Abstracts

Using questionnaires to assess constructs has a long standing tradition in psychological research. Several guidelines and best-practices for constructing questionnaires and scales have been proposed over the years. In most of these, it is recommended to generate more items than the final scale is supposed to include, test this item pool on a sample, and select those items that perform best for the (potentially) final scale. Recent developments have necessitated the use of much shorter scales, making the shortening of established scales a common setting in which items are selected from an original pool. Whether in scale shortening or in initial scale construction, the quality requirements for a valid and reliable scale are manifold and, not seldom, contradicting. Beyond this, modern psychological research is often based on complex study designs, making scales desirable, which are known to be adequate for longitudinal studies, multiple groups, multiple sources of information, or any combination thereof.This thesis presents the stuart approach for item selection, which allows for the simultaneous consideration of a multitude of quality criteria in complex study settings. To this end, item selection is defined as an I-dimensional multiple knapsack problem with assignment restrictions (IMKAR) and an adaptation of the MAX-MIN Ant-System (MMAS) is presented as an algorithmic approach to find solutions for this problem. In this context, item selection is based on generating promising solutions for final scales, evaluating these solutions via confirmatory factor analysis (CFA), and using the results of these analyses to guide the search for better solutions. Within this approach, an ideal measurement model and its restrictions must be defined a priori and solutions are then generated to best accomplish this ideal. Utilizing the CFA approach allows for item selection based on measurement models including multiple facets, multiple occasions, multiple groups, and multiple sources of information simultaneously and optimizing the final solution for criteria of model fit under assumptions of measurement invariance, among others.Because the aim of this thesis is to present an applicable, flexible approach for item selection, an extensive evaluation study was performed to investigate the performance of the chosen algorithmic approach and derive recommendations for applications. These recommendations were then transferred to three applications of item selection: (a) a longitudinal setting, incorporating measurement invariance over time as a crucial component in item selection for a mood scale, (b) a multiple-group setting, aimed at generating a cross-culturally comparable, ultra-short Big Five scale, and (c) a setting including self- and peer-reports in the step of item-selection, to generate a scale which can assess emotional expressivity via multiple sources of information.Overall, the stuart approach proved flexible in the accommodation of a wide variety of study designs, allowing for complex, application-specific