Why Science Is Not Necessarily Self-Correcting

Foundational metascience paper for evaluating psi research credibility. The impediments Ioannidis catalogues — low power, rare replication, publication bias, allegiance bias — are the same ones debated in parapsychology. Both critics (applying these to psi claims) and proponents (applying them to mainstream null results) invoke this analysis.

Self-correction is widely assumed to be a defining hallmark of science, but how often does it actually occur? Reviewing empirical evidence from psychology and biomedicine, Ioannidis argues that self-correction requires active replication effort — yet only ~1% of psychology papers are replications, fewer than 0.2% are independent direct replications, and most yield confirming results (perpetuated fallacies). With average power of 35% and modest bias, unchallenged fallacies may constitute 30–95% of published significant findings. A taxonomy of six discovery–replication paradigms quantifies the problem. Thirteen impediments to self-correction are catalogued, including publication bias, selective reporting, underpowered studies, and editorial bias against replication. Proposed reforms each carry unintended risks unless pursuit of truth remains the overriding priority.