Pilot flight experience and accident severity in Brazilian general aviation: a register-based analysis of CENIPA final reports (2007– 2024)

Description

This dataset contains aggregate research outputs and analysis code supporting the study “Pilot flight experience and accident severity in Brazilian general aviation: a register-based analysis of CENIPA final reports (2007–2024)”, by Daniel Guilherme Marques da Silva and Fabiana Serra de Arruda. The study examines the association between pilot-in-command flight experience and the probability of at least one onboard fatality, conditional on an accident having occurred. It does not estimate accident incidence or exposure-adjusted accident risk. CENIPA final investigation reports were linked to structured occurrence and aircraft records. The revised analytical sample comprises 800 accident-aircraft records from 800 unique occurrences, including 199 records with onboard fatalities. General aviation is used in a broad operational sense; narrower operational definitions are examined in sensitivity analyses. The package includes regression coefficients, confidence intervals, sensitivity analyses, multiple-imputation summaries, missingness statistics, annual report-linkage and inclusion counts, calibration summaries, and model-derived probability curves. It also provides statistical code, variable and output dictionaries, descriptive metadata, file checksums, and instructions for regenerating figures. The primary complete-case model includes 764 records and 192 fatal outcomes. The adjusted odds ratio per approximate doubling of total flight hours is 0.93 (95% CI 0.85–1.01; p=0.100). These observational results do not establish a causal or statistically conclusive inverse association. To minimise indirect-identification and redistribution concerns, the package excludes record-level analytical data, aircraft registrations, occurrence identifiers, individual predictions, report narratives, original CENIPA PDFs, and third-party reference PDFs. The deposited summaries support inspection of the estimates and regeneration of six figures. Independent model refitting and reconstruction of the analytical cohort require additional record-level inputs that are not included. Packaging and automated checks do not replace human validation of source extraction and linkage.

Steps to reproduce

1. Download and extract the ZIP archive. Open a terminal in the extracted folder containing README.md, requirements.txt, data/, and src/. 2. Use Python 3.12 and create a virtual environment: python -m venv .venv Activate it on Linux/macOS: source .venv/bin/activate Or on Windows PowerShell: .venv\Scripts\Activate.ps1 3. Install the pinned dependencies: python -m pip install -r requirements.txt 4. Validate the deposited files and regenerate the figures: python src/reproduce_public.py The script checks SHA-256 file integrity, coefficient-to-odds-ratio transformations, annual inclusion totals, and confidence-band ordering. It regenerates six figures covering experience bands, operation-specific median hours, spline predictions, adjusted odds ratios, separate total/in-type experience curves, and apparent calibration. 5. Inspect the generated PNG files and checks.json in reproduced/. The original deposited tables are not overwritten. Consult OUTPUT_DICTIONARY.csv for table definitions and results/model_results.json for model estimates, diagnostics, and software versions. 6. Distinguish figure regeneration from model refitting. The command above uses aggregate and model-derived outputs; it does not independently re-estimate the statistical models. For model refitting, a legitimately obtained and validated record-level analytical CSV is required. After placing it at results/tables/analysis_dataset.csv, run: python src/analysis/revised_analysis.py --dataset results/tables/analysis_dataset.csv --out refitted Required input variables are described in VARIABLE_DICTIONARY.csv. The exact-date-linkage sensitivity additionally requires results/tables/sample_pdf_ledger.csv and is omitted when that file is absent. Neither record-level input file is included in this deposit. Full archive reconstruction also requires original registry files, reports, extraction outputs, and validated linkage corrections. See REPRODUCIBILITY.md for these limitations and ETHICS_AND_PROVENANCE.md for source provenance and AI-assisted processing.

Institutions

• Universidade de Brasília

  Federal District, Brasília

Categories

Licence