistic) and item fit (Differential item functioning and shape of item characteristic curve statistic) are investigated by fit statistics that are developed based on the Lagrange Multiplier (LM) test. In general, it was found that it is important to not simply apply models that are used in maximum performance measurement to typical performance measurement. To investigate the dimensionality structure it is important to take both general and specific factors into account. Furthermore it is reasonable to expect that responses to typical performance measures follow an ideal-point response process. Four unfolding IRT models and methods to assess the statistical fit of the models were introduced.