Monthly nighttime-lights panel and station-opening registry for mass transit in Lima and Callao, 2012-2026

Published: 1 September 2026| Version 1 | DOI: 10.17632/8txpvzssf6.1
Contributors:
,
,
,
,

Description

This dataset supports a study of how the opening of metro and bus rapid transit stations affects local economic activity in metropolitan Lima and Callao, proxied by satellite-measured nighttime luminosity, and of the conditions under which that inference is credible. It contains three things. First, a hand-collected registry of 76 mass transit stations with commercial opening months verified against primary sources of the Peruvian transport authorities (ATU, AATE and OSITRAN) rather than press coverage; where a station opened first in free trial operation and only later charged fares, both dates are recorded. Second, a monthly panel of 500 m cells covering the metropolitan area from April 2012 to April 2026, with VIIRS Day/Night Band radiance, cloud-free coverage counts, Sentinel-5P nitrogen dioxide, and urban structure covariates from WorldPop, the Global Human Settlement Layer, Copernicus DEM and Dynamic World. Third, the complete build and estimation code: the Earth Engine extraction notebook, the Stata programs that construct the panel, run the matching and produce the staggered difference-in-differences estimates, the permutation inference routines, and the Python scripts that generate the figures and maps. The panel is provided already built, so the analysis can be reproduced without re-running the satellite extraction, and the raw Earth Engine exports are included for anyone who wants to rebuild it from the composites. The station registry is the component that does not exist elsewhere. Opening dates for Lima's network are scattered across authority documents, operator announcements and press reports that frequently disagree, and the distinction between trial operation and fare collection is rarely made explicit. Assembling and verifying it was a substantial part of the work, and it is reusable well beyond the study it was built for.

Files

Steps to reproduce

Unzip replication_package.zip. Every Stata program begins with a line that sets the global ROOT; point it at the folder where you unzipped the package and no other path needs editing. Run the programs in the order given by their file names: 03_code/02_analisis_principal.do builds the estimation panel, runs the propensity-score matching and produces the main Callaway-Sant'Anna estimates 03_code/03_continuacion_final.do remaining robustness exercises 03_code/04_permutacion_inferencia.do cell-level and 5 km spatial block permutation tests, 500 replications each 03_code/05_heterogeneidad_inferencia.do corridor contrast within cohort, ring gradient, minimum detectable effect 03_code/06_figuras_paper.do tables and Stata figures 03_code/07_figuras_evento.py event-study figures 03_code/08_graficos_mapas.py maps The analysis panel (02_build/panel_analisis.dta) is included, so step 02 onwards runs without touching Google Earth Engine. To rebuild the panel from the satellite composites instead, run 03_code/01_pipeline_datos_colab.ipynb against the raw exports in 01_raw/gee_exports/; this requires an Earth Engine account and takes several hours. Software: Stata 17 or later with csdid, drdid, reghdfe, ftools, kmatch, acreg, eventstudyinteract, did_imputation, bacondecomp and honestdid. Python 3.10 or later with pandas, geopandas, matplotlib and numpy. The permutation routines run 500 replications each and take roughly forty minutes on a desktop machine.

Institutions

Categories

Econometrics, Remote Sensing, Urban Transportation

Licence