Railway Supply and Car Dependence in French Municipalities

I developed a reproducible open-data analysis of the relationship between railway service intensity and car dependence in French municipalities.

The study examines whether, among municipalities actually served by rail, a higher level of railway service is associated with a lower share of commuting by car.

The project combines several open datasets published on data.gouv.fr, including modal share, population density, median living standard, territorial classifications and railway supply data from Géofer.

Among the 2,144 municipalities included in the main analysis, a higher railway service intensity is associated with a lower car modal share after controlling for population density, population size, median living standard and position within the French urban-area classification.

In the main specification, doubling railway service is associated with approximately 1.24 percentage points less car modal share.

The result remains statistically significant when standard errors are clustered by department.

The analysis is observational and does not establish a causal effect.

The study is publicly available here:

• Reuse on data.gouv.fr
• Source code and reproducibility repository
• Scientific paper (PDF)

The analysis combines:

• commuting modal-share data;
• population density and population;
• median living standard;
• the French Aires d'attraction des villes classification;
• railway service intensity from the Cerema Géofer dataset.

The initial exploration and identification of relevant datasets was performed using my datagouv-toolkit project.

Railway supply is measured using the number of TER, TGV and Intercités stops observed on a typical weekday in 2022.

Several model specifications are compared, including robust HC3 standard errors, department-clustered standard errors and an analysis by railway-service quartiles.

The repository contains:

• Python scripts for data preparation;
• territorial-control construction;
• statistical models;
• publication figures;
• automated tests;
• the Org manuscript;
• the generated PDF.

Code quality is checked with:

• Ruff for linting and formatting;
• mypy for static type checking;
• pytest for automated testing;
• Bandit for security analysis.

This project turns several independent French open datasets into a reproducible empirical study of structural car dependence.

The main result is a robust statistical association between stronger railway service and lower car modal share, while explicitly avoiding a causal interpretation unsupported by the current cross-sectional design.