Anthony Mveyange, a current EASST visiting fellow, is pursuing a PhD in Economics from the University of Southern Denmark. He is presenting this paper at a seminar on 3/13/2015 at Colorado State University, where he spent two years as a Fulbright scholar.
The transition from low to high growth performance has been a central feature of policy discussions on Africa’s economic prosperity over the last decade. These growth spurts have also brought along growing concerns over income disparities across regions (see, WorldBank, 2012, for more discussions). The existing evidence on income inequality in Africa is polarized between two extremes. At one extreme there is the micro evidence that uses individuals and households as units of analysis, and at the other extreme there is macro evidence that uses countries and supra-nations as units of analysis. However, virtually no evidence exists on regional - the meso level - income inequalities in Africa as a whole. Kim (2008) asserts that the main reason for this vacuum has mainly been the lack of reliable and consistent sub-national income data.
In this paper I employ night lights (henceforth lights) data to circumvent these data limitations and estimate a proxy for regional income inequality in Africa. The main hypothesis is that lights are good proxies for regional income inequality in as much as they are proxies for income (Papaioannou, 2013; Sutton et al., 2007), economic growth (Henderson et al., 2012; Chen and Nordhaus, 2011) and wealth (Ebener et al., 2005). To test this hypothesis, I i) construct regional income inequality (henceforth regional inequality) indices (i.e. Gini and Mean logarithmic de- viations - MLD) based on spatial income data and light data in 423 regions across 32 countries in Africa for years 1995, 2000 and 2005. The estimated indices are based on total and income per capita and total and per capita light intensity. ii) follow Henderson et al. (2012)’s statistical framework to test the extent to which light intensity inequality predicts income inequality in Africa during the same period. This allows statistical inference on whether light-based inequality indices are good proxies for income-based indices. iii) use light-based inequality indices to estimate and show the trends of regional income inequality in Africa during 1992 − 2012.
The relationship between lights and regional inequality is summarized in figure 1. The figure conveys the underlying theoretical conjecture which relates to Elvidge et al. (2009, p.1653) who distinctly offer an interesting implicit insight to the relationship between light and regional inequality. That is, they assert that "area with higher population counts in developing countries would be poorly lit and therefore have higher percentages of poor people (light being considered as a proxy for wealth)". The most direct implication of this assumption, in the context of my study - is that to the extent that lights are positive and strong correlates of income, poorly lit regions tend to have low income and are less wealthy.
Thus, the arrows in figure 1 show the association between wealth, income, lights and regional inequality. That is, wealth predicts the levels of income which in turn determines the intensity of lights. Relative to areas with high light intensity, poorly lit areas will tend to be less wealthy. This hypothesis also applies for within region variations across, for example, counties, districts or the grid cells as discussed later on. I exploit G-Econ data (for income data) and lights data to show lights as potentially alternative data source by estimating and comparing inequality indices based on these two data sources. Figure 1 and 2 respectively present the correlation of income and light-based Gini1 . An interesting observation to emerge from a visual inspection of these panels is a strong positive correlation (about 0.60) of the income and light-based Gini. Moreover, a non-parametric fit suggests a linear relationship.
Figure 4 and 5 present a visual spatial distribution of income and light-based inequality Gini. Figure 4 shows the distribution of the average total income Gini (left panel) and sum of light Gini (right panel) both during 1995 - 2005. Similarly, figure 5 shows the average distributions based on per capita measures during the same period. The one key message hard to miss from all these figures is that in general regions that have had high inequality based on income data also appear to have had high inequality as estimated using lights data. The converse holds true.
With this strong correlation evidence at hand, I then proceed to assess the causal links between lights inequality and income inequality. I find lights as moderately good proxies for regional income disparities in the absence of income data. The fixed effects regressions estimates suggest that one percentage point increase in light intensity Gini and MLD significantly increases Gini and MLD of about 0.02 and 0.01 percentage points across a range of specification tests.
Then, estimates of regional income inequality trends in 7482 regions across 52 African countries during 1992 − 2012 document four important results: i) increasing regional inequality trends across regions in Africa during 1992-2001, ii) declining trends during 2002 − 2012, iii) substantial variations across different regional groupings, indicating the sensitivity of inequality to country and regional differences and, iv) dominant role of the between inequality as a key driver of all these trends.
The choice of lights data is motivated by a recent growing literature which uses these data as a proxy for income, wealth and economic growth in the absence of traditional income data. This literature is sub-divided into four strands3 . First, Michalopoulos and Papaioannou (2013); Papaioannou (2013); Henderson et al. (2012, 2011) and Chen and Nordhaus (2011) who used lights as a proxy for income and economic growth in sub-Saharan Africa. Second, Alesina et al. (2012) used lights to estimate ethnic inequality across countries. Third, Elvidge et al. (2012) who developed a "night light development index" to measure human development and track the distribution of wealth and income across countries. Fourth, Elvidge et al. (2009) and more recently Pinkovskiy and Sala-i Martin (2014) who employed lights to estimate poverty.
However, my study differs from its predecessors in several ways. First, it extends the literature that used lights to estimate income, wealth and economic growth and proposes that lights do a fairly decent job to provide a sense of regional disparities when income data are unavailable. Elvidge et al. (2012) offered a first attempt but did not find any meaningful association. But their test was at country level suffering a potential omitted variable bias, limited variation and time invariance in their cross-sectional set-up. The present paper, on the contrary, not only covers regions which addresses the omitted variable bias but it also employs a panel analysis and exploits grid cell level variation to estimate inequality at regional level thus capturing variations at local scales over time. In addition, the paper is careful to avoid making claims about individual-level4 inequality or poverty. Second, Elvidge et al. (2009) and Pinkovskiy and Sala-i Martin (2014) have recently focused on estimating poverty and not regional inequality as is the case in the present paper. Third, this paper is consistent with Alesina et al. (2012) in as far as the use light data to construct inequality measure is concerned. However, it is different for three main reasons: i) it measures regional inequality unlike ethnic inequality, ii) it explores variation of regional inequality at local scales over a period of relatively longer period - two decades - than in Alesina et al. (2012)’s focus at country level and over a relatively shorter period, iii) it checks the validity of light-based inequality indices not addressed by Alesina et al. (2012) and, iv) it includes light-based decomposable measures of regional inequality to identify the sources of the observed regional inequality, an element absent in all the previous studies cited.
The focus on regional inequality in Africa is interesting for several reasons. First, regional inequality has been associated with conflicts and civil unrest in a number of African countries. Nonetheless, studies on conflict have only used non-income and welfare measures of regional inequality with no success in measuring regional income inequality because of the lack of data. Excellent examples of such studies include Østby et al. (2009) and Sahn and Stifel (2003) who use demographic data on household education indicators, assets and other capabilities indicators to measure and decompose regional inequalities across African countries. Second, because regional inequality can also affect household income inequality (Kim, 2008) estimating it indirectly offers insights on the trends of household income inequality which has been a subject of intense debates in Africa (see, Deaton, 2005, for more details). Third, quantifying regional inequality trends also has clear policy implications especially when combined with growing concerns that the recent economic growth surges in Africa are not inclusive (WorldBank, 2012).
Relative to the existing literature, the main contribution of the present paper is its propo- sition to use lights data to approximate regional income disparities patterns, in the absence of traditional income data, across countries in Africa. Also, the paper complements two long pedi- gree strands of literatures. First, a long standing literature on regional and urban inequality (see Kim, 2008, for detailed discussion of this literature). Second, a recent burgeoning literature on the comparative development which has been preoccupied with understanding, among others, the reasons for income differences within and across countries.
1 In the main paper I also compute and compare the mean log deviation (MLD)
2 With light data I am able to estimate more regions relative to 423 regions sample used for the panel fixed effect analysis. In the panel fixed effect analysis, i only use 423 sample because I filter out good quality income data (from G-Econ) and 19 small African countries and pair them with light data, thus lower sample. However, this is not a concern with light data.
3 Other studies that have documented the use of light to approximate economic activities at the subnational level include Hodler and Raschky (2014) who used data on the intensity of light as a proxy for economic activities and, hence, GDP growth across 126 countries to estimate regional favoritism. For other parts of the world, Villa (2014) used light to approximate the growth of Colombian municipalities, Levin and Duke (2012) compared Israel and the West Bank to show that differences in lights reflect the underlying differences in subnational socio-economic activities across the two countries. In a different study.
4 The unit of analysis for inequality calculation is grid cell. Within a grid cell, any given average value can be the result of uniform or highly unequal income or light distribution
Alesina, A., S. Michalopoulos, and E. Papaioannou (2012). Ethnic inequality. Centre for Eco- nomic Policy Research.
Chen, X. and W. D. Nordhaus (2011). Using luminosity data as a proxy for economic statistics. Proceedings of the National Academy of Sciences 108 (21), 8589–8594.
Deaton, A. (2005). Measuring poverty in a growing world (or measuring growth in a poor world). Review of Economics and Statistics 87 (1), 1–19.
Ebener, S., C. Murray, A. Tandon, and C. C. Elvidge (2005). From wealth to health: modelling the distribution of income per capita at the sub-national level using night-time light imagery. International Journal of Health Geographics 4 (1), 5.
Elvidge, C., K. Baugh, S. Anderson, P. Sutton, and T. Ghosh (2012). The night light development index (nldi): a spatially explicit measure of human development from satellite data. Social Geography 7 (1).
Elvidge, C. D., P. C. Sutton, T. Ghosh, B. T. Tuttle, K. E. Baugh, B. Bhaduri, and E. Bright (2009). A global poverty map derived from satellite data. Computers & Geosciences 35 (8),1652–1660.
Henderson, J. V., A. Storeygard, and D. N. Weil (2012). Measuring economic growth from outer space. American Economic Review 102 (2), 994–1028.
Henderson, V., A. Storeygard, and D. N. Weil (2011). A bright idea for measuring economic growth. The American Economic Review 101 (3), 194–199.
Hodler, R. and P. A. Raschky (2014). Regional favoritism*. The Quarterly Journal of Economics, qju004.
Kim, S. (2008). Spatial inequality and economic development: Theories, facts, and policies.Urbanization and Growth, 133–166.
Levin, N. and Y. Duke (2012). High spatial resolution night-time light images for demographic and socio-economic studies. Remote Sensing of Environment 119, 1–10.
Michalopoulos, S. and E. Papaioannou (2013). Pre-colonial ethnic institutions and contemporary african development. Econometrica 81 (1), 113–152.
Østby, G., R. Nordås, and J. K. Rød (2009). Regional inequalities and civil conflict in sub-saharan africa1. International Studies Quarterly 53 (2), 301–324.