Data and analysis code for: Separating detectability, robustness, and predictive utility: atmospheric pressure and Japanese psychiatric day-care attendance over four years

This record contains the anonymized dataset and the analysis code that reproduce all results reported in the associated manuscript: "Separating detectability, robustness, and predictive utility: atmospheric pressure and Japanese psychiatric day-care attendance over four years" (Okuma T, Ishige M, Kikuchi S).

The study is a retrospective, facility-level observational analysis of daily attendance at a Japanese psychiatric day-care unit over four fiscal years (FY2022–FY2025), examining same-day atmospheric pressure as a candidate predictor and assessing its statistical detectability, robustness to meteorological confounding, and out-of-sample predictive utility.

Files:

• data_all_years.csv — daily records (one row per operating day, Monday–Saturday): attendance count, atmospheric pressure, temperature, precipitation, and relative humidity at lags 0–3, and fiscal year.
• DATA_README.md — data dictionary (column definitions, units, and sources).
• analysis_reproduce_all.py — script reproducing all tables (Tables 1–4 and S1–S3 Tables), diagnostics, and figures.
• README.md — code description and usage.
• requirements.txt — pinned software versions.

Meteorological data were obtained from the Japan Meteorological Agency (Past Weather Data Download service): atmospheric pressure and relative humidity from the Chiba Local Meteorological Observatory, and temperature and precipitation from the Kisarazu observation station (Chiba Prefecture, Japan). Attendance data are anonymized, facility-level daily counts containing no individually identifiable information.

Usage: pip install -r requirements.txt then python analysis_reproduce_all.py data_all_years.csv.