Abstract
This article introduces the Peace Accords Matrix Implementation Dataset (PAM_ID). We present time-series data on the implementation of 51 provisions in 34 comprehensive peace agreements negotiated in civil wars since 1989. We follow the implementation process for up to ten years following the signing of each agreement. The data provide new insights into the types of provisions that are more or less likely to be implemented, how implementation processes unfold over time, how implementation processes relate to one another, and how implementation affects various post-accord outcomes. We outline our coding methodology and case selection, and examine descriptive statistics. We illustrate one potential use of the data by combining eight different provisions into a composite indicator of security sector reform (SSR). A survival analysis finds that implementing security sector reforms contributes to long-term conflict reduction not only between the parties to the accord but also between the government and other non-signatory groups in the same conflict.
Keywords
Introduction
Research on the implementation of intrastate peace agreements and the impact of implementation on post-accord conflict dynamics is a field that may be best described as theory-rich but data-poor. Many of the prominent arguments in the literature invoking implementation processes predated sufficient data to fully test their arguments. Walter (1997: 335), for example, frames credible commitment theory as aiming to provide ‘the conditions under which negotiated solutions, once reached, can be implemented’. Hartzell & Hoddie (2003: 321) argue that cumulative power-sharing has a strong peace-promoting effect, while also noting that peace is unlikely in low implementation settings: ‘Failure to carry out a particular aspect of a peace agreement may occasionally occur even in those instances in which the peace itself proves durable.’
Mattes & Savun (2010: 9) argue that verification mechanisms help conflict actors overcome commitment problems and information asymmetries through the reporting of credible information in the post-accord setting, but acknowledge that such devices must ‘be implemented in order to take effect’. It is not only the verification mechanism itself that relies on being implemented. Verification, the act of monitoring and reporting on levels of compliance, can promote either peace or conflict depending on what is being reported. Where the conflict actors are complying with the implementation of the peace agreement and the verification body is accurately sending this information back to the conflict actors, verification can have a stabilizing and peace-promoting effect. However, where implementation is not taking place, and the actors are not complying with the terms of the agreement, the reporting of this information could constitute a possible reason for returning to war.
Like most types of reform, implementation is likely to be most resisted where it is most needed. Hence, attempts to evaluate the effects of particular provisions without sufficient information on implementation can lead to the making of type I and type II errors. This stems from the fact that accord content tends to be problem-driven and the same processes that motivate conflict actors to negotiate certain content will motivate other actors who may have greater control over the actual implementation process. Stated differently, those policy sectors that are most likely to be targeted with reforms will often be controlled by the very actors who have the most to lose from the reforms being implemented. With respect to interpreting research findings, this means that a provision that fails in one setting could have been more successful in a less challenging setting, and vice versa. These concerns have led some analysts to conclude that any ‘policy debate which ignores implementation is vacuous’ (Bird, 2004: 166). The qualitative nature of most of the scholarship on implementation, though rich, makes it difficult to systematically control for the large number of confounding variables that surround the implementation process (e.g. Bekoe, 2003, 2008; Boltjes, 2007; DeRouen et al., 2010; Stedman, Rothchild & Cousens, 2002).
To assist in closing the gap between theory and evidence regarding peace agreement implementation, we introduce new data from a project on the implementation of 51 types of provisions found in comprehensive peace agreements (CPAs) signed between 1989 and 2012 (Joshi & Darby, 2013). In the next section, we discuss how our dataset differs from existing sources. We then cover the methodology used to generate the data and present some descriptive statistics. Lastly, we give a brief example of how the data can be used by creating a composite indicator of eight security sector reform (SSR) provisions.
Existing implementation data
There are two existing data sources that include information about the content of peace accords and two existing data sources that provide information on the implementation of one or more types of provisions found within peace accords. Högbladh (2012) and Joshi & Darby (2013) provide data on the types of provisions found within peace accords, but do not provide quantitative data on their implementation. Current sources of implementation data cover three types of power-sharing provisions found in peace accords. Hoddie & Hartzell’s (2003) data on Sharing and Dividing Military Power was the first attempt to collect implementation data on a particular type of provision across a sample of peace accords. A second implementation data source is Jarstad & Nilsson’s (2008) Implementation of Pacts (IMPACT) dataset which contains implementation data for three types of power-sharing provisions (military, political, and territorial) found in accords signed between 1989 and 2004.
The new dataset expands upon these existing sources in two main areas. First, the new data cover the implementation of political, military, and territorial power-sharing, plus 48 additional types of negotiated provisions. The second major difference is that the new dataset is a time series and provides annual implementation data for those provisions. The new data’s longitudinal format is an especially important feature given that particular provisions in peace agreements are designed by the actors to be implemented in certain sequences that build confidence, enhance security, and accomplish other reform-oriented objectives. Implementation processes often unfold in a reciprocal fashion where actors tie their continued participation and compliance in the process to their rival’s current level of participation and compliance. For example, following the 1997 General Agreement in Tajikistan, ‘the UTO asserted that its forces could not give up their weapons and demobilize before the opposition received the 30 per cent of governmental positions due to it’ (Smith, 1999: 245). The ability of researchers and practitioners to study basic dynamics, such as how past implementation processes influence future implementation processes, has been limited by data constraints.
Case selection, provisions, and coding rules
List of comprehensive peace agreements
There are potential biases that our selection criteria may create. While it is reasonable to expect that more inclusive and more substantive agreements (CPAs) will experience less resistance or receive more support, we did not consider the outcome of an accord or levels of post-accord violence in the selection process. We find that the sample is only slightly less subject to breakdown than other samples of accords. For example, 29% of our CPAs are followed by at least minor armed conflict compared to 40% in the entire UCDP Peace Agreement Dataset (Högbladh, 2012). 4 Hartzell & Hoddie (2007: 78) have 38 settlement cases (1945–99) of which 34% were followed by renewed armed conflict. Mattes & Savun (2010) have 51 cases (1945–2005) of which 33% were followed by renewed armed conflict.
The PAM project collected implementation information for 51 different types of provisions that represent the corpus of issues addressed in CPAs. These are listed in Table II. We consider a provision to be a goal-oriented reform or stipulation that is costly to one or both conflict actors and that falls under a relatively discrete policy domain (e.g. executive branch reform, police reform). 5 In negotiations, most proposals will fall under the policy jurisdiction of some interest group that will band together and lobby against it. Police reform, for example, will be resisted (at a minimum) by the police officers that will be affected by the reforms. For this reason, we consider that the most meaningful reforms that make their way from negotiations into formal agreements are those that are costly to one or both conflict actors. Costly provisions that are mutually accepted signal commitment to the peace process. We identify six categories of provisions that constrain or bind different constituencies to the implementation process. The grouping into categories permits the conceptual linking to particular sectors in observing the impact of implementation, but the data do not require the adoption of our categorization.
Categories and provisions
When an implementation process was initiated but progress in the year was not considered viable for reaching full implementation, it was coded minimum. A non-viable implementation rate is one that if continued would not likely produce full implementation. If the implementation process fulfills some of the major steps necessary for reaching full implementation, but is still short of completion, it was coded intermediate. Full implementation represents a completed or nearly completed process. Our coding procedure also allows for reversals in the implementation process, which occur with some regularity. If a reversal took place, it was coded as a minor or major reversal. A minor reversal is a retrenchment in the implementation process that falls within the same implementation category (i.e. minimal, intermediate or full) whereas a major reversal suggests a change from one category to another. This data collection methodology yielded a dataset containing a total of 314 accord-year observations.
Three examples from Bangladesh, Angola, and the United Kingdom illustrate aspects of the coding methodology. The Chittagong Hill Tracts Accord called for the formation of an internal verification committee to monitor and report on the implementation of the accord. The historical record for 1998 indicates that the committee was established, and included an appointee of the Prime Minister, the Chairperson of the Task Force on Refugees, and Shantu Larma, President of the PCJSS (the rebel group). A UN report stated that the committee held several meetings in 1998, which were considered non-functional, and no minutes or records were kept. Multiple sources report that the committee did not meet again for 11 years. Based on these narratives, the verification provision was coded as initiated in 1998 and achieving minimum implementation that year.
The Lusaka Protocol in Angola contained a provision for the establishment of a power-sharing government under which UNITA members would receive 17 positions in the executive branch and 70 parliamentary positions. Numerous sources indicate that none of the minister positions for UNITA officials were filled in 1994, 1995 or 1996. In 1997, President Dos Santos announced that his new cabinet contained 11 UNITA positions. Later that year, it was reported that some of the parliamentary seats were filled. Based on these annual narratives, the power-sharing provision in the Lusaka Protocol was coded as not yet initiated in 1994, 1995, and 1996, with initiation and an intermediate level of implementation achieved in 1997.
Illustration of implementation coding for power-sharing transitional government in Good Friday Agreement
Description of data
The new dataset’s ordinal and longitudinal structure allows numerous ways of exploring variation in implementation processes.
Figure 1 displays the top 26 most commonly negotiated provisions in CPAs and their frequency, sorted by how many achieved full implementation by year ten or the last year of observation. 8 It is rather clear from these data that inserting a provision into a peace accord does not guarantee that high levels of implementation will take place. For example, when an accord stipulates changes falling under media reform and judicial reform, they are fully implemented only 7% and 19% of the time, respectively. When accords call for the decentralization of state power, the central government gives up that power as stipulated 15% of the time (decentralization is never initiated 40% of the time). Other types of provisions are consistently implemented in full. When an accord stipulates the granting of amnesty to former combatants (sixth most frequent), it is fully implemented 70% of the time. Both sides release prisoners (fourth most frequent), when called to do so, around 75% of the time. Where a CPA establishes a power-sharing transitional government, the arrangement is fully implemented 65% of the time. Not coincidently, these three provisions are typically scheduled to be implemented in the first year of the implementation process, and are more likely to garner the attention of third-party overseers. This reflects a general pattern in the data as well.
While it is useful to have a broad overview of the most negotiated provisions and how many reach full implementation, Figure 1 conceals the fact that the highest point of implementation reached by most provisions falls somewhere between no implementation and full implementation. Of the 724 provisions that were coded, 353 provisions or 49% were considered fully implemented by year ten or the last year of observation (whichever came first). There were 166 provisions (23%) coded as intermediate at the end of the examination period, 108 provisions (15%) coded minimal, and 97 provisions (13%) never initiated. Just under half of all provisions negotiated in CPAs are fully implemented, and 28% of all provisions are either never initiated or minimally implemented. Figure 2 provides a disaggregated view of the implementation of power-sharing transitional governments. As seen, there were five instances of a power-sharing government being partly established, two with minimal implementation, and three with intermediate implementation. Moreover, 11 of the 17 accords with power-sharing provisions were fully implemented. In all but one case, implementation was initiated. The data also suggest that power-sharing governments, in particular, are vulnerable to reversals (not shown in the figure); there were seven instances of reversals coded for this provision.
The longitudinal structure of the new data allows us to examine implementation processes over time. Figure 3 displays annual levels of implementation for a set of six provisions. The implementation line for each provision represents the average level of implementation each year for that provision across all the CPAs that contain it. As seen in Figure 3, very different trajectories are observed across different provisions. Power-sharing governments, as mentioned, tend to be implemented early (50% established by the end of year one, 60% established by the end of year two, and 80% implemented by year four). For other provisions in Figure 3, progress is slower and the trajectory flatter. At year three, Truth and Reconciliation Most frequently negotiated provisions in CPAs Disaggregation of power-sharing transitional government implementation
Average level of annual implementation for a sample of six provisions
Just as individual provisions take very different implementation paths over time, so it is with peace accords in their entirety when we consider the implementation of each of their respective provisions. Figure 4 depicts the temporal variation in the level of aggregate implementation across six CPAs over a ten-year period. 9 Scholars who have studied these particular peace processes in some depth will immediately recognize familiar inflection points and periods of stagnation that are clearly visible in the data. As already discussed, Northern Ireland’s Good Friday Agreement experienced rapid implementation in the first three years but the process stalled for several years over the decommissioning of IRA weapons and the suspension of the power-sharing arrangement. In the Nepalese case: ‘Considerable progress was achieved in the implementation of this agreement between 2006 and 2008’ (Trachsler, 2012: 2), but the ‘peace process has since reached an impasse’. These charts show how the data from individual provisions can be aggregated to create a ‘holistic’ view of implementation. In the next section, we give an example of how the data might be used to summarize the measurements of several provisions that are theorized as being part of a similar process.
Application of the data
Although provisions in peace accords are usually studied on an individual basis, many are implemented as part of a larger multidimensional process. One such process is security sector reform, or SSR, a concept that gained prominence in the mid-1990s, and refers to that ‘set of policies, plans, programs, and activities that a government Annual variation in the level of aggregate implementation of six CPAs
Based on several definitions of SSR and its elements (Schnabel & Ehrhart, 2005: 6; Newman, Paris & Richmond, 2009: 8), we identify eight provisions in our dataset mentioned as falling under security sector reform (disarmament, demobilization, reintegration, military reform, police reform, ceasefire, paramilitary groups, withdrawal of troops). We found that many of these eight provisions are typically negotiated as a cluster. As seen in Figure 5, none of our CPAs contained only one SSR provision and over 75% of CPAs (26) contained five or more SSR provisions.
Frequency of SSR provision in CPAs
Pairwise correlation of implementation of eight security sector provisions
*p < 0.05.
Implementation of SSR provisions and recurrence of armed conflict
Robust standard errors in parentheses. *p < 0.05, **p < 0.01, ***p < 0.001.
Effects of implementing SSR reform on recurrence of armed conflict over a decade
We estimated three Weibull models (accelerated failure time (AFT) metric form) for each dependent variable. We include a number of standard control variables from the civil war literature. In the majority of cases, there was violence beyond the 25 battle-death threshold in the year that the peace accord was signed. Following convention, all of the cases are allowed to survive at least one year (Jarstad & Nilsson, 2008). This also allows fair time for implementation to get underway. Therefore, the presence or absence of armed conflict begins to be observed in year two.
The results from these models are presented in Table V. We started our empirical tests by including in Model 1 the implementation of military reform, which can include military power-sharing or the downsizing of the military. Military reform, by itself, is statistically significant which is consistent with earlier findings (Hoddie & Hartzell, 2003; Jarstad & Nilsson, 2008). Model 2 includes the newly created SSR implementation index (excluding military reform) alongside the implementation of military reform, and Model 3 includes the SSR index with military reform included. The SSR index is highly significant (p < 0.001) in every model with the relationship running in the expected direction; military reform, however, falls from statistical significance in Models 2 and 4. This example illustrates that the implementation of related provisions can have a ‘cumulative’ effect (Hartzell & Hoddie, 2003) where the combined effects of several interrelated provisions is greater than the effect of any individual indicator. This same logic suggests that interdependency can be disabling when a failure occurs in one provision that negatively impacts concurrent implementation processes. Interestingly, in Models 5 and 6 our results suggest that high levels of SSR implementation reduce conflict not only between the groups that negotiated the accord but also between the government and non-signatory groups (i.e. other factions or splinter groups within the same conflict ID). These results are presented graphically in Figure 6 (derived from Models 3 and 6). We now turn to some concluding remarks.
Conclusion
Mediators and civil war actors often spend years in negotiations trying to reach mutually acceptable terms on a wide variety of conflict issues. Theoretical efforts to understand civil war have analyzed violence as a result of bargaining failures that prevent the warring parties from being able to reach mutually acceptable terms ex ante. Despite the practical and theoretical importance attributed to the particular set of terms that are reached between conflict actors in negotiations, the amount of research devoted to the implementation of these terms has been limited due to data constraints. The structure of the data we introduce allows researchers and practitioners the ability to study implementation at several different levels of analysis. Researchers can focus on a particular agreement and its provisions, or one type of provision across all the accords that contain it, or a group of provisions and their joint implementation at a particular point in time. The data also offer a systematic methodology for evaluating and measuring agreement implementation in contemporary peace processes. Given the impressive body of scholarship produced in the last ten years on just a handful of the types of provisions found in CPAs, we envision many potential applications of the new data and a deepening of our understanding of implementation processes writ large.
Footnotes
Replication data
The dataset, codebook, and code files for the empirical analysis in this article can be found at http://www.prio.org/jpr/datasets and 
.
Acknowledgements
The PAM project was established by John Darby with funding from the Kroc Institute and the United States Institute of Peace (USIP-149-06F). Additional funding came from the National Science Foundation (0921818). We thank Peter Wallensteen, Erik Melander, and Desirée Nilsson for excellent comments and suggestions.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
