Causal Claims in Economics

Researchers Prashant Garg and Thiemo Fetzer have published a groundbreaking paper titled 'Causal Claims in Economics,' using a structured multi-stage AI workflow to analyze the entire corpus of modern economic literature. Their system processes 44,852 papers spanning 1980-2023, converting each into standardized 'evidence-annotated claim graphs'—directed networks where nodes represent economic concepts and edges represent stated relationships, with each edge labeled by its evidentiary basis (causal vs. non-causal). This represents a major step toward solving economics' scaling bottleneck: representing paper claims in comparable, aggregable form. The complete data, code, prompts, and workflow documentation are publicly available on GitHub, enabling replication and further research.

The analysis reveals a dramatic shift in economic research methodology over three decades. The share of edges in claim graphs supported by causal inference designs (like randomized controlled trials or natural experiments) surged from just 7.7% in 1990 to 31.7% in 2020—a fourfold increase highlighting the 'credibility revolution' in empirical economics. Crucially, the researchers found that measures of causal narrative structure and causal novelty are positively associated with publication in top-five economics journals and long-run citations, whereas non-causal counterparts show weak or negative relationships. This suggests the field increasingly rewards rigorous causal identification. The open-source AI pipeline demonstrates how large language models (LLMs) can be systematically applied to meta-science, potentially transforming how we synthesize knowledge across disciplines.