Which night lights data should we use in economics, and where?

Popular DMSP night lights data are flawed by blurring, top-coding, and lack of calibration. Yet newer and better VIIRS data are rarely used in economics. We compare these two data sources for predicting GDP, especially at the second subnational level, for Indonesia, China and South Africa. The DMSP data are a poor proxy for GDP outside of cities. The gap in predictive performance between DMSP data and VIIRS data is especially apparent at lower levels of the spatial hierarchy, such as for counties, and for lower density areas. The city lights-GDP relationship is twice as noisy with DMSP data than with VIIRS data. Spatial inequality is considerably understated with DMSP data, especially for the urban sector and in higher density areas. A Pareto adjustment to correct for top-coding in DMSP data has a modest effect but still understates spatial inequality and misses key features of economic activity in big cities.

Automated summary

Popular DMSP night lights data are flawed and less effective in predicting GDP compared to newer and better VIIRS data, especially at lower levels of spatial hierarchy and in lower density areas. The relationship between city lights and GDP is noisier with DMSP data, while spatial inequality is considerably understated, particularly in urban sectors and higher density areas. Adjustments made to correct for top-coding in DMSP data only have a modest effect and still miss key features of economic activity in big cities.