Embedded Files
As economics scales with AI, a key bottleneck is evidence synthesis in a comparable, aggregable form
As economics scales with AI, a key bottleneck is evidence synthesis in a comparable, aggregable form
We created an evidence‑annotated claim graphs for 44,852 economics working papers (1980–2023).
We created an evidence‑annotated claim graphs for 44,852 economics working papers (1980–2023).
WHAT IS A “CLAIM GRAPH”?
WHAT IS A “CLAIM GRAPH”?
A claim graph is a paper‑level, machine‑readable map of what a paper states.
Nodes are standardized economic concepts (mapped to JEL codes). Directed edges are the relationships the authors claim between concepts.
Each edge is annotated with its evidentiary basis, including whether it is supported by credible causal designs (e.g., DiD, IV, RDD, RCTs, event studies, synthetic control) or by non‑causal evidence.
WHAT THIS PROJECT COVERS
WHAT THIS PROJECT COVERS
Corpus: 44,852 NBER and CEPR working papers (due to full text availability)
Years: 1980–2023
Unit of analysis: relationships (edges), not just paper‑level tags
Output: evidence‑annotated claim graphs + paper‑level measures (complexity, novelty, positioning)
HEADLINE RESULTS
HEADLINE RESULTS
1) The share of causal edges rises sharply over time (7.7% in 1990 → 31.7% in 2020).
2) Causal narrative complexity and causal novelty predict top‑5 publication and long‑run citations.
3) Non‑causal complexity and novelty are weakly related or negative.
Graph of Landmark Economics Papers
Graph of Landmark Economics Papers
Esther Duflo; Abhijit Banerjee; Rachel Glennerster; Cynthia G. Kinnan
Esther Duflo; Abhijit Banerjee; Rachel Glennerster; Cynthia G. Kinnan
Raj Chetty, Nathaniel Hendren, Patrick Kline, Emmanuel Saez
Raj Chetty, Nathaniel Hendren, Patrick Kline, Emmanuel Saez
Blue edges are based on causal inference methods, Orange on non-causal methods.
Proportion of Claims in Paper that are Causal
Proportion of Claims in Paper that are Causal
Across economics working papers, the share of causal edges rises strongly over time: 7.7% (1990) → 31.7% (2020), continuing upward through 2023.
By field, before and after 2000
By field, before and after 2000
Breakdown by Methods
Breakdown by Methods
Breakdown by Fields
Breakdown by Fields
WHAT PREDICTS PUBLICATION AND CITATIONS?
WHAT PREDICTS PUBLICATION AND CITATIONS?
Causal narrative structure is rewarded:
Causal narrative structure is rewarded:
Papers with richer causal structure (more causal edges / deeper causal chains) are more likely to publish in top‑5 journals.
Novel causal relationships (new causal edges or paths) are strongly associated with top‑5 publication.
By contrast, expanding non‑causal structure is weakly related or negative.
We find a similar pattern for citations:
We find a similar pattern for citations:
Causal complexity and causal novelty robustly predict higher citations.
Conceptual positioning also matters: papers anchored in central, widely recognized concepts tend to receive more citations.
IMPORTANT NOTE
IMPORTANT NOTE
We extract what papers claim and how they support those claims. We do not adjudicate whether claims are true.
Last updated: <February 2026>.
Last updated: <February 2026>.
This is v2.0 of our project. For our v1.0 data please contact us.
Questions?
Questions?
Contact [team@causal.claims] to get more information on the project.
We would love to hear from you!
Page updated
Google Sites
Report abuse