Data analysis involves many decisions, including study design, data preparation, and statistical model selection. However, a single analysis represents only one of many possible outcomes, raising questions about the impact of undocumented and at times arbitrary choices. Multiverse analysis addresses this issue by conducting all—or a large set of—meaningful analyses and presenting the results in summary form to assess the robustness of conclusions to alternative modeling decisions. The approach addresses two fundamental problems in research: the lack of transparency and the dependence of analysis results on data-analytic decisions. We will also discuss how to implement the approach, it’s advantages over more traditional analysis approaches, as well as limitations and open challenges, including statistical inference and computational requirements.
Reinhard Schunck is Professor of Sociology at the University of Wuppertal. He works primarily in the field of social stratification and inequality, concentrating on migration and family related processes, and has a focus on quantitative methods.