This article confronts the ongoing statistical crisis undermining scientific credibility—highlighting p-hacking, publication bias, and the misapplication of statistical methods as central culprits. Drawing on the foundational critique by John Ioannidis that "most published research findings are false," the article delves into the cultural and methodological roots of the replication crisis and calls for a comprehensive reform of scientific practice.
A central theme is the rejection of traditional point-null hypothesis testing, which under continuous models assigns zero probability to any specific value. The authors argue for replacing this outdated framework with interval null hypotheses that emphasize practical significance and empirical relevance. This proposed shift, informed by the Rao–Lovric Theorem, also offers a resolution to the Jeffreys–Lindley Paradox and seeks to unify frequentist and Bayesian methodologies.
The paper calls for institutional and systemic change. It promotes transparency through preregistration, open data, and detailed reporting; reproducibility through replication studies; and technological progress through advanced software, machine learning, and dynamic reporting tools. Education reforms are also emphasized, encouraging the integration of statistical literacy and best practices in curricula across scientific disciplines.
Journals and funding agencies are urged to shift incentives toward methodological rigor over novelty, fostering a culture where reproducibility and transparency are core scientific values. The paper also advocates for the notion of contextual significance: interpreting p-values not as binary decision tools but as indicators of context-specific relevance.
Ultimately, this work envisions a new statistical paradigm: one that is transparent, reproducible, collaborative, and grounded in real-world meaning. For a complete exploration of reform strategies and theoretical insights, consult the full encyclopedia article in the International Encyclopedia of Statistical Science.