Horizontal Inequality & Conflict

>>>Economic Horizontal Inequality is Linked to Conflict Persistence: High-Resolution Evidence from Satellite-Derived Wealth Measures in Africa

Skip to the findings

Using annual satellite-derived wealth estimates at 6.72-km resolution across Africa, this study separates the start of conflict from what happens after armed violence is underway. It finds no statistically significant onset association, but consistent links to rebellion continuation, casualty intensity, and shock vulnerability.

1 · Adel Daoud2 · 3

  1. 1Linköping University
  2. 2Institute for Analytical Sociology, Linköping University, and Chalmers University of Technology
  3. 3Department of Government, University of Texas at Austin

Last updated: · 80 pages · Preprint

Read the preprint Explore data and replication code

DOI: 10.2139/ssrn.6963338

Spatial resolution
6.72 × 6.72 km cells
Group-year analysis
1992–2020
Group-years
5,893, including 60 onsets
Shock analysis
1995–2024
Wealth-index shocks
5,352
Data backbone
Satellite IWI + GeoEPR + UCDP GED
01 / Core finding

What does satellite wealth mapping show about group inequality and armed conflict?

Economic horizontal inequality appears less as a general trigger of new conflicts than as a condition associated with what happens after conflict becomes active. Across politically relevant ethnic groups in Africa, the study finds an onset null, but stronger associations with active-rebel years, continuation of ongoing rebellions, casualty intensity, and vulnerability to localized wealth shocks.

Economic horizontal inequality

Economic differences between identity groups. The paper compares each politically relevant ethnic group’s population-weighted mean International Wealth Index (IWI) with its national mean.

φ = group mean IWI ÷ national mean IWI − 1

How to read φ

φ < 0 indicates relative deprivation; φ > 0 indicates relative advantage. A value of −0.5 means a group’s predicted mean IWI is 50% below the national mean; +0.5 means 50% above.

IWI is a 0–100 asset and living-standards index. It is not observed income or monetary wealth.

What the resolution changes: the populated 6.72-km cells cover about 45.2 km² each—roughly 270 times finer in area than the 1° production grids used in classic horizontal-inequality research. That makes group wealth position and localized shocks visible on the same measurement backbone.

02 / Conflict cycle

Does group wealth inequality predict conflict onset—or conflict persistence and severity?

Not onset in these data. The onset estimates are imprecise and their 95% confidence intervals cross the no-association value of 1; political marginalization is the stronger onset predictor. By contrast, both relative advantage and relative deprivation are associated with active-rebel years and higher fatality counts, while relative deprivation predicts whether an ongoing rebellion continues into the next year.

Economic horizontal inequality across four stages of the conflict cycle A forest plot on a logarithmic ratio scale. Onset confidence intervals cross one for relative advantage and deprivation. Relative advantage and deprivation estimates are above one for active-rebel years and fatality counts. Only relative deprivation is clearly above one for continuation of an ongoing rebellion. Inequality across the conflict cycle Per-standard-deviation ratios from full models; horizontal lines are 95% confidence intervals Conflict onset Advantage OR 1.31 · 0.77–2.07 Deprivation OR 1.23 · 0.76–2.00 Active-rebel year Advantage OR 1.82 · 1.31–2.49 Deprivation OR 1.27 · 1.00–1.61 Rebellion continuation Advantage OR 1.05 · 0.56–2.04 Deprivation OR 1.78 · 1.24–2.63 Fatality count Advantage IRR 1.69 · 1.25–2.28 Deprivation IRR 1.49 · 1.16–1.93 0.5 1 2 3 Effect ratio per 1-SD increase · logarithmic scale
Figure 1. The relationship changes across the conflict cycle. Circles show relative advantage (φ+); diamonds show relative deprivation (φ−). The dashed line is the null ratio of 1. OR = odds ratio; IRR = incidence-rate ratio. The fatality-count model is limited to active-conflict group-years with at least one fatality. Estimates are associational.
Exact per-standard-deviation estimates shown in Figure 1
Conflict stageMeasureRelative advantage, φ+Relative deprivation, φ−Plain-language reading
Conflict onsetOdds ratio1.31 [0.77, 2.07]1.23 [0.76, 2.00]Neither confidence interval excludes 1.
Active-rebel yearOdds ratio1.82 [1.31, 2.49]1.27 [1.00, 1.61]Groups farther from the national mean appear more often in rebel-active years.
Rebellion continuationOdds ratio1.05 [0.56, 2.04]1.78 [1.24, 2.63]Relative deprivation—not advantage—predicts year-to-year continuation.
Fatality countIncidence-rate ratio1.69 [1.25, 2.28]1.49 [1.16, 1.93]Both directions of divergence track deadlier fighting within active conflicts.
59%

In a model-based associational calculation, setting both inequality components to zero reduces predicted fatalities across active group-years from roughly 158,000 to 66,000—a 59% difference. This is not the share of deaths proven to have been caused by inequality.

03 / Local wealth shocks

When do localized wealth shocks correspond to more conflict?

Dense wealth-shock scenarios do not correspond to more conflict events on average. The country-level contrasts are generally close to zero, neutral, or negative. The estimated contrast becomes positive only where local group wealth sits more than about 50% above or below the national mean—a range covering roughly 7% of populated grid cells on average.

Figure 2. Direction of the estimated dense-minus-sparse wealth-shock contrast across local economic horizontal inequality. This band summarizes sign and threshold, not effect magnitude. Estimates at the most extreme inequality values have wider uncertainty and should be interpreted cautiously.

The analysis compares an annual sparse scenario with 70 shocks (the 25th percentile) to a dense scenario with 466 shocks (the 90th percentile). A shock is a three-year proportional fall in predicted IWI exceeding five within-country standard deviations among negative changes. These are sharp changes in a model-derived living-standards index, not directly observed income losses.

Are satellite-measured wealth shocks simply drought shocks?

Mostly not under the paper’s drought definition. Among 5,352 wealth-index shocks, 33% coincide with at least moderate drought and 8% with severe drought—almost the same rates as year-matched non-shock cells. Put differently, about two-thirds carry no moderate-drought signature.

Figure 3. Drought coincidence among shock and year-matched non-shock cells, on a common 0–100% scale. The near-equality shows why satellite-predicted wealth shocks should not be treated as repackaged drought events; it does not prove that the remaining shocks are unrelated to climate.
04 / Reverse direction

Does conflict exposure reduce satellite-measured local wealth?

The staggered event study does not detect a post-exposure decline in predicted local IWI relative to unexposed cells in the same country and year. The result bounds a detectable neighborhood-scale decline; it does not establish that conflict never destroys wealth. Small pre-trends and persistence in the satellite prediction model counsel caution.

Overall post-exposure estimate
+0.20 IWI points
Ten-year IWI estimate
+0.13 [−0.06, +0.32]
Two-way fixed-effects check
0.00 mean post-effect

The design compares 5,973 conflict-exposed 6.72-km cells with never-exposed cells, using the year before first exposure as the reference. The cell’s relative national wealth position changes only −0.016 after ten years. This quiet reverse-direction result also reduces concern that visible conflict damage mechanically drives the paper’s satellite wealth measure.

05 / Research design

How was the relationship between group inequality, wealth shocks, and conflict studied?

The paper combines one high-resolution wealth backbone with three complementary designs. Annual satellite-predicted IWI is overlaid with dynamic ethnic settlement areas, political status, population, and georeferenced conflict events, allowing onset, ongoing activity, severity, shocks, and the reverse wealth trajectory to be compared on consistent spatial units.

  1. Measure relative group wealth

    Population-weighted IWI means are computed for politically relevant ethnic groups and their countries. Their ratio, φ, separates relative advantage from relative deprivation. Spatial dispersion within settlement areas is measured separately.

  2. Separate the conflict cycle

    Firth logistic models estimate conflict onset and active-rebel years for 1992–2020. Discrete-time hazards separate entry from continuation. A hurdle model estimates any fatalities and fatality counts within active conflicts.

  3. Compare shocks and reverse direction

    A spatiotemporal counterfactual framework compares dense and sparse shock scenarios for 1995–2024, including conditional contrasts by φ. A Callaway–Sant’Anna event study then traces local IWI after first conflict exposure.

Core data sources

Annual machine-learning predictions of the International Wealth Index; GeoEPR ethnic settlement polygons, political access, and rebel links; UCDP Georeferenced Event Dataset conflict locations and fatalities; LandScan population weights; and SPEI drought measures. Processed data, source documentation, and replication code are publicly linked below.

Data and code
Processed data, documentation, scripts, tables, and figures are available in the public replication repository.
Validation
Group wealth positions correlate with an independent satellite-derived wealth product: Pearson r = 0.56 across 190 groups.
Funding and interests
The authors declare no relevant funding and no competing interests.
Ethics
The study uses secondary aggregate geospatial and event data and involved no new interaction with human participants.
06 / Interpretation

What can this study establish—and what can it not establish?

The paper provides measurement-enabled evidence about where inequality is associated with conflict dynamics; it does not offer a definitive causal account of why wars begin. The strongest reading is a distinction among onset, persistence, severity, and conditional shock vulnerability—not a claim that inequality universally causes armed conflict.

  • Group-year estimates are associational. Predictors and outcomes are measured contemporaneously, so simultaneity and reverse causality remain possible.
  • Spatial contrasts depend on assumptions. A causal reading requires exchangeability, positivity, correct treatment-model specification, and the framework’s spatial and temporal interference assumptions.
  • IWI is model-derived. Satellite predictions measure an asset and living-standards index, not observed household income or monetary wealth; local shocks are sharp changes in those predictions.
  • Outcomes are conditional and geographically bounded. Fatality estimates condition on active conflict, and the Africa-only scope limits generalization beyond the continent.
  • The reverse-direction estimate is a bound. Small pre-trends, smooth annual predictions, and the exclusion of conflict events outside populated prediction cells preclude a sharp causal conclusion about wealth destruction.

Short answers that prevent overreading

Does the paper show that economic inequality causes armed conflict?
No. The group-year findings are associational, and the spatial contrasts have a causal interpretation only under stated identifying assumptions.
Does the 59% estimate mean inequality caused 59% of battle deaths?
No. It is a model-based associational comparison of predicted fatalities when the two relative-wealth components are set to zero.
Are satellite wealth shocks the same as income or drought shocks?
No. They are unusually large three-year declines in predicted IWI. Only 33% coincide with moderate drought, nearly matching the 32% rate in comparison cells.
Does the event study prove that conflict does not destroy wealth?
No. It finds no detectable decline in this annual satellite measure at this neighborhood resolution; short-run or unmeasured losses may still occur.
07 / Implications

Why does the onset-versus-dynamics distinction matter for peace and inequality policy?

If group-based inequality is associated more with keeping conflict active and making it deadlier than with starting it, inequality reduction is relevant to peace-sustaining policy as well as prevention. The paper connects reduced inequalities (SDG 10) with peaceful and inclusive societies (SDG 16), treating them as operationally coupled rather than separate goals.

The results motivate testing whether reducing economic disparities between identity groups—and rapidly protecting highly unequal regions from localized economic disruption—can shorten or dampen ongoing violence. Adaptive social protection, insurance, and rapid transfers are plausible instruments, but the study does not estimate the effects of those policies.

08 / Paper, data, and citation

Where can readers access, cite, and reproduce the research?

The DOI provides the persistent paper link, and the public replication repository contains processed data, source documentation, and analysis code. Because the DOI record still carries the earlier SSRN title, readers citing the present revision should retain the visible version date below.

Suggested citation

Fagerlind, Johannes, Adel Daoud, and Connor Jerzak. 2026. “Satellite Wealth Mapping Reveals Where Group Inequality Turns Economic Shocks into Armed Conflict.” Current manuscript, 13 August 2026. https://doi.org/10.2139/ssrn.6963338.

Related Work

Adel Daoud, Connor T. Jerzak. Planetary Causal Inference: Understanding the Environment, Society, and Economy through Earth Observation and AI Systems. A Book Project, 2026+.
@article{daoud2026+planetary,
  title={Planetary Causal Inference: Understanding the Environment, Society, and Economy through Earth Observation and AI Systems},
  author={Daoud, Adel and Connor T. Jerzak},
  journal={A Book Project},
  year={2026+},
  volume={},
  pages={},
  publisher={Under contract with Cambridge University Press}
}

–

Adel Daoud, Cindy Conlin, Connor T. Jerzak. Chinese vs. World Bank Development Projects: Insights from Earth Observation and Computer Vision on Wealth Gains in Africa, 2002-2013. World Development, 202: 107328, 2026.
@article{daoud2026chinese,
  title={Chinese vs. World Bank Development Projects: Insights from Earth Observation and Computer Vision on Wealth Gains in Africa, 2002-2013},
  author={Daoud, Adel and Cindy Conlin and Connor T. Jerzak},
  journal={World Development},
  year={2026},
  volume={202},
  pages={107328},
  publisher={}
}
[Overview][Data][>]

–

Warren Zhu Fucheng, Connor T. Jerzak, Adel Daoud. Optimizing Multi-Scale Representations to Detect Effect Heterogeneity Using Earth Observation and Computer Vision: Applications to Two Anti-Poverty RCTs. Proceedings of the Fourth Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR), 2025.
@article{fucheng2025optimizing,
  title={Optimizing Multi-Scale Representations to Detect Effect Heterogeneity Using Earth Observation and Computer Vision: Applications to Two Anti-Poverty RCTs},
  author={Fucheng, Warren Zhu and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the Fourth Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR)},
  year={2025},
  volume={},
  pages={},
  publisher={}
}
[Overview][Data][Code]
[Video][>]

–

Markus B. Pettersson, Connor T. Jerzak, Adel Daoud. Debiasing Machine Learning Predictions for Causal Inference Without Additional Ground Truth Data: 'One Map, Many Trials' in Satellite-Driven Poverty Analysis. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-2026), 40: 39106-39115, 2026.
@article{pettersson2026debiasing,
  title={Debiasing Machine Learning Predictions for Causal Inference Without Additional Ground Truth Data: 'One Map, Many Trials' in Satellite-Driven Poverty Analysis},
  author={Pettersson, Markus B. and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence (AAAI-2026)},
  year={2026},
  volume={40},
  pages={39106-39115},
  publisher={}
}
[Overview][Code][>]

–

Connor T. Jerzak, Ritwik Vashistha, Adel Daoud. Effect Heterogeneity with Earth Observation in Randomized Controlled Trials: Exploring the Role of Data, Model, and Evaluation Metric Choice. ArXiv Preprint, 2024.
@article{jerzak2024effect,
  title={Effect Heterogeneity with Earth Observation in Randomized Controlled Trials: Exploring the Role of Data, Model, and Evaluation Metric Choice},
  author={Jerzak, Connor T. and Ritwik Vashistha and Adel Daoud},
  journal={ArXiv Preprint},
  year={2024},
  volume={},
  pages={},
  publisher={}
}
[Overview]

–

Satiyabooshan Murugaboopathy, Connor T. Jerzak, Adel Daoud. Platonic Representations for Poverty Mapping: Unified Vision-Language Codes or Agent-Induced Novelty?. Proceedings of the 7th International Conference on Social Computing (ICSC 2026), of Communications in Computer and Information Science, 2026.
@article{murugaboopathy2026platonic,
  title={Platonic Representations for Poverty Mapping: Unified Vision-Language Codes or Agent-Induced Novelty?},
  author={Murugaboopathy, Satiyabooshan and Connor T. Jerzak and Adel Daoud},
  journal={Proceedings of the 7th International Conference on Social Computing (ICSC 2026), of Communications in Computer and Information Science},
  year={2026},
  volume={},
  pages={},
  publisher={Springer Nature}
}
[Overview][Data][Slides]
[Video]

–

Connor T. Jerzak, Fredrik Johansson, Adel Daoud. Image-based Treatment Effect Heterogeneity. Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR), 213: 531-552, 2023.
@article{jerzak2023image,
  title={Image-based Treatment Effect Heterogeneity},
  author={Jerzak, Connor T. and Fredrik Johansson and Adel Daoud},
  journal={Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR), Proceedings of Machine Learning Research (PMLR)},
  year={2023},
  volume={213},
  pages={531-552},
  publisher={}
}
[Overview][Data][Code][>]

–

Kazuki Sakamoto, Connor T. Jerzak, Adel Daoud. A Scoping Review of Earth Observation and Machine Learning for Causal Inference: Implications for the Geography of Poverty. Hall, Ola and Ibrahim Wahab (eds.), Geography of Poverty, 2026.
@article{sakamoto2026scoping,
  title={A Scoping Review of Earth Observation and Machine Learning for Causal Inference: Implications for the Geography of Poverty},
  author={Sakamoto, Kazuki and Connor T. Jerzak and Adel Daoud},
  journal={Hall, Ola and Ibrahim Wahab (eds.), Geography of Poverty},
  year={2026},
  volume={},
  pages={},
  publisher={Edward Elgar Publishing (Cheltenham, UK)}
}
[Overview][Data][Video][>]

–

Connor T. Jerzak, Fredrik Johansson, Adel Daoud. Integrating Earth Observation Data into Causal Inference: Challenges and Opportunities. ArXiv Preprint, 2023.
@article{jerzak2023integrating,
  title={Integrating Earth Observation Data into Causal Inference: Challenges and Opportunities},
  author={Jerzak, Connor T. and Fredrik Johansson and Adel Daoud},
  journal={ArXiv Preprint},
  year={2023},
  volume={},
  pages={},
  publisher={}
}
[Overview][Data][Code]

–

Connor T. Jerzak, Adel Daoud. CausalImages: An R Package for Causal Inference with Earth Observation, Bio-medical, and Social Science Images. ArXiv Preprint, 2023.
@article{jerzak2023causalimages,
  title={CausalImages: An R Package for Causal Inference with Earth Observation, Bio-medical, and Social Science Images},
  author={Jerzak, Connor T. and Adel Daoud},
  journal={ArXiv Preprint},
  year={2023},
  volume={},
  pages={},
  publisher={}
}
[Overview][Code]