Most empirical research involves a long chain of analytical decisions, creating a “garden of forking paths” in which different choices can lead to different findings. Yet published studies typically present only a few carefully-curated results, raising concerns about transparency and credibility. Multiverse analysis addresses this problem by systematically exploring many reasonable analytical alternatives, often across thousands of model specifications. The approach reveals which decisions matter most, and how data and assumptions work together to produce empirical findings.

This talk introduces the method in intuitive terms and illustrates it with real-world datasets. By looking across many plausible analyses rather than a single path, we gain a clearer picture of the uncertainty in our results and strengthen the credibility of reported findings.