Verification of Peace Accords and Military Expenditures in Post-Conflict Societies

ABSTRACT

Why is it that some governments ending a civil war in a negotiated settlement succeed in reducing military spending while others fail? Civil wars ending in peace agreements result in relatively low military expenditures; however, not all governments succeed in the reduction. I argue that implementing a third-party verification mechanism of peace accords helps reduce military spending in post-conflict societies because the verification mechanism facilitates the peace accord implementation by enabling reciprocal implementation and by increasing the cost of noncompliance through active information flow. Implementation of peace agreements reduces threats posed by both former and outside rebel groups. This makes the government decrease the military expenditure allocated to appease internal security threats. I tested this argument using 32 civil wars with a comprehensive peace agreement between 1992 and 2011. The results indicate that initiating a verification mechanism leads to lower military spending.

KEYWORDS:

• Military expenditure
• civil war
• verification mechanism
• peace agreement
• negotiated settlement

JEL CODES:

• D74
• H56
• H53

Acknowledgement

I would like to thank Joakim Kreutz, Theodora-Ismene Gizelis, and participants at the ISA 2022 Annual Convention for their helpful comments and suggestions on this project. I also greatly thank the editor and anonymous reviewers for excellent suggestions and comments.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Notes

1. The unit of analysis is conflict episode. Information on conflict episodes and their terminations were taken from UCDP Conflict Termination Dataset version 2-2015 (Kreutz 2010).

2. The unit of analysis is conflict episode. Information on conflict episodes and their terminations were taken from UCDP Conflict Termination Dataset version 2-2015 (Kreutz 2010).

3. Refworld’s ‘Freedom in the World 2009 – Cote d’Ivoire;’ accessed on 31 March 2021. Available at: https://www.refworld.org/docid/4a6452c11e.html.

4. Mozambique – ONUMOZ Background; accessed on 2 September 2021. https://peacekeeping.un.org/sites/default/files/past/onumozFT.htm.

5. Mozambique – ONUMOZ Background; accessed on 2 September 2021. https://peacekeeping.un.org/sites/default/files/past/onumozFT.htm.

6. United Nations Security Council. ‘Report of the Secretary-General on the United Nations Operation in Mozambique.’ S/1994/89.

7. Although there were originally 34 peace agreement cases, 2 cases were dropped as information on Polity2, which I use as a control variable, was not available for Bosnia and Lebanon.

8. In PAM_ID, some peace agreements exited the sample before the 10^th year of implementation, including the Ouagadougou Political Agreement signed on 4 March 2007 in the Ivory Coast; Abidjan Peace Agreement, on 30 November 1996 in Sierra Leone; and Sudan Comprehensive Peace Agreement, on 9 January 2005 in Sudan.

9. These observations are before those where other missing variables are removed.

10. As I will mention later, I include year dummies for one model and an interaction between the main independent variable and UN personnel variable. VIF scores for year dummies and an interaction term were higher than 3.

11. I include a lagged dependent variable only in one model because including a lagged dependent variable could produce biased estimates (Keele and Kelly 2006). Indeed, many studies do not include a lagged dependent variable (Albalate, Bel, and Elias 2012; Collier and Hoeffler 2007; Phillips 2014).

12. In calculating the marginal effect, mean (median) values were used for continuous (dummy) variables.

13. The information on military personnel was taken from Correlates of War National Material Capabilities (v6.0) Dataset (Singer 1987; Singer, Bremer, and Stuckey 1972).

14. As the main results in Table 1 had greater Adjusted R² score, I present a model with population as the main result in Table 1.

15. As the main results in Table 1 had greater Adjusted R² score, I present a model without log transformation as the main result in Table 1.

16. I assign the same model numbers as in Table 1.

17. The values of the other continuous variables are held at their mean values, while median values were plugged in for dichotomous variables.

18. I calculate the number of years each conflict lasted using the information on conflict start date listed in the PAM_ID (Joshi, Quinn, and Regan 2015a).

19. The cumulative number of battle-related deaths was calculated using the UCDP Battle-Related Deaths Dataset (Pettersson and Öberg 2020). Because this dataset only covers information since 1989, for conflicts that started before 1989, cumulative deaths were calculated since 1989.

20. This variable takes the value 1 if the Polity 2 score from the Polity V Dataset (Marshall and Robert Gurr 2018) is greater than five and 0 otherwise.

21. Using the threshold for mean difference to 0.1, all covariates were balanced.

22. As mentioned earlier, because information on battle-related deaths was only available since 1989, I controlled for this variable in the model.