Subgroup Mining Signals P‑Hacking

Researchers commonly split samples and search for subgroups until an outcome reaches statistical significance; because interaction effects require much larger samples than main effects, these subgroup discoveries are especially likely to be flukes and fail replication. Identifying fields or papers with unusually many subgroup‑only significant results offers a scalable signal of p‑hacking and compromised evidence. — Flagging subgroup‑only findings would help journalists, policymakers, and funders distinguish robust results from likely data‑dredged artifacts and shape norms (preregistration, reporting) to reduce false positive science.