Automated reproducibility assessments in the social and behavioral sciences using large language models

2026-06-11

AI Summary

This study addresses the high cost and limited scalability of traditional manual reanalyses in assessing reproducibility in social and behavioral science research. It introduces, for the first time, an automated reproducibility auditing framework based on large language models (LLMs) that parses original publications, extracts research claims, and automatically generates analysis code to reproduce Cohen’s d effect sizes. The framework enables both quantitative and qualitative comparisons among original findings, human-led reanalyses, and model-generated results. Evaluated on 76 studies, the approach reproduced the original qualitative conclusions in 96% of cases and achieved a 41% replication rate for effect sizes—significantly outperforming human reanalyses, which showed 74% qualitative consistency and 34% effect size replication—thereby substantially enhancing the efficiency and breadth of reproducibility assessment.

Abstract

Reproducibility in the social and behavioral sciences is typically evaluated by independent researchers who reanalyze the original data to assess whether the published findings can be recovered. However, such approaches are resource-intensive and difficult to scale. Here, we show that large language models (LLMs) can automate reproducibility assessments. Using N=76 published studies with predefined claims from the behavioral and social sciences, we compare LLM-generated analysis with the original findings and human reanalysis. For 7 studies, the LLM could not produce a viable effect size estimate. For the remaining studies, our LLM pipeline recovered the original effect sizes in 41% of studies using a +/-0.05 tolerance in Cohen's d. Further, our LLM pipeline reached the same qualitative conclusion as the original study in 96% of cases, where conclusions indicate whether the reanalysis supports the original claim. For comparison, human reanalysts recovered the original effect sizes in 34% of studies and reached the same qualitative conclusion in 74% of cases. Together, these results show that LLMs can serve as a scalable tool for automated reproducibility assessment and provide a foundation for systematic auditing of empirical results in the social and behavioral sciences.

Problem

• reproducibility
• social sciences
• behavioral sciences
• automated assessment
• large language models

Innovation

• large language models
• reproducibility assessment
• automated reanalysis
• social sciences
• effect size recovery