Half of social science papers don’t replicate. Now what.

HALF OF SOCIAL-SCIENCE STUDIES FAIL TO REPLICATE. That is not a bug, it is the current baseline. In April 2026, the Systematizing Confidence in Open Research and Evidence (SCORE) project released results from 865 replication attempts across economics, psychology, and political science. Only 54% of original findings held up under rigorous retesting. The data, published in Nature, confirms what many suspected but few could quantify: published results are not reliable by default. Stanford’s 2026 Data Science CORES Symposium doubled down on solutions, with Russell Poldrack and others advocating mandatory pre-registration, shared code, and standardized effect-size reporting. The American Enterprise Institute simultaneously pushed for scaled replication in education research, where evidence drives billion-euro policy decisions. This matters deeply in the Nordics. Our public institutions, Finnish education boards, Swedish health agencies, Danish innovation funds, routinely cite social science to justify spending, regulation, and product design. If half the foundation is sand, interventions fail quietly and budgets bleed. Worse, policymakers often lack the statistical literacy to distinguish robust from fragile findings. The replication crisis is not academic theater; it is a silent tax on public trust and product efficacy. Start this week: audit one paper you plan to cite or implement. Use the SCORE Replication Index (publicly available via Phys.org) to check its replication status. If no replication exists, demand the raw data and analysis script before acting. If the authors refuse or cannot provide them, treat the finding as provisional, not actionable.

scandinavi.ai

researched · 6 sources

11 JulResearchreaches nearby