HLER: Human-in-the-Loop Economic Research via Multi-agent Pipelines for Empirical Discovery

Large language models (LLMs) have enabled agent-based systems that attempt to automate scientific research workflows. Existing approaches often pursue fully autonomous discovery, in which AI systems generate research ideas, conduct experiments, and produce manuscripts with minimal human involvement. However, empirical research in economics and the social sciences poses additional challenges: research questions must be grounded in available datasets, identification strategies require careful design, and human judgment remains essential for evaluating economic significance. We introduce HLER (Human-in-the-Loop Economic Research), a multi-agent architecture that supports empirical research automation while preserving critical human oversight. The system orchestrates specialized agents for data auditing, data profiling, hypothesis generation, econometric analysis, manuscript drafting, and automated review. A central design principle is dataset-aware hypothesis generation, in which candidate research questions are constrained by dataset structure, variable availability, and distributional diagnostics. This prevents infeasible or hallucinated hypotheses that frequently arise in unconstrained LLM ideation. HLER further introduces a two-loop research architecture: a question quality loop in which candidate hypotheses are generated, screened for feasibility, and selected by a human researcher; and a research revision loop in which automated review triggers re-analysis and manuscript revision. Human decision gates are embedded at key stages, including question selection and publication approval, allowing researchers to steer the automated pipeline. We evaluate the framework on three empirical datasets, including longitudinal survey data. Across 14 pipeline runs, dataset-aware hypothesis generation produces feasible research questions in 87% of cases, compared with 41% under unconstrained generation, and the system successfully completes end-to-end empirical manuscripts at an average API cost of $0.8-$1.5 per run. A detailed case study on the China Health and Nutrition Survey illustrates the full workflow from data inspection to a revised draft manuscript. These results suggest that Human-AI collaborative pipelines may provide a practical path toward scalable empirical research. The HLER working paper repository is publicly available at https://github.com/maxwell2732/hler-working-papers.