Weak Ties, Strong Impact: An Analysis of the Rwandan Genocide
By Gemma Hauser
Edited by Sofia Gobin and Zoe Laxton, Peer Reviewed by Alex Bagdade
- Summary of argument
This paper argues that widespread civilian participation in the 1994 Rwandan Genocide was not solely the product of top-down elite control, external pressures, or psychosocial conformity, but was significantly enabled through weak social ties that diffused genocidal violence across communities. Drawing from Granovetter’s theory of weak ties, which conceptualizes these ties as low-frequency, low-intensity social relationships that serve as bridges between otherwise disconnected groups, I reframe genocide as a decentralized, networked event rather than a strictly hierarchical one.
While strong ties may have catalyzed initial mobilization through elite rhetoric and militia organization, it was weak ties—connections between loosely affiliated individuals—that
facilitated the rapid diffusion of violence between social clusters. The paper specifically
theorizes that community meetings, such as Umuganda, and intermediary figures, such as local
political actors and militia recruiters, operated as brokers. These actors were structurally
positioned to channel genocidal messaging from central authorities into local settings, enabling
ordinary civilians to coordinate and participate in acts of violence.
This theory is empirically tested through a network analysis of actors and violent events
using the UCDP Georeferenced Event Dataset (UCDP-GED). The analysis focuses on
identifying whether actors with high betweenness centrality but low degree centrality—the
operational definition of weak tie brokers—are disproportionately associated with regions where
One-sided violence against civilians was more intense and widespread. This has implications for understanding how violence proliferates not simply through elite coercion, but via decentralized relational mechanisms.
- Research design
The central causal claim is that weak ties–measured by high betweenness and low degree centrality–enabled the horizontal diffusion of violence by linking otherwise unconnected social groups. Most conventional analysis, such as Hintjens (1999), argue that “the nature of the Rwandan state must be seen as absolutely central” (Hintjens 241), with genocide taking place under the aegis of the state, where “the main organizers were a northern elite, united by senior positions in the army and the top civil service” (Hintjens 249). These accounts emphasize external influences and domestic pressures, inadvertently overlooking how to explain the scale to which participation diffused across communities. This research addresses that gap by showing how weak tie brokers facilitated the horizontal spread of violence between otherwise unconnected social groups. conventional narratives of the genocide
- Causal Framework
Granovetter’s weak tie hypothesis posits that individuals connected by low-frequency,
low-intimacy relationships can serve as bridges across social clusters, allowing for the spread of
ideas and behaviours beyond the reach of tightly knit groups. In Rwanda, these brokers were not
national leaders, but local actors such as Interahamwe recruiters, political party organizers, or
heads of community meetings, who could influence civilians otherwise disconnected from
formal institutions. This diagram illustrates the mechanism through which weak tie brokers
facilitated the horizontal diffusion of genocidal violence, with light blue representing main flow
causal variables and light grey representing moderating factors.
- Dependent variable
The dependent variable, civilian participation in violence, is operationalized as the
density and frequency of one-sided violent events (violence against civilians) across regions.
This measure captures both the intensity and spatial distribution of violence.
- Independent variable
The independent variable is the presence of weak ties, operationalized as actors with high
betweenness centrality and relatively low degree centrality. They are expected to serve as
brokers, having fewer direct connections (degrees) but serving as key bridges in the network
(betweenness).
- Control variables
- Geographic region (to account for geographic variation)
- Population density
- Proximity to elite actors (e.g., national political leaders, major military groups)
- Number of Umuganda (community meetings)
These controls help isolate the effect of weak ties from structural and demographic
factors that may also influence violence levels.
- Justification for Network Approach
The causal question is relational: how did violence diffuse between groups? Network
analysis allows for the identification of intermediary actors who occupy structurally important
positions as brokers. My essay focuses on betweenness centrality (how often a node connects
others) and uses degree centrality descriptively to distinguish elites from weak ties.
The initial phases of my theory confused the use of centrality measures. This version
clarifies that betweenness is the theoretical centrepiece, while degree is used to refine actor classification and rule out elite dominance. A network approach is essential for revealing how
otherwise disconnected actors were bridged by community figures.
- Proposed inferential framework
Although formal statistical inference is not conducted, this paper uses an inferential logic
based on spatial comparison. If the theory holds:
- Regions with a higher density of weak tie brokers (high betweenness/low degree) will show higher civilian violence
- Elite-dominated regions will show a weaker association with civilian participation
- Description and visualization of the network
The empirical analysis focuses on the independent variable: the presence and structure of
weak tie brokers in the actor-event network. Using UCDP-GED data, a network was constructed
with nodes representing actors and edges representing co-involvement in recruitment,
coordination, or attacks. The dataset consists of georeferenced conflict events occurring within Rwanda between April and July 1994, corresponding to the approximately 100-day period. The graph includes actors such as the Government of Rwanda, Interahamwe, various militias, and civilian groups. Descriptive network statistics were used to calculate degree centrality (number of direct ties) and betweenness centrality (how often a node lies on the shortest path between others). Actors with high betweenness and low degree were flagged as weak tie brokers. These actors are interpreted as community-level intermediaries who played a key role in connecting otherwise disconnected actors.
Figure A1 illustrates the overall structure of actor interactions during the 1994 Rwandan
Genocide. While elite actors like the Government of Rwanda and Interahamwe occupy central
positions with multiple connections, the presence of smaller, directionally linked nodes
highlights a more distributed flow of coordination. Actors identified as weak tie brokers (i.e.,high betweenness and low degree) are labelled red, while all other nodes appear in blue. The
figure’s visual pattern aligns with my theory: violence diffused horizontally through
intermediaries embedded in community-level networks rather than exclusively directed from
Kigali’s political core.
Tutsi
It is also beneficial to compare degree and betweenness scores for each actor, illustrating
that brokers tend to fall in the low-degree, high-betweenness quadrant, consistent with
Granovetter’s theory. These actors include provincial militia organizers and local civilian
figures. Figure A2 is a scatterplot where red-labelled actors appear in the low-degree but high-
betweenness zone, confirming their structural role as intermediaries linking otherwise unconnected actors. This supports and further illustrates the horizontal diffusion of violence as
enabled by weak ties.
More specifically, actors such as Tutsi, FDLR-RUD, and M23 are located on the
periphery of the network in terms of direct connections, yet occupy crucial bridge positions
between major factions like the Interahamwe, FDLR-FOCA, and Civilians. Their location is
representative of Granovetter’s argument that weak ties serve as conduits through which
information and behaviour traverse social clusters, enabling broad behavioural diffusion without
centralized control. At the same time, the Tutsi, Hutu, and other identity-based categories fall
along the zero mark in both degree and betweenness centrality, suggesting that they were not
central in organizing violence. This spatial absence from network brokerage roles shows how intermediaries, rather than identities alone, functioned as a channel through which genocidal
participation spread.
- Conclusion
The findings of this analysis suggest that the diffusion of genocidal violence during the
1994 Rwandan Genocide cannot be fully explained by elite command structures alone. Instead,
weak tie brokers–identified as actors with high betweenness and low degree centrality – played
a critical role in horizontally disseminating violence across communities. The network structure
revealed that decentralized relationship pathways, rather than strictly vertical hierarchies, were
essential in enabling mass civilian participation. These results reinforce the argument that weak
social ties, amplified through intermediary actors embedded in local settings, were a key
mechanism driving the widespread and decentralized nature of the genocide.
Works Cited
Bonnier, Eric, et al. “The dynamics of violence: Evidence from the Rwandan genocide.” Journal
of Development Economics, vol. 147, 2020, 102559.
Granovetter, Mark S. “The Strength of Weak Ties.” American Journal of Sociology, vol. 78, no.
6, 1973, pp. 1360–1380. Stanford University, snap.stanford.edu/class/cs224w-
readings/granovetter73weakties.pdf.
Hintjens, Helen M. “Explaining the 1994 Genocide in Rwanda.” The Journal of Modern African
Studies, vol. 37, no. 2, 1999, pp.241-286. JSTOR, https://www.jstor.org/stable/161847.
Lemarchand, René. “Managing Transition Anarchies: Rwanda, Burundi, and South Africa in
Comparative Perspective.” The Journal of Modern African Studies, vol. 32, no. 4, 1994,
pp. 581-604.
Moore, Barrington, Jr. Injustice: The Social Bases of Obedience and Revolt. Basic Books, 1978.
Nyseth Nzitatira, Hollie, Et al. “Analyzing Participation in the 1994 Genocide in Rwanda.”
Journal of Peace Research, vol. 60, no. 2, 2023, pp. 291-306. SAGE Publications,
https://doi.org/10.1177/00223433221075211.
O’Loughlin, John, Micheal D. Ward, Corey L. Lofdahl, Jordin S. Cohen, David S. Brown, Kristian S. Gleditsch, and Micheal Shin. “The Diffusion of Democracy, 1946-1994.”
Annals of the Association of American Geographers, vol. 88, no. 4, 1998, pp. 545-574.
Institute of Behavioral Science, University of Colorado Boulder,
ibs.colorado.edu/johno/pub/old_pubs/O%27Loughlin%20et%20%%20al%201998%20A
Nnals.pdf.
Palmer, Alan. “Colonial and Modern Genocide: Explanations and Categories.” Ethnic and Racial
Studies, vo. 21, no. 1, 1998, pp. 89-115.
Uppsala Conflict Data Program (UCDP). Rwanda: UCDP Conflict Encyclopedia. Uppsala
University, https://ucdp.uu.se/country/517. Accessed February 2025.
Appendix: R Script
# Load necessary libraries
library(igraph)
library(tidyverse)
library(ggrepel)
library(readr)
# Load the UCDP-GED data
data <- read_csv("~/Desktop/gedevents-2025-01-31.csv")
# Build edge list from actor columns
edges <- data %>%
select(source_actor = side_a, target_actor = side_b) %>%
filter(!is.na(source_actor), !is.na(target_actor)) %>%
distinct()
# Create a directed graph
g <- graph_from_data_frame(d = edges, directed = TRUE)
# Simplify graph (remove loops and multiple edges)
g <- igraph::simplify(g, remove.multiple = TRUE, remove.loops = TRUE)
# Calculate centrality measures
V(g)$degree <- degree(g, mode = "all")
V(g)$betweenness <- betweenness(g, directed = TRUE, normalized = TRUE)
# Identify weak tie brokers (high betweenness, low degree)
threshold_betweenness <- quantile(V(g)$betweenness, 0.75)
threshold_degree <- quantile(V(g)$degree, 0.25)
V(g)$weak_tie_broker <- ifelse(
V(g)$betweenness >= threshold_betweenness & V(g)$degree <= threshold_degree,
TRUE, FALSE
)
# Layout for network graph
layout1 <- layout_with_fr(g)
# Create data frame for centralities
centrality_df <- data.frame(
actor = V(g)$name,
degree = V(g)$degree,
betweenness = V(g)$betweenness,
weak_tie_broker = V(g)$weak_tie_broker
)
# Subgraph of weak tie brokers and their immediate neighbors
weak_nodes <- V(g)[V(g)$weak_tie_broker]
subg_nodes <- unique(c(weak_nodes, neighbors(g, weak_nodes, mode = "all")))
g_sub <- induced_subgraph(g, subg_nodes)
layout_sub <- layout_with_fr(g_sub)
Causal logic flowchart
> library(DiagrammeR)
>
> # Create causal flowchart using Times New Roman
> grViz("
+ digraph causal_flow {
+
+ graph [layout = dot, rankdir = TB]
+
+ node [shape = box, style = filled, fillcolor = lightblue, fontname = 'Times New Roman']
++ A [label = 'Weak Tie Brokers\n(Local Political Actors, Recruiters)']
+ B [label = 'Facilitation of Genocidal Messaging']
+ C1 [label = 'Exposure to Propaganda\n(Radio, Meetings)']
+ C2 [label = 'Social Peer Pressure\n(Community Norms)']
+ D [label = 'Coordination Across Isolated Groups']
+ E [label = 'Increased Civilian Participation\nin Violence']
+
+ # Moderator factors
+ M1 [label = 'Geographic Region\n(Moderator)', fillcolor = lightgrey]
+ M2 [label = 'Population Density\n(Moderator)', fillcolor = lightgrey]
+ M3 [label = 'Proximity to Elite Actors\n(Moderator)', fillcolor = lightgrey]
+
+ # Connect nodes
+ A -> B
+ B -> C1
+ B -> C2
+ C1 -> D
+ C2 -> D
+ D -> E
+
+ # Connect moderators
+ M1 -> B
+ M2 -> D
+ M3 -> B
+ }
+ ")
Figure A1: Directed Network of Actors in the Rwandan Genocide
# Plot full network
plot(
g,
layout = layout1,
vertex.label = V(g)$name,
vertex.label.cex = 0.8,
vertex.label.color = "black",
vertex.size = scale(V(g)$betweenness, center = FALSE, scale = TRUE) * 15,
vertex.color = ifelse(V(g)$weak_tie_broker, "tomato", "skyblue"),
edge.arrow.size = 0.4,
edge.color = "grey50",
main = "Figure A1: Directed Network of Actors in the Rwandan Genocide"
)
Figure A2: Scatterplot of Degree vs Betweenness Centrality
# Scatterplot with repelling labels
ggplot(centrality_df, aes(x = degree, y = betweenness)) +
geom_jitter(aes(color = weak_tie_broker), width = 0.2, height = 0.01, size = 3) +
geom_point(data = subset(centrality_df, weak_tie_broker == TRUE), shape = 21, fill = "tomato", size = 4, stroke = 1) +
geom_text_repel(
data = subset(centrality_df, weak_tie_broker == TRUE),
aes(label = actor),
size = 3.2,
point.padding = 0.25,
box.padding = 0.4,
max.overlaps = Inf,
segment.color = "grey60",
segment.size = 0.4,
min.segment.length = 0
) +
scale_color_manual(values = c("FALSE" = "skyblue", "TRUE" = "tomato")) +
labs(
title = "Figure A2: Centrality Comparison of Network Actors",
subtitle = "Weak tie brokers are highlighted in red",
x = "Degree Centrality (Connections)",y = "Betweenness Centrality (Broker Role)",
color = "Weak Tie Broker"
) +
theme_minimal(base_size = 12) +
theme(legend.position = "bottom")