Abstract

The rapid proliferation of academic journals has intensified concerns about the quality and integrity of published research. While the peer-review system remains a cornerstone of scholarly communication, its effectiveness is inconsistent across outlets. In some cases, inadequate review processes have permitted a substantial volume of methodologically weak or statistically unsound papers to enter the scientific literature. This degradation of quality poses a significant, yet often unmeasured, risk.
This risk is now being amplified by the advent of Large Language Models (LLMs) and AI-powered research tools. These systems are trained on existing corpora of published work. When that corpus is polluted by low-quality studies, AI outputs risk reinforcing errors, biases, and misinformation. This propagation threatens to mislead the public, distort policy discussions, and misdirect young researchers.
From the dual perspective of a statistician and an active researcher, this presentation will: (1) Present empirical case studies illustrating the decline in methodological and statistical rigor in certain segments of the published literature. (2) Propose a critical role for Official Statistics agencies in this ecosystem. We argue that these agencies are uniquely positioned to establish and champion standards for data quality, methodological transparency, and reproducible analysis.
The conclusion will outline a concrete pathway for National Statistical Offices to engage with publishers, academic institutions, and AI developers to uphold statistical integrity in the digital age.