Networks and Dynamics of Violent Political Mobilization
Historical Evidence from Spain
Giacomo Lemoli and Sergi Martínez
sergi.martinez@gess.ethz.ch | sergimartinez.github.io
Motivation
We reached the highest number of state-based conflicts since 1946 (PRIO, 2024).
What leads volunteers to take arms?
What drives conscripts’ compliance?
The composition of soliders changes through conflicts:
Volunteers and (reservists) compliers differ in their motivation (e.g., Eck 2014)
Our contribution: Context and network embeddedness matter differently in different contexts
Literature and theoretical expectations
Intrinsic Motivations
Extrinsic incentives: Elite-Led & Coercive
Gap: Existing scholarship treats motivations as constant, but these drivers of participation change as conflict evolves
Phase 1, conflict onset: Intrinsic motivations dominate
Early mobilizers (hardliners) via identity and ideological networks (Staniland 2012; Sanı́n and Wood 2014; Abramson and Qiu 2024; Kalmoe 2020)
Phase 2, conflict escalation: Extrinsic motivations rise
As territorial control shifts and armies need to fill the tanks through forced recruitment (Eck 2014), survival and fear of retribution displace ideological drives (Kalyvas 2006; Rozenas, Talibova, and Zhukov 2023)
The Spanish Civil War in Navarre
1936-1939: Civil War
Failed fascist coup that divided the country for 39 months
Rebels, Francoists vs. Loyalists contenders of the Republic
Both sides killed and displaced civilians using indiscriminate and selective techniques
About 500,000-800,000 deaths
known as the “dress rehearsal for WWII”
Conservative and catholic, but diverse in terms of social fabric, identity, and industrialization (Caspitegui 2005)
July 18, 1936: Franco staged the coup.
In 10 days, elect. and paramilitary networks joined 10K volunteers in Pamplona.
Posterior compulsory conscription
August ’36, stable territorial control.
Both sides called for conscripts.
Rebels: 13 call ups on 12 cohorts (1929-1941).
Municipality-level evidence
First-movers - Electoral networks (Turnout ’36 and Right-wing vote share ’36) → voluntary enlistment
Volunteers vs. conscripted
% Right-wing ’36 on voluntary enlistment
Individual-level evidence
34’ voting-age census, n ≈ 160,000, 1/2 males (80K over 23yo)
File of Navarrese fighters, n ≈ 12K>25 y/o in 1936 (71% merged) (App B)
One tie = Interfamily marriage
2 measures from social networks: influence and dependence
Eigenvector centrality: Relative centrality in the network → Power (Correlated with higher class, + Carlist surnames) (Bandiera, Larreguy, and Mangonnet 2026)
Minimum path distance (MPD): Relative proximity to the most central node → Reachability (threat) (Atwell and Nathan 2022)
Implications
Intrinsic incentives (social and ideological incentives) → volunteers
Extrinsic incentives (fear) → compliance joining later, when called-up and possibly reported
\[ \Pr(\text{Volunteer}_i) = \alpha + \beta_1 \text{Age}_i + \beta_2 \text{Age}_i^2 + \beta_3 \text{Cen}_{f(i)} + \beta_4 \text{MPD}_{f(i)} + \beta_5 \ln(\text{Pop}_m) + \beta_6 \text{Carl}_{m,36} + \beta_7 \text{Right%}_{m,36} + \varepsilon_{m,i} \]
| Joining | voluntarily | |
|---|---|---|
| Fam. Centrality | 0.023** (0.010) | 0.004 (0.008) |
| Fam. Proximity | -0.045*** (0.013) | -0.016 (0.012) |
| (log) Population | -0.006*** (0.002) | |
| Carlist club (pre-war) | 0.012*** (0.004) | |
| Right % '36 | 0.046*** (0.010) | |
| Num.Obs. | 73203 | 73029 |
\[ \Pr(\text{Late joiner}_i) = \alpha + \beta_1 \ln(\text{Pop}_m) + \beta_2 \text{Carl}_m + \beta_{3} \text{Right \%}_{m,36} + \beta_4 \text{Age}_i + \beta_5 \text{Age}_i^2 + \]
\[ \beta_6 \text{Conscripted}_i + \beta_7 \text{Centrality}_{f(i)} + \beta_8 \text{MPD}_{f(i)} + \beta_{9} (\text{Cen}_{f(i)} \times \text{Cons}_i) + \beta_{10} (\text{MPD}_{f(i)} \times \text{Cons}_i) + \varepsilon_{m,i} \]
| Joining | late | |
|---|---|---|
| (log) Population | -0.005 (0.004) | -0.005 (0.004) |
| Carlist club | 0.011 (0.007) | 0.011 (0.007) |
| Right % '36 | -0.008 (0.023) | -0.009 (0.023) |
| Fam. Centrality | -0.026* (0.015) | -0.018** (0.008) |
| Fam. Proximity | 0.066*** (0.021) | 0.020* (0.011) |
| Conscripted cohort | 0.085*** (0.006) | 0.027 (0.018) |
| Fam. Centrality × Conscripted | -0.043 (0.043) | |
| Fam. Proximity × Conscripted | 0.191*** (0.063) | |
| Num.Obs. | 71342 | 71342 |
Compliance with the draft to avoid imprisonment, get rid of stigma and police persecution (Leira-Castiñeira 2020).
As the way to switch sides (Rego 2014; Leira-Castiñeira and Domı́nguez-Almansa 2018).
Conclusion
The interwar period backsliding was endogenous to the pre-war scenario, separately, in political and social terms:
Pre-war electoral networks → volunteer mobilization
Threat of retribution → compliance
Literature
Roots of democratic backsliding and conflict mobilization (e.g., Kalmoe 2020; Dippel and Heblich 2021), joining lit. on interwar fascism (e.g., Satyanath, Voigtländer, and Voth 2017).
Role of networks (Bai, Jia, and Yang 2023; Cruz, Labonne, and Querubin 2020; Atwell and Nathan 2022; Bandiera, Larreguy, and Mangonnet 2026), now on conflict mobilization
Violent political mobilization (Humphreys and Weinstein 2008; A. M. Arjona and Kalyvas 2012; Eck 2014) and conflict processes (Kalyvas 2006; Gates 2017).
Thanks for listening!
References
Appendix
If electoral-patronage networks mobilized right-wing hardliners as first movers:
Right % ’36 to predict the share of first movers.
Turnout ’36 to predict the share of first movers.
\(ShareVolunteers_{m} = \beta_1 ShareRight_{1936,m} + \beta_2 Turnout_{1936,m} + X_{cp} \gamma + \varepsilon_{m}\)
| Volunteers July'36 | Placebo w/post-Aug '36 | |
|---|---|---|
| * p < 0.1, ** p < 0.05, *** p < 0.01 | ||
| % Right 1936 | 0.132*** | -0.120** |
| (0.033) | (0.050) | |
| % Turnout 1936 | 0.097** | -0.000 |
| (0.044) | (0.042) | |
| Num.Obs. | 255 | 255 |
| R2 | 0.500 | 0.195 |
Insurgent record
|
Census record
|
|||||||
|---|---|---|---|---|---|---|---|---|
| Municipality | S1 (ins.) | S2 (ins.) | Name (ins.) | Age (ins.) | S1 (cen.) | S2 (cen.) | Name (cen.) | Age (cen.) |
| Lumbier | Caminos | Irabarren | Javier | 30 | Caminos | Iribarren | Javier | 27 |
| Garralda | Larraz | Sánchez | Ambrosio | 29 | Larruiz | Sánchez | Ambrosio | 24 |
| Yerri | Unanua | Ilzarbe | Teofanes | 26 | Unanua | Ilzarbe | Teogenes | 23 |
| Sanguesa | Galarza | Maestu | Javier | 30 | Galarza | Maestre | Javier | 25 |
| Falces | Mendoza | Antón | Gregorio | 26 | Mendoza | Antón | Gregorio | 23 |
| Araiz | Goicoechea | Loiti | Fermin | 27 | Goicoechea | Loidi | Fermin | 25 |
| Arruazu | Arratibel | Bergara | Pedro | 28 | Arratibel | Vergara | Pedro | 25 |
| Villava | Echepare | Ibañez | Fermin | 29 | Echapare | Ibáñez | Fermin | 27 |
| Lumbier | Machinandiarena | Torrea | Jose | 27 | Machinandiarena | Torrero | Jose | 25 |
| Tafalla | Urian | Reta | Juan | 27 | Uríen | Reta | Juan | 24 |
| Aranguren | Zabalza | Giménez | Tomas | 26 | Zabalza | Jiménez | Tomas | 23 |
| Estella | Aramendía | Garialde | Joaquin | 44 | Aramendía | Garlalde | Joaquin | 40 |
| Peralta | Lorea | García | Julian | 43 | Lorca | García | Julian | 40 |
| Ulzama | Larrainzar | Vallanueva | Juan | 29 | Larrainzar | Villanueva | Juan | 23 |
| Estella | Nuin | Mendoza | Herminio | 29 | Nuin | Mendoza | Herminio | 23 |
Matches are approximate: surnames and given names may differ slightly across sources due to transcription variation, dialectal spelling, and OCR error. Age gap ≤ 10 years. Non-within-municipality matches < 3.5%.