The documentation for this data set can be found in the following open access peer-reviewed publication:
Galimberti, J. K., S. Pichler, and R. Pleninger. 2023. Measuring inequality using geospatial data. The World Bank Economic Review, 37(4):549-569, https://doi.org/10.1093/wber/lhad026.
Notes on the data and their interpretation:
The Global Database of Light-based Geospatial Income Inequality (LGII) Measures, Version 1 data set contains Gini-coefficients of inequality for 234 countries and territories from 1992 to 2013. The measurement unit is the Gini-Coefficient (Range: 0-1), with higher values representing higher inequality. These measures are constructed using worldwide geospatial satellite data on nighttime lights emission as a proxy for economic prosperity, matched with varying sources of data on geo-located population counts. The nighttime lights data were supplied by the National Oceanic and Atmospheric Administration (NOAA), National Centers for Environmental Information (NCEI), Earth Observation Group (EOG), and Operational Linescan System (OLS) instruments. The population data used consisted of CIESIN's Gridded Population of the World (GPW) collection, and the Oak Ridge National Laboratory (ORNL) LandScan (LSC) data set. The nighttime lights and population data were combined to produce an array of geospatially-informed Gini-coefficients, which were then weighted to optimize their correlation with a benchmark - specifically, the Standardized World Income Inequality Database (SWIID), to generate a parsimonious composite inequality metric.
These data are generated using Operational Linescan System (OLS) instruments and specifically employ an unfiltered version (referred to as 'average visible') that captures even low-intensity emissions from less densely populated regions. A constant elasticity model, uniformly implemented at the pixel level, is employed to convert luminosity into economic output indicators. The elasticity parameter is empirically determined through an agnostic calibration approach, guided by extant measures and quality assessments of income inequality. Two primary sources furnish the population data integral to the calculations. The first is the GPWv4.11 data set, assembled using population census figures gathered from a multitude of organizations, including national statistical agencies, and is conjoined with spatially explicit administrative boundary data sets. The second is the LSC data set, which employs a multi-variable disaggregation technique to apportion census counts within administrative regions, informed by auxiliary data including land cover, transportation networks, topography, and high-resolution satellite imagery. The aggregation of these luminosity and population data sets produces an array of geospatially-informed Gini-coefficients. To generate a parsimonious composite inequality metric, the Gini values are subsequently weighted to optimize their correlation with a benchmark - specifically, the Standardized World Income Inequality Database (SWIID). The weighting factor, denoted as "lambda", is adjustable, enabling customization to achieve superior cross-country (coded as L100 in the data set) or within-country (coded as L000) congruence with income inequality metrics. Intermediate measures are also provided, with a recommended lambda calibration of 0.5 (coded as L050 in the data set).
The magnitudes of the Light-based Geospatial Income Inequality (LGII) Gini-coefficients are generally higher than Gini-coefficients based on income surveys – on average, about 0.3 points higher than the benchmark SWIID. This difference can be attributed to spatial aggregation and the use of night lights as a proxy for income.
The LGII Gini-coefficients are recommended for interpretation comparatively across countries. One of the merits of this data set is that it has light-based Gini-coefficients that are more highly correlated with income-based Gini-coefficients than the previous literature, especially in cross-country comparisons. It should be noted that even inequality measures based on income surveys are subject to uncertainties, mostly due to income mismeasurement. The calibration of the LGII Gini-coefficients takes into account a measure of uncertainty around the income inequality benchmark.
The time series presented in these data should be treated cautiously. The LGII measure does not capture time variation in the benchmark income inequality as well as it captures variations across countries. Inequality is a slowly moving phenomenon, so the time variation observed over the 1992-2013 period is relatively small compared to the variation across countries. Trends in inequality are normally studied across several decades of data, sometimes centuries, which is feasible only for a handful of countries. Additionally, the data sources used in this data set are not ideal for comparisons over time; both night lights and gridded population (especially LandScan) have issues here.
Finally, estimates for small geographies, such as small island states, need to be treated with caution. This is because the number of observations available for computation of the Gini-coefficients is small, especially in terms of census areas.
Recommended data set citation:
Galimberti, J. K.1,2,3, S. Pichler2,4,5,6, and R. Pleninger2,7. 2023. Global Database of Light-based Geospatial Income Inequality (LGII) Measures, Version 1. Palisades, New York: NASA Socioeconomic Data and Applications Center (SEDAC). https://doi.org/10.7927/kd8b-2376. Accessed DAY MONTH YEAR.
1 Asian Development Bank, Economic Research and Development Impact Department, Philippines
2 KOF Swiss Economic Institute, ETH Zurich, Switzerland
3 Centre for Applied Macroeconomic Analysis, Australian National University, Australia
4 University of Groningen, Netherlands
5 Aletta Jacobs School of Public Health, Netherlands
6 IZA Bonn, Germany
7 World Bank, USA
The authors wish to thank Professor Eric Edmonds, Editor at World Bank Economic Review, and anonymous referees; Richard Bluhm, Bruno Caprettini, Florian Eckert, Vera Eichenauer, Harry Garretsen, Martin Karlsson, Melanie Krause, Philip Vermeulen, and Nicolas Ziebarth for helpful comments and suggestions; participants at the World Inequality Conference in Paris in 2021, DENS 2020 in St. Gallen, YSEM 2021 in Zurich and the research seminars at the Competent in Competition and Health center (CINCH) in Essen, at Auckland University of Technology, at the University of Auckland, at the University of Hamburg, and at the University of Groningen; Kytt MacManus and Greg Yetman of the Center for International Earth Science Information Network (CIESIN) for their hospitality; and the NASA Socioeconomic Data and Applications Center (SEDAC) for their support with the distribution of the data set generated by this project. The subnational borders data used in this work were accessed during a visit to CIESIN. This project received funds from the Management, Technology and Economics (MTEC) Foundation Grant, for which the authors gratefully acknowledge their support. The views expressed in this paper are those of the authors, and do not necessarily represent the views of their corresponding institutional affiliations.