When the Counter-Insurgent Becomes the Killer

Wagner, the AES Coup Belt, and the State-Led Civilian-Targeting Turn in the Sahel, 2018–2025

Across Mali, Burkina Faso, and Niger—which formed the Alliance des États du Sahel in 2023 and upgraded it to a confederation in 2024—the perpetrators of mass civilian killing have shifted faster than the fighting itself. Since Wagner and Africa Corps personnel arrived, state forces and their external partners account for a growing share of civilian deaths. This report examines how clearly that shift appears in the ACLED (Armed Conflict Location & Event Data) record.

26,977
Total ACLED events, AES core 2018–2025
61,007
Total reported fatalities
34.5%
Civilian-targeting share of events
7.76×
Mali state-force civilian killing post/pre Wagner
6×
Mali external-force civilian killing post/pre Wagner

Introduction

Between August 2020 and July 2023, the central Sahel saw one of the post-Cold War's most concentrated coup waves [1]. Mali (twice), Burkina Faso (twice), and Niger replaced civilian governments with military juntas, pivoting to Russian partners — first Wagner, then Africa Corps [2][3]. The conventional read is geopolitical. The more basic question is who is now killing whom.

ACLED records for Western Africa (2018-01-01 to 2025-04-25) [4] frame the question: Did Russian deployment into the AES core coincide with a major shift in who is targeting civilians, distinct from the ongoing jihadist insurgency? We distinguish state forces, external forces, and non-state armed groups pre- and post-Russian arrival. Two robustness checks follow: a placebo (re-fit on West African countries without Russian forces) and a breakpoint test (anchored to French withdrawal instead of Russian arrival). We test whether post-coup violence reflects jihadist resurgence or state-led victimization.

Data and Methods

ACLED's Data Export Tool covers Western Africa (2018-01-01 to 2025-04-25). AES filtering yields 26,977 events; 9,300 carry the civilian-targeting flag, accounting for 22,744 fatalities [4]. When sources disagree, ACLED records the lower number, undercounting deaths. Eck's critique is addressed below [5].

Country-year merges add the Powell-Thyne coup dataset [1] and V-Dem v16 for regime indicators. Cleaning derives four actor-role categories (state, non-state armed group, external force, civilian/other) and three breakpoints: first coup, French withdrawal, Russian arrival.

Pre/post ratios use a month-level bootstrap (2,000 replications, 95% CIs). A negative-binomial regression with country fixed effects, an event-volume control, and a linear time trend yields the headline incidence-rate ratio (IRR), re-fit on 13 non-AES countries as a placebo (null expected if Wagner-specific). Burkina Faso (15-month) and Niger (12-month) have shorter post-windows than Mali (41); we flag this rather than averaging through it.

Findings

Finding 1 — Monthly conflict events and the post-coup level shift

Figure 1 plots total monthly conflict events per country with first-coup (dashed) and Russian-deployment (dotted) markers, plus orange horizontal pre/post-Wagner mean lines per panel labelled with their numerical values. Mali's monthly event count nearly doubles (87 → 169), peaking at 150–200/month from mid-2022. Burkina Faso climbs from 127 to 148 post-Africa-Corps; Niger from 48 to 66. Burkina Faso's series, however, peaks at the 2022 coup and settles before Africa Corps arrives, so its rise is coup-driven, not Russian-driven. The civilian-targeted-fatality decomposition that follows in Figure 2 is the analytical outcome variable; this figure establishes the overall conflict envelope. Figure 1b animates the same data year-by-year.

That Burkina Faso's and Niger's events visibly rose before their own coups fits Wilén's contagious civil-military imbalance: once Mali's August 2020 junta survived ECOWAS sanctions, regional officers updated their beliefs about coup viability and state-led violence rose [6]. Coup-contagion theory predicts this clustering [7]. Whether the rise is insurgent or state-led is Finding 2's question.

Finding 2 — Civilian-targeted fatalities, decomposed by perpetrator role

Figure 2 plots monthly civilian-targeted fatalities by perpetrator role across the AES core as stacked bars, with each panel's pre→post-Wagner role means displayed in the top-left textbox. In Mali's panel, the pre-Wagner bars are dominated by non-state killings. After Wagner, monthly averages shift sharply: state 10.70→83.10 (Mar 2022 Moura spike: 497), external 2.30→13.78, non-state only 54.21→64.32. Burkina Faso's panel shows post-Africa-Corps state-force escalation (state 30.81→83.40) alongside continued non-state activity. Niger's brief post-window is too short for confident inference. Mali's asymmetry is the central empirical claim.

In Mali, if post-coup violence were jihadist resurgence, non-state killings would be rising. Instead, they are holding flat. Stanton's strategic-violence framework fits this asymmetry: hierarchical regime forces respond to audience-cost calculations differently from loose insurgent groupings [8]. Mali fits Kalyvas's indiscriminate-violence pattern in zones a regime cannot fully control [9]. Hultman documents this for rebels [10]. Marten describes Wagner's African expansion [2].

Finding 3 — Bootstrap pre/post ratios, regression, and placebo

Figure 3 plots bootstrap pre/post fatality ratios with 95% CI error bars per perpetrator type, faceted by country. Mali's state and external bars sit far above 1 (7.76× and 6.00×, CIs 4.71–14.24 and 1.95–30.07). The non-state bar at 1.19× has a CI (0.88–1.62) crossing 1. Burkina Faso's state and external bars rise but its CIs all cross 1. Niger's state and external bars fall. The pooled negative-binomial returns IRR = 1.74 (1.11–2.74, p = 0.017; n = 264). The placebo on 13 non-AES countries returns IRR = 1.15 (p = 0.356, CI contains 1).

The placebo's silent CI isolates the AES effect from a continent-wide trend. Lyall & Wilson predicted these bars: external forces with firepower but no local intelligence cannot distinguish combatants from civilians, defaulting to indiscriminate force [11]. Mali's Russian-backed sweeps fit Fjelde & Hultman — government violence intensifies against Fulani communities JNIM recruits from but the regime cannot reach [12]. Russia's lack of human-rights conditionality on its partners — unlike France — incentivizes this shift.

Finding 4 — V-Dem regime trajectories

Figure 4 plots V-Dem v16's polyarchy index and regime category for each AES country with coup markers. Polyarchy drops step-wise at each coup year (2020-08-18, 2022-01-23, 2023-07-26; vlines mark year), falling below 0.3 by 2024 — V-Dem's electoral-autocracy band — while regime category transitions to closed autocracy (0). The French-exit breakpoint check returns IRR = 1.48 (p = 0.097), weaker than the Russian-arrival estimate.

These polyarchy collapses fit Roessler's coup–civil war trap, Mali the clearest case: a fractured officer corps since the 2012 Sanogo coup, with successive junta cycles (2012 Sanogo, 2020/2021 Goïta) each purging prior officers and widening the army's northern gap, inviting insurgent advance [13][14]. Reform would require trusting the officers just purged, so juntas chose force. Singh's coordination framework, applied to post-coup officer loyalty, explains why this produces civilian harm: visible sweeps against "internal enemies" lock in pro-coup officer loyalty as slow population-centric COIN cannot [15]. That the Wagner-arrival breakpoint outperforms the French-exit confirms the substitution drove the change.

Finding 5 — Mali geographic shift, pre vs post Wagner

Figure 5 overlays Mali's civilian-targeted events pre-Wagner (blue) and post-Wagner (red), sized by fatalities; the top-left textbox displays the admin1-share aggregation that the dot cluster alone obscures. Mopti's share drops from 0.54 to 0.31, while Ségou (just south) rises from 0.10 to 0.16. Figure 5b colors events by perpetrator year-by-year — state forces (pink) emerge alongside persistent non-state activity post-Wagner. Variance is 75% within countries, 25% between: the action is inside each country, not between.

Sahelian counter-insurgency operations have treated Fulani communities as a suspect category — operations target them regardless of JNIM membership, and targeting drives recruitment that retroactively confirms the category [16]. As JNIM consolidated Mopti, contestation shifted south to Ségou (mixed Fulani–Bambara) — where Kalyvas predicts indiscriminate violence in zones a regime cannot fully control [9]. The variance pattern confirms regime-driven change, not cross-country drift.

Finding 6 — Text analysis of ACLED notes (TF-IDF, VADER, LDA)

Figure 6 ranks words distinctive of post-Wagner vs pre-Wagner Mali notes (TF-IDF, a word-frequency score). Post-Wagner is dominated by Wagner-era actors and operations — wagner, fama (Mali's army), sahel, mercenaries, patrol. Pre-Wagner features jihadist names ('katiba', 'isgs') and Dogon-massacre terms ('dogon', 'villagers'). Figure 7 tracks monthly note tone per country (VADER sentiment, −1 negative to +1 positive) with orange horizontal pre/post-Wagner mean lines per panel labelled with their numerical values. All three lines stay around −0.5. Mali's pre-post mean shift is just 0.02 (significant at p = 0.025 on n = 4,090+6,939 but tiny in magnitude). Figure 8 plots eight word-clusters (LDA) pre- vs post-Wagner: 'mopti-fulani-militiamen' more than halves while 'fama-wagner-village' more than doubles — independently surfacing F5's Mopti-Ségou shift.

Three text methods converge: post-Wagner notes name Wagner-era actors, not jihadist-faction descriptors. State forces now operate alongside Wagner / Africa Corps personnel, so mercenary terms dominate ACLED's vocabulary [2]. A Random Forest classifier separately reaches 94% accuracy distinguishing civilian-targeted from non-civilian-targeted notes (Logistic Regression: 93%), confirming ACLED's text field carries machine-detectable signal. The flat sentiment locks this in: tone did not change, only the actors named — ruling out a reporting-volume artifact [5].

Data Explorer

The Data Explorer extends the main findings. Figure 9's treemap shows post-Wagner Mali's battles becoming more lethal (red 'armed clash' block) and air/drone strikes more prominent. Figure 5b above animates the Ségou cluster's emergence as a step change after 2021-12-01 — a deployment pattern consistent with the broader post-coup Sahel literature.

Open the pyLDAvis interactive topic-explorer in a new tab (course-taught Session-2 idiom).

Discussion

Four alternatives merit engagement. Autonomous jihadist resurgence is hardest to rule out, but the perpetrator breakdown contradicts it: non-state killing barely changes while state and external forces rise sharply — wrong shape for a jihadist story. ACLED reporting bias [5] could inflate post-coup state-violence counts, but worsening press conditions more plausibly suppress reporting of state perpetration — biasing against our finding. Bargaining failure [17] explains state escalation but not the unchanged insurgent line. Ethnic mobilization explains target selection at best, not timing — which tracks Russian arrival, not ethnic structure (cf. Fearon & Laitin's caution about ethnic-grievance accounts) [18].

Reverse causation is harder: Wagner may be symptom, not cause. Juntas may pick Russians for the freer hand. Niger's post-Wagner window shows no significant rise, confirming Mali drives the pooled IRR. The placebo's silence is our strongest evidence, not conclusive proof.

Conclusion

The AES case tests whether civilian killing can shift onto state and external partners while the insurgency stays constant. ACLED's record, broken down by perpetrator type, says yes — provisionally. Whether Burkina Faso's and Niger's shorter post-Wagner windows converge on Mali's pattern will settle whether Wagner is a cause or only a symptom. Until then, the data are a warning, not a verdict.

References

  1. [1] Powell, Jonathan M., and Clayton L. Thyne. 2011. "Global instances of coups from 1950 to 2010: A new dataset." Journal of Peace Research 48 (2): 249–259. https://doi.org/10.1177/0022343310397436
  2. [2] Marten, Kimberly. 2023. "Whither Wagner? The Consequences of Prigozhin's Mutiny and Demise." Survival 65 (5): 45–64. https://doi.org/10.1080/00396338.2023.2261245
  3. [3] McCullough, Aoife, and Adam Sandor. 2023. "Briefing: How a mutiny became a(nother) coup: The politics of counterinsurgency and international military partnerships in Niger." African Affairs 122 (489): 587–601. https://doi.org/10.1093/afraf/adad034
  4. [4] Raleigh, Clionadh, Andrew Linke, Håvard Hegre, and Joakim Karlsen. 2010. "Introducing ACLED: An Armed Conflict Location and Event Dataset." Journal of Peace Research 47 (5): 651–660. https://doi.org/10.1177/0022343310378914
  5. [5] Eck, Kristine. 2012. "In data we trust? A comparison of UCDP GED and ACLED conflict events datasets." Cooperation and Conflict 47 (1): 124–141. https://doi.org/10.1177/0010836711434463
  6. [6] Wilén, Nina. 2024. "Contagious coups in Africa? History of civil–military imbalance." African Affairs 123 (491): 243–261. https://doi.org/10.1093/afraf/adae011
  7. [7] Houle, Christian, Mark A. Kayser, and Jun Xiang. 2016. "Diffusion or Confusion? Clustered Shocks and the Conditional Diffusion of Democracy." International Organization 70 (4): 687–726. https://doi.org/10.1017/S002081831600028X
  8. [8] Stanton, Jessica A. 2016. Violence and Restraint in Civil War: Civilian Targeting in the Shadow of International Law. Cambridge: Cambridge University Press. https://www.cambridge.org/core/books/violence-and-restraint-in-civil-war/E12CE71566F30A88B86D5DD16C89A2EA
  9. [9] Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge: Cambridge University Press. https://www.cambridge.org/core/books/logic-of-violence-in-civil-war/F23B73D7C5301D7BAAC5DA8DEC68352F
  10. [10] Hultman, Lisa. 2007. "Battle Losses and Rebel Violence: Raising the Costs for Fighting." Terrorism and Political Violence 19 (2): 205–222. https://doi.org/10.1080/09546550701246866
  11. [11] Lyall, Jason, and Isaiah Wilson III. 2009. "Rage Against the Machines: Explaining Outcomes in Counterinsurgency Wars." International Organization 63 (1): 67–106. https://doi.org/10.1017/S0020818309090031
  12. [12] Fjelde, Hanne, and Lisa Hultman. 2014. "Weakening the Enemy: A Disaggregated Study of Violence against Civilians in Africa." Journal of Conflict Resolution 58 (7): 1230–1257. https://doi.org/10.1177/0022002713492648
  13. [13] Roessler, Philip. 2011. "The Enemy Within: Personal Rule, Coups, and Civil War in Africa." World Politics 63 (2): 300–346. https://www.cambridge.org/core/journals/world-politics/article/abs/enemy-within/B0E6C58E8FAC5D0BBD27BB22B4D0F0FF
  14. [14] Roessler, Philip. 2016. Ethnic Politics and State Power in Africa: The Logic of the Coup–Civil War Trap. Cambridge: Cambridge University Press. https://www.cambridge.org/core/books/ethnic-politics-and-state-power-in-africa/EDB6C9E47C42D58B3B53F4FC1F5DE0FC
  15. [15] Singh, Naunihal. 2014. Seizing Power: The Strategic Logic of Military Coups. Baltimore: Johns Hopkins University Press. https://www.press.jhu.edu/books/title/10802/seizing-power
  16. [16] Charbonneau, Bruno. 2021. "Counter-insurgency governance in the Sahel." International Affairs 97 (6): 1805–1823. https://doi.org/10.1093/ia/iiab182
  17. [17] Fearon, James D. 1995. "Rationalist Explanations for War." International Organization 49 (3): 379–414. https://www.cambridge.org/core/journals/international-organization/article/abs/rationalist-explanations-for-war/E66BCE3D6BBE2DDA9F4DDC7F03C20E69
  18. [18] Fearon, James D., and David D. Laitin. 2003. "Ethnicity, Insurgency, and Civil War." American Political Science Review 97 (1): 75–90. https://www.cambridge.org/core/journals/american-political-science-review/article/abs/ethnicity-insurgency-and-civil-war/B1D5D0E03488E3DDA48D2EE6CDE7DBE7