A0920
Title: Exploring meaningful analytical choices using the vibration of effects framework
Authors: Simon Lemster - LMU Munich (Germany) [presenting]
Abstract: In many studies in biomedical research, contradictory results are found even when investigating the same research question. Effects often vary considerably across studies, not only due to sampling variability but also due to the multiplicity of possible analysis strategies. These include the exact definition of exposure and outcome, the selection of adjustment variables, or data preprocessing decisions, such as treatment of outliers and missing data. The influence of these analysis decisions can be investigated with the vibration of effects framework. Using the frequently studied example of body composition and cardiovascular mortality, this approach is demonstrated on data from a German population-based cohort study. Findings show that while a positive association between body composition and cardiovascular disease often emerges, the strength and direction of the effect are highly sensitive to analytic decisions. In some model specifications, the association may even reverse. However, not all specifications are equally plausible from a causal or methodological perspective. Some choices may introduce bias, for example, through adjustment for colliders or inappropriate data handling. Therefore, restricting the analysis to a subset of theoretically justifiable decisions may reduce the observed vibration of effects. This could increase both credibility and practical utility of multiverse analyses in applied statistical research.
