On Measuring Some of the People Some of the Time with Some of the Items: The Search for Stability and Variation in Item Sets

Psychological assessment is shaped by the items used and the persons assessed. Both items and persons typically represent a random or representative sample of a much larger item and person pool. However, most of the focus on psychological measurement rests on the person sampling side. Item sampling from larger item pools is still a black box. In this dissertation, I present the advantages of new state-of-the-art item and person sampling procedures in the context of personality development research (manuscript 1). Measurement in personality development faces many psychometric problems. First, the theoretically assumed measurement models do not fit the data when tested with confirmatory factor analysis. Second, measurement invariance across age, which is necessary for a meaningful interpretation of age-associated personality differences, is rarely accomplished. And third, the continuous moderator variable age is often artificially categorized. I show how Ant Colony Optimization can be used to select indicators that provide adequate model fit and measurement invariance across age (manuscript 2). I also apply a combination of the item sampling approach Genetic Algorithm and the person sampling approach Local Structural Equation Modeling to identify the items that provide the most prototypical measurement of personality within restricted age samples (manuscript 3). These manuscripts address two sides of the measurement invariance problem (i.e., the DIF paradox): If normative age-associated differences should be studied, measurement invariant indicators across age need to be selected. If the measurement within restricted age ranges should be optimized, indicators that maximize model fit and measurement variance across age need to be sampled. The novel item sampling procedures can be applied in any assessment context to optimize psychometric requirements (e.g., model fit, reliability, difficulty). The person sampling method Local Structural Equation Modeling can also be applied to any measurement to study the robustness across continuous moderator variables (e.g., cognitive abilities, SES). In the epilogue, I discuss implications for personality measurement and provide an outlook on future research.