Compatibility of content coverage and one-dimensional statistical modelling by item response theory models with local item dependencies and an application in personality psychology

点击翻译按钮获取中文摘要

Psychological constructs, such as intelligence or depression, are typically multidimensional and their measurement requires sufficient content coverage. Nevertheless, in practice one-dimensional measurement models are commonly employed. This is often sensible because a multidimensional statistical approach does not always offer advantages (higher model-complexity, fit not necessarily improved). In addition, a one-dimensional construct is easier to interpret in concrete scientific investigations. Hence, this research project aims to further the development of methods for the construction of one-dimensional scales, with these methods incorporating underlying multidimensionality through local item dependence. The approach builds on a recently published method of Edward Ip that projects multidimensional models into unidimensional models. The method will be initially generalized to the case of polytomous items. Then, a simulation study will compare different robust person parameter estimators. Since the discrimination parameters define the construct in the multidimensional case (in contrast to the one-dimensional case where discrimination merely determines measurement accuracy), different rotations will be implemented and investigated. These rotations will be used to identify discrimination parameters with desirable properties in regard to statistics and content coverage. The newly developed methods are to be applied in personality psychology. Through this project, a psychometric scale for susceptibility will be developed. Susceptibility (also named sensitivity or vulnerability) is multi-dimensional and a suitable scale should inhibit sufficient content coverage. Nevertheless, one-dimensional modelling offers advantages within the context of the investigation of the diathesis-stress-model and the differential susceptibility hypothesis, especially with respect to interpretability and complexity. Because susceptibility is often studied in the framework of gene x environment interactions, we will also include genetic markers (the polymorphisms BDNF Val66Met, DRD2/ANKK1 Taq Ia, 5-HTTLPR, and DRD4 VNTR) for external validation in the development of a suitable scale.