Economic Development in Pixels

Stylized graphic of the globe with varying intensity of light

Spatially disaggregated data on key indicators of development are crucial to the study of wide-ranging questions regarding economic progress, inequality, and violence and conflict, among others. Such data are also vital for the effective design of poverty alleviation policies targeting places that need them most.

While developed countries have made strides in collecting and distributing economic data at the subnational level, these indicators are lacking in developing countries. This fact is especially true for Sub-Saharan Africa. As a result, researchers have turned to using satellite imagery of luminosity at night (nightlights), as they are a natural proxy for economic development.

In Barcelona School of Economics Working Paper 1433, “Economic Development in Pixels,” John D. Huber and Laura Mayoral show that using nightlights as a proxy for development in empirical analysis can lead to misleading conclusions, as nightlights suffer from particularly problematic errors. In response, the authors combine a broad array of spatial datasets—not just night-lights—with modern machine-learning techniques to create a new, finely disaggregated measure of economic well-being. This metric is both economically interpretable and free of the biases that afflict night-light data.

Nightlights and Development

Nightlights have become a widely-used tool for the study of economic development in places with little to no coverage of traditional economic data. Nightlights have several characteristics that make them a very attractive proxy:

  1. Nightlights are measured in very small pixels that translate to cells of approximately 1 km2. This allows researchers to aggregate them at essentially any level: provinces, counties, municipalities, neighborhoods, etc.
  2. Yearly nighlight data exist for the whole world since 1992, providing a better time coverage than available through national accounts in most developing countries.

However, there are two important limitations of nightlights as proxy for development:

  1. Satellite images lack economic interpretation. A brighter pixel signals more activity than a dimmer one, but it says nothing about whether a place is above or below a poverty line.
  2. Nightlights suffer from non-classical measurement error. If the error in night-light readings were unrelated to true living standards, it would be “classical” and could be corrected with standard tools. In reality, the error varies systematically with development and therefore cannot be treated so easily.

One of the main reasons for this error is when satellites misclassify dimly lit areas as having no light at all. Satellites record complete darkness across large swathes of the developing world. However, these areas are not always unlit. In sparsely populated regions with faint artificial lighting, satellite sensors cannot distinguish weak human lights from background glow, so the pixels are coded as zero.

For example, between 2006 and 2018, satellites classified about 92% of Sub-Saharan Africa as “dark.” This darkness causes widespread misclassification, tagging some economically active areas as poor and, occasionally, exaggerating prosperity elsewhere. Such errors introduce biases that standard correction techniques cannot readily fix.

A New Measure of Spatial Consumption per Capita

The authors of this study introduce a new, spatially disaggregated measure of economic well-being that is readily interpretable in economic terms and free of the non-classical measurement error that plagues night-light data. Their fine-grained estimate of per-capita consumption is built in three stages.

  1. They combine World Bank figures with Demographic and Health Survey (DHS) data to translate the DHS asset index into a monetary measure of consumption.
  2. They use that variable to train a random-forest model—drawing on a wide range of predictors including night-lights, remoteness, geography, disease environment, CO₂ emissions, population, and ecosystem characteristics.
  3. The trained model generates estimates of log consumption per capita for every 10 × 10 km cell.

The result is the Spatially Explicit Development (SED) dataset: annual, cell-level measures of consumption and poverty for 42 Sub-Saharan African countries between 2003 and 2018.

The new dataset is an improvement with respect to previously available measures such as nightlights. Figure 1 (below) illustrates how the SED differs from nightlights data.

Panel A, on the left, shows a choropleth or shaded map of nightlights covering Tanzania and its borders. Panel B, on the right, shows a choropleth of SED covering the same area.

Note: Panel (a) shows the map of (log) nightlights in 2013. Dark areas on the map have a value of 0. Panel (b) shows the map of consumption per capita in 2013 using the estimates developed in this paper.

The first striking feature in the Panel A is the vast expanse of darkness: 89% percent of cells are dark, and population estimates suggest 54% of Tanzanians live in those cells.

The map also shows that the lit areas are almost exclusively large towns and along major roads, and that there is almost no variation in brightness levels when one crosses a national border. 

Panel B shows consumption per capita using the new SED data. In addition to the fact that this data has a direct economic interpretation, there are several important differences from the map of nightlights data:

  • It captures significant variation within the dark areas of the nightlights map
  • It reveals several areas that appear dark in Panel A but have relatively high economic development, such as the southwest area of Tanzania near Lake Malawi
  • It clearly shows how consumption per capita changes as one moves away from cities or major roads
  • It captures sharp changes in consumption per capita in some borders, such as the border between Tanzania and Mozambique, and not in others, such as that between Tanzania and Kenya

Institutions in the Spotlight

Beyond offering a clearly interpretable metric of per-capita consumption, the authors’ new measure also sidesteps the specific non-classical errors that plague night-light data.

To illustrate its value, they re-examine two influential studies on the role of institutions in development. Michalopoulos and Papaioannou (2013, 2014) used night-lights as a proxy for prosperity and concluded that (i) centralized pre-colonial ethnic institutions spur growth, while (ii) modern national institutions—rule of law, corruption control, and similar factors—have little effect.

When the same models are run with the authors’ Spatially Explicit Development (SED) data in place of night-lights, the conclusions reverse. Consistent with the broader literature that highlights the importance of national institutions, the re-analysis shows that areas with stronger state-level institutions in Sub-Saharan Africa are significantly wealthier, whereas ethnic centralization has negligible impact.

The authors trace this reversal to the non-classical measurement error embedded in night-light imagery—an issue not unique to the original studies but common to night-light data across much of the developing world.

A Guiding Light for Policy Intervention

Although satellite night-lights are widely used as a proxy for economic activity, they pose serious problems in econometric analysis because they suffer from non-classical measurement error.

Huber and Mayoral address this issue by integrating night-lights with a wide range of spatial datasets and constructing a new, fine-grained measure of well-being. Their metric avoids the biases inherent in night-lights and expresses outcomes in clear economic units, allowing empirical results to be interpreted meaningfully.

This approach yields a more nuanced picture of local development, useful not only to researchers, but also to policymakers who need to pinpoint where progress is occurring and where targeted intervention is most urgent.