Economists and other social science researchers increasingly use satellite-detected night-time lights, as one of the most popular “big data” sources. The most widely used series of night-time lights data are from the Defense Meteorological Satellite Program (DMSP), which was initiated in the 1960s to observe clouds to aid US Air Force weather forecasts. Initial use of these data by social science researchers was as a proxy for economic activity at the national or aggregated regional level but increasingly these data are used to evaluate local impacts of interventions and to estimate local inequality. When measurement errors in these data were originally considered it was in a framework that just required that the errors were independent of errors in conventional economic statistics. However, more recent studies use DMSP data directly as a proxy and so the nature of their measurement error becomes important because under certain circumstances these errors could cause bias that distorts conclusions.

This talk provides two such examples: first, when estimating local inequality in China and the United States the level of inequality is understated and a misleading trend is introduced, because of spatially mean-reverting errors in the DMSP data. Second, in a difference-in-differences evaluation of the impact of a sanction on North Korea the sanction impact is understated due to mean-reverting errors and bottom-coding in the DMSP data. These errors reflect some of the inherent limitations of DMSP data. Where possible, applied economists and other social scientists should switch to using newer, more accurate, night-time lights data that were designed for research purposes, even if that means they have to work with shorter time-series.

John Gibson is a Professor of Economics at the University of Waikato, New Zealand. His research interests include Economic Theory and Applied Economics, especially Accounting, Finance and Economics Operations.