Chinese vs. World Bank Development Projects: Insights from Earth Observation and Computer Vision on Wealth Gains in Africa, 2002-2013

TL;DR

This study addresses whether international development projects improve living conditions in Africa by leveraging an EO–ML causal framework and a satellite-derived wealth index. It analyzes World Bank and Chinese aid across 36 countries (2002–2013) in 9,899 neighborhoods, pairing pre-treatment daytime satellite imagery with a rich covariate set to estimate sector-specific ATEs via inverse probability weighting. The main finding is that both donors raise wealth on average, with larger and more consistent gains for China, and that image-based adjustment generally reduces estimated effects compared to tabular models, highlighting the role of spatial selection in aid placement. The results underscore the value of earth observation and machine learning for causal inference in development, while revealing distinct allocation mechanisms between donors and meaningful sector heterogeneity; the approach also emphasizes the importance of robust identification and sensitivity analyses for informing policy relevance.

Abstract

Debates about whether development projects improve living conditions persist, partly because observational estimates can be biased by incomplete adjustment and because reliable outcome data are scarce at the neighborhood level. We address both issues in a continent-scale, sector-specific evaluation of Chinese and World Bank projects across 9,899 neighborhoods in 36 African countries (2002-2013), representative of ~88% of the population. First, we use a recent dataset that measures living conditions with a machine-learned wealth index derived from contemporaneous satellite imagery, yielding a consistent panel of 6.7 km square mosaics. Second, to strengthen identification, we proxy officials' map-based placement criteria using pre-treatment daytime satellite images and fuse these with tabular covariates to estimate funder- and sector-specific ATEs via inverse-probability weighting. Incorporating imagery often shrinks effects relative to tabular-only models. On average, both donors raise wealth, with larger and more consistent gains for China; sector extremes in our sample include Trade and Tourism (330) for the World Bank (+12.29 IWI points), and Emergency Response (700) for China (+15.15). Assignment-mechanism analyses also show World Bank placement is often more predictable from imagery alone (as well as from tabular covariates). This suggests that Chinese project placements are more driven by non-visible, political, or event-driven factors than World Bank placements. To probe residual concerns about selection on observables, we also estimate within-neighborhood (unit) fixed-effects models at a spatial resolution about 67 times finer than prior fixed-effects analyses, leveraging the computer-vision-imputed IWI panels; these deliver smaller but, for Chinese projects, directionally consistent effects.

Figures (17)

• Figure 1: Visualizing the shape of the data objects.
• Figure 2: (a) Traditional DAG with unobserved confounder $U_i$. (b) With satellite image $\mathbf{M}_{i(\overline{t}-1)}$, a function $f_U(\cdot)$ maps $\mathbf{M}_{i(\overline{t}-1)}$ into the latent confounder $U_{i(t-1)}$, blocking the back‐door path when conditioned upon. $X_{i(t-1)}$ are observed covariates and $R_{i(t-1)}$ are confounders not present in the image.
• Figure 3: African aid by start year and funder from 2002-2013.
• Figure 4: African aid projects 2002-2013 by sector and funder.
• Figure 5: Geographical distribution of World Bank and Chinese projects. Note that some aid projects cannot be distinguished visually at the continental scale.