Regional Inequality from Outer Space

Downloads

All data are free to download. Each dataset comes in two identical forms — Stata .dta (with embedded variable and value labels) and plain .csv.

⇩ Download all data (ZIP)stata_codebook.do

Run stata_codebook.do in Stata once to attach long-form per-variable notes to the .dta files.

Load directly in code

Every file loads straight from GitHub (raw URLs — robust and stable) — no manual download needed (except pyreadstat, which reads local files). Swap the file name to load any of the six datasets.

Stata

* Stata 14+ : `use` reads an https URL directly
global BASE "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_kuznets_dmsp/data/"
use "${BASE}Table_3_data.dta", clear
describe        // variable + value labels
notes           // long-form documentation (after running stata_codebook.do)

Python

!pip install -q pyreadstat
# Python : pandas reads a .dta URL directly (values + variable labels)
import pandas as pd
BASE = "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_kuznets_dmsp/data/"
df = pd.read_stata(BASE + "Table_3_data.dta")

# load all six datasets at once
files = ["Prediction_Data", "Table_2_data", "Table_3_data",
         "Table_4_data", "Table_B4_data", "Figure_5_data"]
data = {f: pd.read_stata(BASE + f + ".dta") for f in files}

# pyreadstat exposes the richest metadata but reads LOCAL files -> download first
import pyreadstat, urllib.request
urllib.request.urlretrieve(BASE + "Table_3_data.dta", "Table_3_data.dta")
df, meta = pyreadstat.read_dta("Table_3_data.dta")

Copy and paste this snippet in Google Colab app. https://colab.research.google.com/notebooks/empty.ipynb

R

# R : haven::read_dta auto-downloads an https URL
library(haven)
BASE <- "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/python_kuznets_dmsp/data/"
df <- read_dta(paste0(BASE, "Table_3_data.dta"))   # labels via attr(df$var, "label")

Overview & sources

Companion data for the post Regional Inequality from Outer Space, a Python replication of Lessmann & Seidel (2017): predict regional GDP per capita from DMSP-OLS nighttime lights, build five population-weighted inequality indices, and estimate the spatial (regional) Kuznets curve and its determinants across up to 180 countries, 1992–2012.

Data sources

Cite this data

Please cite both this dataset/replication and the original study.

APA

Mendez, C. (2026). Regional inequality from outer space: Predicting GDP from nighttime lights and building inequality indices in Python [Data set]. https://carlos-mendez.org/post/python_kuznets_dmsp/

Lessmann, C., & Seidel, A. (2017). Regional inequality, convergence, and its determinants — A view from outer space. European Economic Review, 92, 110–132.

BibTeX

@misc{mendez2026kuznetsdmsp,
  author       = {Mendez, Carlos},
  title        = {Regional Inequality from Outer Space: Predicting GDP from Nighttime Lights and Building Inequality Indices in Python},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/python_kuznets_dmsp/}},
  note         = {Data set; replication of Lessmann and Seidel (2017)}
}

@article{lessmann2017regional,
  author  = {Lessmann, Christian and Seidel, Andr\'{e}},
  title   = {Regional inequality, convergence, and its determinants---A view from outer space},
  journal = {European Economic Review},
  volume  = {92},
  pages   = {110--132},
  year    = {2017}
}

Variable explorer search & filter all 53 variables

Type to filter by name or label, or use the chips to filter by type. Each row shows a mini distribution of the variable. Click a column header to sort.

Cross-file variable index

Which file each variable appears in (● = present).

Construction & formulas

All five inequality indices are computed within each country-year, across that country's regions, on predicted regional income y = pred_GDP_pc_Region, weighted by the regional population share p_i = Pop_Region_i / Pop_Country. Let ybar = sum_i p_i * y_i be the population-weighted mean.

• Population-weighted Gini (GINIW_pred_GDP_pc, 0–1): GINIW = ( sum_i sum_j p_i p_j |y_i - y_j| ) / (2 * ybar).
• Coefficient of variation (COVW_pred_GDP_pc): COVW = sqrt( sum_i p_i (y_i - ybar)^2 ) / ybar.
• Generalized entropy GE(alpha) (GE_1W, GE_0W, GE_m1W): for alpha not in {0,1}, GE(alpha) = 1/(alpha(alpha-1)) * sum_i p_i [ (y_i/ybar)^alpha - 1 ]; GE(1) = Theil = sum_i p_i (y_i/ybar) ln(y_i/ybar); GE(0) = mean log deviation = sum_i p_i ln(ybar/y_i); GE_m1W uses alpha = -1.

Other constructed variables:

• Log nighttime light (log_Light_ppix_Region): ln(meanDN), with the region mean DN set to 0.01 when it is 0 so the log is defined (DMSP DN range 0–63).
• Prediction model (eq. 1): a random-effects regression of log_GDP_pc_Region on log_Light_ppix_Region, the top-/low-coded pixel counts, log_area, log_region, log_region_X_log_area, the World Bank region dummies, the satellite-era dummies, and log_GDP_pc_Country; pred_GDP_pc_Region is the back-transformed fitted value.
• Transport cost (paper's proxy, not stored): area * price_gasoline.

The six datasets

Switch datasets with the tabs. Each shows the full variable dictionary plus a sortable statistics table with mini distributions and data coverage.

Expand all datasets expand to search (Ctrl/⌘+F) or print across all six datasets

Known limitations & caveats

• DMSP top-coding / saturation. Digital numbers cap at 63, so the brightest city cores are censored; log_N_pix_top_cod_1_ppix controls for this.
• Sensor drift. Six+ DMSP sensors span 1992–2010 with differing calibration and aging; the satyear_1…satyear_7 dummies absorb era effects.
• Sparse determinants. Several Table 4 determinants are observed for far fewer country-years than the core panel (e.g. price_gasoline N=1,366, School_enrollment_secondary N=2,566), so those regressions run on shifting subsamples — compare descriptively. The licensed ICRG "bureaucratic quality" index is not included.
• Scale of Giniall. The interpersonal income Gini is on a 0–100 scale, whereas the regional indices are on a 0–1 scale — do not mix them without rescaling.
• code_Coutry_Region spelling. The misspelling "Coutry" is kept so the key matches the authors' original replication archive.