Skip to main content

Night lights and regional GDP

Abstract

Based on evidence from national data, Henderson et al. (Am Econ Rev 102(2):994–1028, 2012) suggest that growth of night lights can proxy reliably for growth of regional GDP in low-income countries where GDP data is frequently lacking or of poor quality. Using regional data in two large emerging economies, Brazil and India, this paper finds, however, that the relationship between night lights growth and observed GDP growth varies significantly—in both statistical and economic terms—across regions. The same applies to advanced economies like the United States and Western Europe. The paper accounts for measurement issues with regard to the night lights data and considers several extensions of the empirical model in order to analyze if and under which circumstances the relationship between night lights and GDP growth is stable. Yet parameter instability typically persists, while the stable relationship among urban counties in Brazil represents the major exception.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2

Notes

  1. Examples of these studies are Alesina et al. (2015) who show that ethnic inequality within countries (Gini index of average per capita lights intensities across the homelands of ethnic groups) hinders aggregate economic development (country-level GDP); Gennaioli et al. (2013) who find that night lights are related to human capital in a similar way as regional per capita income for a large cross-country sample of regions; Michalopoulos and Papaioannou (2013) who find a positive association between more complex pre-colonial institutions and current night lights intensity within African countries; Hodler and Raschky (2014) who find that leaders of countries with poor institutions use foreign aid for favoritism, indicated by higher effects of foreign aid on per capita lights intensities at the leaders’ birthplaces; or Small et al. (2011) who find that Zipf’s law holds for night lights all over the world.

  2. In a similar vein, Chen and Nordhaus (2011) argue that changes of night lights have informational value for countries with poor quality of national income accounts.

  3. While we call these local units counties for expositional convenience here, we will use the smallest administrative units for which GDP data is available in the empirical implementations: districts in India, municipalities in Brazil, counties in the United States and NUTS3 regions in Western Europe.

  4. Like Henderson et al., we average the initial and final GDP and lights densities of these growth rates over two years to mitigate the effects of outliers. The GDP growth rate, for example, is calculated as y i  = [ln(YD iT  + YD iT–1) – ln(YD it+1+YD it )]/(T – t – 1), where YD denotes GDP density (per km²) and T and t are, respectively, the last and the first year for which we have data for region i. Unlike Henderson et al., we use compound growth rates because time periods for which data is available differ across counties, notably in India and Western Europe.

  5. For the case of Western Europe these regions are actually countries (EU Member States).

  6. In these unrestricted regressions, the parameter β 0 will report the GDP-lights growth nexus in the reference region, whereas the parameters of the interaction terms will report deviations of the respective regions from the reference region.

  7. We use Huber–White robust covariances. χ² tests based on spatial heteroscedasticity and autocorrelation consistent (HAC) covariances (Kelejian and Prucha 2007) yield even stronger results (lower p values).

  8. More precisely, w ij  = [1/D ij ²]/Σ j [1/D ij ²], where D ij is the Euclidean geographic distance between counties i and j.

  9. We will return to this issue in Sect. 3 where we explore several additional possibilities to achieve parameter stability.

  10. The data is available for download at http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html.

  11. http://planningcommission.nic.in/plans/stateplan/ssphd.php?state=ssphdbody.htm.

  12. East India comprises all counties (districts) of the states of Bihar, Jharkhand, Orissa and West Bengal; North India those of Chhattisgarh, Haryana, Himachal Pradesh, Madhya Pradesh, Punjab, Uttar Pradesh and Uttarakhand; Northeast India those of Arunachal Pradesh, Assam, Manipur, Meghalaya, Mizoram and Sikkim; South India those of Andra Pradesh, Karnataka, Kerala and Tamil Nadu; and West India those of Maharashtra and Rajasthan.

  13. The 13 Western European countries are Austria, Belgium, Germany, Denmark, Finland, France, Ireland, Italy, the Netherlands, Portugal, Spain, Sweden and the United Kingdom. Luxembourg is excluded from the regressions in this section because it comprises a single NUTS3 region. It is included, however, in the panel estimations of short-term elasticities provided in the “Appendix”. Greece is excluded because Greek data may not be as reliable as the Penn World Tables data quality grade of B suggests. The questionable reliability is, among others, indicated by the poor data on public debt reported to the EU Commission during the financial crisis.

  14. While India and Brazil are rated C for data quality on the A–D scale of the Penn World Tables, more advanced OECD countries are mostly rated A. See the online Appendix of Chen and Nordhaus (2011: Table SI-4).

  15. The detailed results are available upon request. We do not test for parameter stability within the smaller countries because these tests are less reliable due to the small numbers of regional observations.

  16. More precisely, we use the specification in columns (3) of Tables 1, 2, 3 and 4 for all regressions reported in Sects. 3.1, 3.2, 3.3 and 3.4. That is, we always include region-specific constants, while we omit spatial lags which typically proved to be statistically insignificant before. For the sake of brevity, we only report the results of the χ² tests on parameter stability in the tables shown in this section. Detailed regression results are available from the authors on request.

  17. We are most grateful to an anonymous reviewer for suggesting several of the following modifications and extensions of our regression model.

  18. Note that even if controlling for income density were to achieve parameter stability, reliable predictions of regional GDP growth would require information on initial income levels, which will often not be available for regions in developing countries.

  19. For data description and download go to http://ngdc.noaa.gov/eog/dmsp/download_radcal.html.

  20. Western Europe as defined in Sect. 2 consists of 13 individual countries for which there is, to the best of our knowledge, no internationally comparable delineation of labor market regions or similarly meaningful regions. For India, we attempted to base the estimations on some 475 urban agglomerations (as of the 2011 Census). However, shape files do not appear to be readily available for these urban agglomerations and it proved impossible to figure out which districts are lying inside or outside these urban agglomerations. We therefore restrict ourselves to distinguishing urban and rural Indian districts by defining thresholds of lights intensity (see Sect. 3.4 below).

  21. Focusing on the estimates for the reference regions, the coefficients of night lights growth for the reference regions in Brazil and the United States with population growth as the dependent variable are significantly positive and of similar size as for the baseline results with GDP growth as the dependent variable. By contrast, the coefficient of night lights growth for the reference region in India appears to be unreasonably high, whereas the corresponding coefficient is negative, though insignificant, for the reference region in Western Europe. When considering the growth in per capita income as the alternative dependent variable, most coefficients of night lights growth for the reference regions tend to be negative (except for Western Europe), which appears to be counterintuitive.

  22. Alternatively, we classified a county to be urban if the share of unlit pixels (light intensity = 0) in this county is below a predefined threshold. The results, available from the authors upon request, are very similar to those reported below.

  23. For a narrow definition of urban areas, i.e., a lights intensity threshold of 63, the parameters of the lights intensities are estimated to be around 0.7, which is substantially higher than the value of 0.3 suggested by the estimates of Henderson et al. (2012).

  24. The shares of manufacturing and agriculture are based on value added data at the county level for India and Brazil; these shares are based on employment data for the United States. For the Western European countries no internationally comparable data on sector shares at the NUTS-3 level could be obtained for the period under consideration.

  25. For instance, population size and density enter significantly positive for the United States and Western Europe but are insignificant for India and significantly negative for Brazil. The sector shares are mostly negatively related with GDP growth.

  26. Due to data availability problems already mentioned, the estimations for Western Europe do not include the sector shares.

  27. There is just one estimation for which the χ² test is significant “only” at the four percent level (Brazil with non-saturated night lights data and all counties as single observations (row 4b, sample A).

References

  • Alesina, A., Michalopoulos, S., & Papaioannou, E. (2015). Ethnic inequality. Journal of Political Economy, forthcoming.

  • Berliant, M., & Weiss, A. (2013). Measuring economic growth from outer space: A comment. (MPRA Paper No. 51713) (posted 26 November 2013). http://mpra.ub.uni-muenchen.de/51713/

  • Chen, X., & Nordhaus, W. D. (2011). Using luminosity data as a proxy for economic statistics. Proceedings of the National Academy of Sciences, 108(21), 8589–8594.

    Article  Google Scholar 

  • Gennaioli, N., La Porta, R., Lopez-de-Silanes, F., & Shleifer, A. (2013). Human capital and regional development. Quarterly Journal of Economics, 128(1), 105–164.

    Article  Google Scholar 

  • Henderson, J. V., Storeygard, A., & Weil, D. N. (2012). Measuring economic growth from outer space. American Economic Review, 102(2), 994–1028.

    Article  Google Scholar 

  • Hodler, R., & Raschky, P. A. (2014). Regional favoritism. Quarterly Journal of Economics, 129(2), 995–1033.

    Article  Google Scholar 

  • Kelejian, H. H., & Prucha, I. (2007). HAC estimation in a spatial framework. Journal of Econometrics, 140(1), 131–154.

    Article  Google Scholar 

  • Michalopoulos, S., & Papaioannou, E. (2013). Pre-colonial ethnic institutions and contemporary African development. Econometrica, 81(1), 113–152.

    Article  Google Scholar 

  • Small, C., Elvidge, C. D., Balk, D., & Montgomery, M. (2011). Spatial scaling of stable night lights. Remote Sensing of Environment, 115(2), 269–280.

    Article  Google Scholar 

  • Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge, MA: MIT Press.

    Google Scholar 

Download references

Acknowledgments

The authors thank Marcus Berliant and David Weil for helpful comments and Michaela Rank for excellent research assistance. They are most grateful to an anonymous reviewer for various helpful suggestions.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Mareike Söder.

Appendix: Stability of short-term elasticities

Appendix: Stability of short-term elasticities

In this appendix, we report the estimation results for the short-term relationship between observed GDP and night lights intensity as well as the corresponding tests of parameter stability across regions. We estimate essentially the same model as Henderson et al. (2012: Table 2) for panels of annual data for districts in India, municipalities in Brazil, counties in the United States and NUTS3 regions in Western Europe. More specifically, we estimate, separately for each country,

$$ \ln Y_{it} = \alpha + \beta_{0} \ln L_{it} + \delta_{i} + \delta_{t} + u_{it} , $$
(3)

where lnY it and lnL it denote the natural logs of GDP and night lights intensity in county i and year t, δ i and δ t county- and year-fixed effects, α a global intercept, β 0 the elasticity of GDP with respect to night lights and u it the error term that may be heteroscedastic. We estimate Eq. (3) using the panel fixed effect estimator, accounting for heteroscedasticity in the errors by clustering the standard errors at the county level. We test the stability of β 0 across regions in the same way as in the cross-section growth regressions in Sect. 2: We add a set of interaction terms between lights, lnL it , and dummies for all (but one) regions, D r , r = 2, …, R, to Eq. (3), and test if the parameters of these interaction terms are jointly zero by means of a robust χ2 test (based on the clustered covariances).

The results for India, Brazil, the United States and Western Europe are shown in Tables 9, 10, 11 and 12. Stability of the parameter β 0 is clearly rejected in all four cases.

Table 9 Stability of short-term elasticity of GDP with regard to lights for India across five regions
Table 10 Stability of short-term elasticity of GDP with regard to lights for Brazil across five regions
Table 11 Stability of short-term elasticity of GDP with regard to lights for the United States across eight BEA regions
Table 12 Stability of short-term elasticity of GDP with regard to lights for Western Europe across 14 countries

About this article

Verify currency and authenticity via CrossMark

Cite this article

Bickenbach, F., Bode, E., Nunnenkamp, P. et al. Night lights and regional GDP. Rev World Econ 152, 425–447 (2016). https://doi.org/10.1007/s10290-016-0246-0

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10290-016-0246-0

Keywords

  • Night lights
  • Regional GDP
  • Stability of lights elasticities
  • Emerging economies
  • Developed economies

JEL Classification

  • E01
  • O11
  • O47
  • R11