From Words to Deeds: The Implementation of Power-Sharing Pacts in Peace Accords

Abstract

Is the implementation of power-sharing provisions the key to durable peace? This study analyzes whether the implementation of political, military, and territorial power-sharing provisions contributes to peace. We introduce a new dataset (IMPACT), which contains unique information on the implementation of power-sharing pacts in peace accords in the post–Cold War period. Based on the logic of costly signaling, we demonstrate that when the parties engage in costly concessions by implementing military and territorial provisions, peace is more likely to prevail. In contrast, the implementation of political pacts is less costly, and hence does not increase the prospects of peace.

Keywords:

• power sharing
• peace agreements
• civil war
• conflict resolution

Acknowledgment

The research has been conducted within the project Dilemmas of Democratization in Post-War Societies, headed by Anna K. Jarstad. We are grateful for the funding provided by the Swedish Research Council. The authors wish to thank Mats Hammarström, Thomas Ohlson, Ralph Sundberg, participants of the research seminar at the Department of Peace and Conflict Research, Uppsala University, and three anonymous reviewers for valuable comments on earlier versions of this article. The authors have contributed equally to this article.

Notes

¹ A political pact regulates the power in the central government, a military pact concerns the integration of the military forces, and a territorial pact provides for some form of decentralization. The pacts are defined in more detail in the section covering the research design

² For a recent related study, see Hartzell (2006)

³ Mukherjee finds that offers of political power sharing make war less likely to resume if they are made following military victories (in contrast to military stalemates)

⁴ It should be noted that Walter (2002b) does take into account the issue of implementation as she examines the effect of these pacts on the successful implementation of settlements, but as we will explain below, there are limitations to her approach

⁵ This study is also presented in Hartzell and Hoddie (2007). In both these studies they also take into account the costs associated with the initiation of negotiations and the signing of a settlement

⁶ Related to this, Werner and Yuen (2005) suggest that peace can be fragile if agreements are imposed, as such deals may not reflect the conditions on the ground

⁷ There is a rich literature on different types of territorial solutions (e.g., Coakley, 2001; McGarry, 2007; McGarry & O'Leary, 2004; Tir, 2003; Weller & Wolff, 2005). Note that our study does not include any cases of partition (Kaufmann, 1996), but other territorial solutions such as autonomies and decentralization where the state borders remain intact

⁸ If not otherwise mentioned, the data used in this article comes from the UCDP (2006). The dataset TOPAD includes information on political, military, and territorial pacts in all peace agreements signed in the period 1989–2004 (Nilsson, Svensson, & Sundberg, 2006). IMPACT is based on this information and additional data has been collected, for instance, concerning if and when these pacts are implemented

⁹ It should be noted that in this study we do not incorporate agreements that merely outline a process for how to resolve the incompatibility, or agreements that are to be confirmed in a forthcoming agreement. A list of the peace agreements that meet our criteria is available in Nilsson (2006). To this list of peace agreements, covering the period 1989–2003, we have added the settlement signed in Ivory Coast in 2004, as we have extended the time period by one year

¹⁰ The peace agreements can also be censored prior to the end of the observation period. In some of the conflicts, the warring parties may sign a peace agreement and then reach another peace agreement without having resorted to violence. In an effort not to overestimate the effect of the duration of peace, the dependent variable is in such cases censored at the time of the signing of another agreement. In other words, if the same signatories sign another peace agreement, and stick to peace, the subsequent agreement is seen as replacing the previous agreement. To exemplify, in the conflict in Djibouti, the FRUD faction signed a peace agreement in 2000, remained at peace, and then reached another peace agreement in 2001. In this case the parties are considered to be at peace for one year after the first agreement, and at peace for four years after the second agreement. Without censoring the cases at the time of the signing of the subsequent agreement, peace would be coded as having prevailed for five and four years, respectively

¹¹ In order to code the implementation of the pacts, a wide range of sources has been consulted, including information from Keesing's Record of World Events, Reuters, Amnesty, IRIN and UCDP. For a more detailed account of the material used, see the codebook (Jarstad, Nilsson, & Sundberg, 2006)

¹² Any time period is admittedly a somewhat arbitrary choice. Another approach, instead of using a five-year implementation period, could have been to examine whether the pacts were put into place in line with the time frame agreed upon in the peace agreement. But as pointed out by Hoddie and Hartzell (2003) in their study of the implementation of military power-sharing provisions, not all agreements provide details concerning when the provisions should be in place, which creates problems of comparability across cases, and it is therefore better to evaluate the implementation according to the same time frame

¹³ In contrast to Walter (2002a), in this article, we do not consider a military provision which merely stipulates that the parties are to retain control over their own forces to qualify as a military pact, as such an arrangement does not concern the sharing of power

¹⁴ We use the Polity2 variable where values of transitions have been converted to conventional polity scores. The scale ranges from –10 to +10, and Democracy is coded ‘1’ if the score is from 6 to 10, and Autocracy is coded ‘1’ if it has a value between –6 and –10. The regimes that have values in between these two ranges are considered to be semi-democracies

¹⁵ It is worth noting that if one excludes the two territorial pacts where the parties are granted control over the territory they have de facto control over and where full implementations thereby is achieved immediately upon the signing of the agreement, the average duration of implementation is increased from 18 to 21 months

∗Statistically significant at p < 0.1

∗∗Statistically significant at p < 0.05

∗∗∗Statistically significant at p < 0.01. Two-tailed tests are used. Note that the variables on implementation refer to pacts that are implemented in full

¹⁶ The Cox proportional hazards model is based on the assumption that the risk is proportional over time. In the literature on duration models, it is recommended that this assumption is tested (Box-Steffensmeier & Jones, 2004; Box-Steffensmeier, Reiter, & Zorn, 2003). Hence, to examine whether this assumption is violated, we have carried out tests based on the Schoenfeld residuals, but neither the global nor the covariate tests show any signs of violations

¹⁷ At the same time, it should be noted that the effect is in the expected direction, and hence, while political pacts are not shown to significantly decrease the risk of the signatories engaging in violence, this type of provision is not found to have a detrimental impact on the prospects for peace

¹⁸ This result is somewhat sensitive to the controls included, and in some of the alternative specifications the effect of Military Pact is statistically significant at p < 0.05

¹⁹ While we have coded whether a pact is partially implemented or not, we do not have information on the date when the fully implemented pacts reaches partial implementation. Therefore, at this stage we cannot include this aspect in our statistical analysis, without the risk of biasing the results. We still feel confident in these results as a close inspection of the data clearly indicate that the territorial pacts that are reaching only partial implementation, rarely are followed by further armed conflict

²⁰ These results are robust to a number of alternative specifications and tests. For instance, the results remain virtually the same when clustering the robust standard errors on country instead of conflict. Furthermore, instead of using a lagged value of the implementation, we also measured the implementation in the same year as the dependent variable, but this does not change the main findings

²¹ As can be seen in Table 1, the hazard ratios are close to zero. More precisely, the hazard ratio for the variable Military Implementation is 3.29e-18, and the corresponding value for Territorial Implementation is 5.96e-18

²² We also conducted some alternative models. For instance, we control for whether the opposition was ethnically mobilized or not, and the number of warring parties in the conflict. The results remain virtually the same. When controlling for the economic development measured as the log of GDP, the variable Military Pact has a hazard ratio below one and is significant at p < 0.10, but this result is not significant in the models where we include the variables capturing the implementation of pacts. Furthermore, we also estimate a model where we replace the variables Democracy and Autocracy with a continuous measure Polity, (ranging from 10 to +10), which is used together with the square of this variable to account for a possible curvelinear relationship of the previous political regime. While the results are basically the same, it should be noted that when using this alternative measure in Model 3, Military Pact is then statistically significant at the 0.10 level, indicating a decreased risk of peace failing

²³ Military pacts are more common following conflicts over government, 45%, whereas only 22% of the territorial conflicts contained such a provision. For a closer discussion on this, see Jarstad and Sundberg (2007)

²⁴ UN peacekeeping forces and non–UN peacekeeping forces do not show any direct effects on the risk of peace failing in these models

²⁵ Of the 46 peace agreements with only one pact, there are 17 agreements containing a political pact, 9 with a military pact, and 20 with a territorial pact

²⁶ The effect of implemented pacts in combination with other factors is assessed by introducing an interaction term (e.g., Political Implementation∗Territorial Implementation)

²⁷ Measured as the number of power sharing pacts (political, military, and territorial) the agreement entails. Hence, this variable ranges from 0 to 3

²⁸ There are some recent attempts to address selection problems in duration models, such as the duration model with selection by Boehmke and colleagues (2006). However, there are limitations to their approach, for example, such a selection model only allows for time-varying covariates in the first stage of the model (not the second). To use a model with time-varying covariates is very important for our study as our main independent variables vary over time, namely the implementation of different types of pacts. Without such an approach we risk losing valuable information about the implementation