Political Parties and the State in Civil War

@ Emory IR Journal Club

Varun Karekurve-Ramachandra

USC

Aidan Milliff

FSU

Drew Stommes

NYU

Elections during civil wars

• Most lit on elections and war focuses on onset, termination, post-conflict
• However, elections often a political “front” for incumbents during conflict
• Do elections shape ongoing violence? If so, how?

Existing theory is fairly ambiguous

There is a lot of research on democracy and civil war

Most of it focuses on the start and the end, but:

1. There is a lot of democratic exercise that goes on during war
2. In separatist conflicts especially, democratic contestation seems to be a way that national parties try to re-establish dominance (Baluchistan, Turkey, etc.)

So: How does this contestation (which is often between incumbents and the moderate separatists) shape violence?

Actually surprisingly hard to make a guess

Theory 1: How Do National Party Legislators Matter?

Reps of the “center” could increase violence:

• Seen as symbolic outsiders
• Could support aggressive policies, increase “effort”

Reps of the “center” could decrease violence:

• Could attract more resource commitment
• Could make COIN more effective

Most tests of this focus on one of the following situations: 1) Not a civil war (Nellis and Siddiqui, Nellis Weaver and Rosenzweig), or not a separatist civil war (Milliff and Stommes) 2) Not a legislator (Dell, Zukerman Daly, etc). Could go either way.

Theory 2: ICTs and Counterinsurgency

A manual telephone exchange

Technology changes the COIN landscape

(Shapiro and Weidmann 2015; Shapiro and Siegel 2015; Berman and Matanock 2015; Pierskalla and Hollenbach 2013; Dafoe and Lyall 2015; Walter 2017)

Where ICT infrastructure is good, effects should be larger because ICT:

• Improves tip sharing with MLAs and security forces

• Eases COIN coordination

• Allows monitoring of militants

Punjab

• High-intensity separatist conflict (1970s–1990s)

• (well) over 10k deaths between state, insurgents, civilians

• Three state-level elections contested during conflict; largely two-party contests

  • INC (National)
  • SAD (Regional, ethnic)

Annual state security force fatalities in Punjab Crisis (Staniland and Stommes, 2019)

We use data from a long-running separatist conflict in India to adjudicate.

• Punjab crisis is an ethnic separatist insurgency that killed at the very least 10k, mostly in the 1980s and 1990s before it was quashed by a totally brutal COIN campaign that has become a “template” for subsequent Indian counterinsurgency in Kashmir for example

• Elections throughout the conflict, though the “spoils” are irregular.

• Contestation between a major national party (national gov for all but 18 months of conflict) and a conservative regional party that is kind of hte “moderate” counterpoint to the militancy

• We want to know whether having a legislator from one or the other affects conflict, as measured by data that paul and drew collected

Measuring the Effects of Legislators

RD Design

• Winner of “close” elections as if random
• Intuition is simple, estimation is not so simple in practice, see Stommes et al. (2023)
• We compare: insurgent violence in ACs narrowly won by INC vs. ACs narrowly lost

Included Constituencies

RD allows us to identify the effect of having a legislator from one party vs. another on insurgent violence against security forces.

looks at “close” elections, which means it leaves out some constituencies. The bandwidth includes all those in dark blue.

Violence in Punjab is geographically concentrated; important to show that we’re not measuring the effect of legislators on violence int he least violent spots

Main Results

P(Any Fatality) ~ INC Win

Fatality Count ~ INC win

INC Legislators Reduce Insurgent Violence Production

.4 is the conservative per-legislator fatality estimate. Baseline is .66

Left plot: any fatalities Right plot: count Effect of having an INC legislator vs other in each year after election Not immediate, but durable Effects probably blow up because of presidents rule stable across all the junk like differnet kernels, covs vs no, bandwitdhts

What Doesn’t Drive the Main Results?

INC effect not driven by:

• Hindu- vs. Sikh-majority ACs
• Higher vs. Lower HDI
• Urban vs. Rural ACs
• Rebel operation timing
• Incumbents vs. new MLAs
• Particular elections
• Particular districts

Block-Jackknife Plot: Dropping each district to check for “importance” to overall effect.

Lots of checks to understand this result these are not for causal ID unless you think one of them jointly determines an INC win (conditional on close elex) and violence

Three Clues About The INC Effect

1. Violence reductions go with decrease in indiscriminate “sweeps”/“nakas”

   • Substitution toward selective violence?

2. Violence reductions benefit both state and national security orgs

   • National party MLAs using de facto authority

3. Violence reductions largest in areas with good ICT Infrastructure

   • Could help with info gathering, COIN coordination, and direct surveillance

1. Tip sharing a la Shapiro and Weidmann (2015)
2. COIN coordination improved
3. Militants unaware of surveillance potential
4. Augments info sharing via “patron-client relations” (Chima, 2014) that INC re-activated

Discussion

• National party politicians significantly decrease violence in a separatist conflict
• Mechanism analysis points to:
  • Classic information \(\rightarrow\) selective violence pathway
  • Better coordination/influence over security bureaucracy
  • Interesting ICT Pathway
• If this is right, selective violence \(\neq\) “friendly” COIN

Thanks!

amilliff@fsu.edu

aidanmilliff.com