Implementation of the Ant Colony Optimization Algorithm for the development of short-scales for determinants of health behavior

点击翻译按钮获取中文摘要

BACKGROUND: The usage of e- and m-health interventions in population-based behavioral prevention and epidemiological research facilitates the use of multi-behavioral tests. The applicability of these approaches in a population as short intervention depends on the burden of the assessment and therefore limits its effect at the population level. This highlights the increased demand for psychometrically robust short scales. The construction of short-scales presents psychometric challenges to researchers and conventional methods of item selection, such as confirmatory factor analysis (CFA) and the use of item response theory (IRT) for scaling, can not address these problems adequately. Automatic metaheuristic optimization algorithms could be used instead as time efficient methods to take these problems into consideration and select itemsets for psychometrically solid short-scales.GOALS: In the course of this project, I want to use the ant colony optimization (ACO) algorithm to develop valid and reliable short-scales for the assessment of self-efficacy and decisional balance in regard to health-related behaviors. Therefore, the following issues will be addressed: 1) To what extent is the ACO algorithm an adequate method of item selection in the domains of alcohol and tobacco consumption, as well as physical activity? 2) Are short-scales that were optimized with the ACO algorithm comparable to or even more reliable than short-scales constructed with conventional methods? 3) Are the scales invariant across different points of time and can the ACO algorithm be used to select measurement invariant itemsets?METHOD: Data will be used from 5 projects of the research collaboration “Early interventions in health risk behaviors” (EARLINT) with up to n = 12.372 subjects from the domains of alcohol consumption, tobacco consumption and physical activity. The ACO algorithm will be used as an automatic and time efficient optimization method. The short-scales which are optimized by this algorithm will be compared to scales that were developed by CFA and IRT scaling. Additionally, longitudinal data will be used to establish measurement invariance across different points of time by using multiple group CFA, thus examining whether the scales are comparable across different points of time.EXPECTED BENEFIT: Considering the limited time resources in clinical practice for surveys and the concurrent frequent use of multi-behavioral tests, it is clear that psychometrically reliable short-scales are required. The ACO algorithm can be used to develop reliable short-scales for these assessments. The project aims to provide the following: 1) data analysis scripts for the implementation of the ACO algorithm for further research, 2) comparison of the newly constructed short-scales with currently applied scales, 3) use of longitudinal data for the establishment of measurement invariance in the short-scales across different points of time.