Statistical Testing

Large Language Models (LLMs) such as ChatGPT are transforming how scientists conduct and validate research, offering promise as tools to improve scientific reproducibility. However, computational reproducibility and error detection remain expensive and labor-intensive. We experimentally test how collaboration between researchers and LLM assistants influences the reproduction of quantitative social science findings across different levels of AI autonomy. We randomly assigned 288 researchers to 103 teams working under three conditions: human-only, AI-assisted (using ChatGPT as a collaborative tool), or AI-led (ChatGPT operating with minimal human oversight). Teams reproduced published results from leading social science journals, detected coding errors, and proposed robustness checks. Human-only and AI-assisted teams achieved comparable reproduction rates (94% vs. 91%) and performed similarly on most outcomes, except human-only teams identified significantly more major coding errors. Both substantially outperformed AI-led teams, which achieved only a 37% reproduction rate, detected fewer errors across all categories, proposed weaker robustness checks, and required more time. This autonomous approach, however, likely represents only a lower bound of AI capabilities. Despite rapid model advances, expert human judgment currently remains indispensable for reliable empirical verification. While AI assistance did not degrade most outcomes, it provided no measurable advantages and was associated with reduced detection of major errors. However, the 37% autonomous reproduction rate indicates that AI could provide value in settings where scale or cost constraints preclude human review of papers, even though general-purpose LLMs offer no immediate advantages for human-supervised verification.

The same dataset can be analysed in different justifiable ways to answer the same research question, potentially challenging the robustness of empirical science. In this crowd initiative, we investigated the degree to which research findings in the social and behavioural sciences are contingent on analysts’ choices. We examined a stratified random sample of 100 studies published between 2009 and 2018, in which, for one claim per study, at least five reanalysts independently reanalysed the original data. The statistical appropriateness of the reanalyses was assessed in peer evaluations, and the robustness indicators were inspected along a range of research characteristics and study designs. We found that 34% of the independent reanalyses yielded the same result (within a tolerance region of ±0.05 Cohen’s d) as the original report; with a four times broader tolerance region, this indicator increased to 57%. Of the reanalyses conducted, 74% reached the same conclusion as the original investigation, 24% yielded no effects or inconclusive results and 2% reported the opposite effect. This exploratory study indicates that the common single-path analyses in social and behavioural research should not be simply assumed to be robust to alternative analyses. Therefore, we recommend the development and use of practices to explore and communicate this neglected source of uncertainty.