Job-specific experience, time into shift, and injury severity in U.S. mining: A 26-year analysis of MSHA data
Description
This repository provides supplementary methods, results, and a reproducibility package for a study of job-specific experience, time into shift, and injury severity in U.S. mining during 2000–2025. The analysis integrates publicly available Mine Safety and Health Administration (MSHA) Open Government Accident Injuries records with historical Part 50 files and annual employment data from MinesProdYearly and ContractorProdYearly. Historical records were linked by DOCUMENT_NO, with agreement on mine identifier and injury date required before importing demographic information. The source snapshot was retrieved on 2 October 2026 and preserved with file-level checksums. The initial dataset contained 271,042 records for 2000–2025. Sequential eligibility restrictions, linkage, complete-case selection, and exclusion of selected disease and inflammation codes yielded a primary sample of 197,159 injury reports from 12,121 mines. Outcomes comprised 62,244 minor injuries (degree 06), 131,716 injuries involving lost or restricted work (degrees 03–05), and 3,199 fatal or permanently disabling injuries (degrees 01–02), including 1,067 deaths and 2,132 permanent disabilities. The main exposures are job-specific experience and time into shift. Time into shift was calculated from reported injury and shift-start times, allowing a midnight crossing, with the primary analysis restricted to 0–16 hours. Covariates include age, recorded sex, total mining experience, calendar year, contractor status, commodity, operational subunit, annual mine employment, and shift-start period. The package contains retained source archives and data definitions, scripts for reconstructing the linked analytical sample, selection and linkage audits, missingness summaries, model coefficients and covariance matrices, numerical tables, figures, and supplementary documentation. Statistical outputs cover sequential multinomial regression, restricted cubic splines, mine-clustered standard errors, nonlinearity and interaction tests, and standardized severity probabilities. Sensitivity analyses address alternative time windows and severity comparisons, zero-time records, accident mechanism, operational strata, contractor clustering, exclusion of 2025, and experience mismatch. These data describe severity conditional on a reported injury. They do not provide exposure-specific denominators for injury incidence or establish causal effects of experience or shift timing. Actual shift duration, individual work histories, and repeated-worker identifiers were unavailable. To reproduce the analysis, extract the package, follow README.txt, install the dependencies listed in requirements_revision.txt, and run scripts/reproduce_revision.py. The retained source snapshot should be used because live MSHA files may subsequently be updated.