Replication data and code for “From Fragmented Public Data to Health-Effect Diagnostics: An AI-Agent and Machine-Learning Framework for China’s Sponge City Programme”

Description

This dataset contains the structured data, derived analysis outputs, source register, and reproducibility materials for the manuscript “From Fragmented Public Data to Health-Effect Diagnostics”. It corresponds to the frozen v2d analysis version. The package includes the authoritative city-year panel, adjudicated sponge-city onset registry, public-source health extractions, imputation flags and confidence grades, climate and latitude classifications, DID/event-study/robustness/synthetic-control outputs, the recomputed Table 10 files, and an auditable Table 5 analysis script. Original PDFs and other third-party publications are not redistributed; the source register provides URLs or source descriptions for traceability. Empty cells indicate unavailable values, and imputation status is retained in the panel.

Steps to reproduce

Use the v2d-labelled master panel in 01_master_panel as the authoritative input. Treated-city cohorts use onset_p50, falling back to policy_start_year; untreated cities use cohort 0. For the main staggered-treatment analysis, use the Callaway-Sant’Anna group-time ATT with not-yet-treated controls, doubly robust estimation, a varying base period, no anticipation, simple post-treatment aggregation, city clustering, and an unbalanced panel. Run 14_method_docs/run_table5_cs_v1.py with Python and the listed packages. For Table 10, use the v2d panel and the accompanying recomputation CSV and JSON metadata; the interaction model includes standardized latitude, its interaction with treatment, city and year fixed effects, and city-clustered standard errors.

Institutions

• Affiliated Hospital of Qingdao University

  Shandong, Qingdao

Categories

Funders

• The Affiliated Hospital of Qingdao University

  Grant ID: Horizontal research project

Licence