When Deadlock Brings Peace: The Impact of Negotiation Length on Post-Agreement Implementation

Transcription

1 When Deadlock Brings Peace: The Impact of Negotiation Length on Post-Agreement Implementation Deniz Cil University of Maryland Abstract While previous studies show that conflict among the same groups is less likely to recur if implementation of an agreement is successful, little work has focused on identifying the factors that lead to successful implementation. In other words, following a negotiated settlement to a civil war, what causes warring parties to fulfill their promises of implementing reforms in different issue areas instead of reneging or returning to violence? Similarly, why are some peace agreements fully implemented while others are only partially or never implemented? The main argument advanced is that dynamics at both the negotiation and implementation stage should be taken into account to fully understand successful transitions to peace, and the incentive of parties to continue implementation or not. This paper, in particular, focuses on one of the important pre-agreement factors rounds of negotiations as an explanation for post-agreement outcomes. When parties value future gains from an agreement and anticipate few enforcement problems in the postagreement period, it takes longer to reach an agreement. In these cases, the level of implementation will be higher. The observable implication is a tougher and longer negotiation process preceding the final agreement. This proposition is tested using original data on pre-agreement negotiations and agreement implementation that includes all final agreements signed between 1989 and 2009 in civil wars. Preliminary findings support argument presented. 1 Paper prepared for Journeys in World Politics Workshop University of Iowa, March 31-April 2, 2016 Working draft, please do not cite. 1 This material is based upon work supported by the National Science Foundation under the Dissertation Improvement Grant No. ( ) and Smith Richardson Foundation World Politics and Strategy Fellowship. 1

2 1. Introduction Why did the El Salvadorian government successfully implement most of the provisions in the Chapultepec Agreement of 1992, while Lomé Peace Agreement of 1999 signed between Sierra Leone government and the RUF failed immediately even though both agreements had provisions on constitutional, electoral, political and military reforms, included measures of guaranteed power sharing between combatants while UN peacekeepers/observers United Nations Observer Mission In El Salvador (ONUSAL) and United Nations Mission in Sierra Leone (UNAMSIL), were present to oversee the agreement? What makes the parties, especially governments, fulfill their promises to make certain reforms in various issue areas ranging from political to economic and military? In other words, why are some agreements fully implemented while others are only partially or never implemented? There is no doubt that the successful implementation of agreements is crucial for durable peace. Previous studies (see; Hartzell and Hoddie 1999 and 2003, Jarstad and Nillson 2008) show that conflict among the same groups is less likely to recur if the implementation of an agreement is successful. Since most of the agreements included in previous studies (as well as in this paper) are the result of a series of strategic negotiations, and consist of concessions made by both sides on various issue areas to end the conflict, once the provisions are followed through it is highly likely that peace will be durable. What is more puzzling and missing in most of these studies is how the parties achieve successful implementation. Even though the implementation stage is as important and as dynamic as the earlier stages in conflict resolution, we do not yet have a comprehensive theory and empirical analysis of the post-conflict years following peace agreements. Existing work in this area is composed of case studies looking in depth at the implementation processes of some civil wars (e.g, Stedman et.al 2

3 2002, Hampson 1996), or books/articles focusing on individual aspects of transitions such as postconflict elections, disarmament and reintegration programs (e.g, Hoddie and Hartzell, 2010). This paper seeks to build on this prior work by providing a general theory of the implementation process and testing implications using new data on the implementation of peace agreements. I argue that the negotiations on an agreement and implementation of that agreement are two different yet interdependent stages of bargaining between the parties involved in the conflict. Expectations about the implementation stage are likely to affect the way parties bargain to reach an agreement, thus the first stage of bargaining should be taken into account when explaining the implementation process. When parties expect that implementation of an agreement will be successful or when they anticipate fewer enforcement problems, they will negotiate longer and carefully in order to reach best possible agreement. 2 This paper presents the results of preliminary analysis to answer the question of what explain variation in the level of agreement implementation. This paper aims to contribute to conflict termination literature by focusing on the process of implementation. Additionally, given the relationship found between the successful implementation and duration of peace in various studies (i.e. Jarstad and Nillson, 2008), explaining what leads to successful implementation also has larger implications for conflict recurrence. The following section reviews the literature on the negotiated settlement of civil conflicts and peace duration. The third section presents the theory 2 While this paper focuses on aggregate level of implementation, progress in implementation of agreements over time is the second part of the larger project in which factors affecting actors incentives during the implementation stage is examined. The general theory put forward is that the driving force behind keeping promises and continuing the implementation process depends on the costs of non-compliance. Non-compliance can be costly because of the continued threat of renewed violence by one of the parties (self-enforcement). Non-compliance can also be costly due to external pressures coming from donors (e.g. foreign aid) or international organizations (e.g. sanctions). In the presence of these pressures, parties are more likely to continue making progress. Furthermore, the changes in parties bargaining power in the implementation stage due to exogenous (e.g., external support, deployment or withdrawal of third party mission etc.) or endogenous (e.g., disarmament of forces, redistribution of economic resources or political power etc.) factors creates room for renegotiation, which in turn influence parties willingness to continue with the implementation of a given agreement or not. 3

4 and hypotheses, while the fourth section describes the data and the model, followed by a discussion of results. The final section lays out the steps for improving the analysis presented here as well as prospects for further research. 2. Literature Review One of the two general approaches explaining civil conflict termination focuses on factors that facilitate initiating negotiations and/or signing agreements (e.g. Walter 1999 and 2002; Zartman 1989), while the other one focuses on the design and implementation of peace agreements with an aim to explain when peace is more durable (e.g. Hartzell 1999; Toft 2009; Jarstad and Nilsson 2008). Groups in civil war cooperate as long as agreements guarantee long term benefits. To achieve a settlement parties should send costly signals to their adversaries, and more power (political, military and/or economic) should be distributed among rivals to overcome commitment problems and guarantee future security (e.g., Walter 1999; 2002, Hartzell and Hoddie 2008). The guaranteed power sharing, as opposed to competitive elections and alike, is a necessary but not sufficient condition for civil war settlement because rebel groups know that the government will renege on the terms of the agreement once it becomes more powerful, therefore refuse to sign an agreement in the first place even if all root causes are addressed (Walter 1999). Several studies show that third party security guarantees, are the most important factor that brings groups to the negotiating table and lead to successful settlement since they solve the enforcement problem, and ensure durable peace (Walter 2002, Fortna 2004, Quinn et. al 2007). In addition to the third party security guarantees, several other studies show that the design of a peace agreement, i.e. types of provisions and power sharing, and security sector reforms, affect peace duration following civil wars. For example, more institutionalized settlements in terms of 4

5 addressing security concerns of rebel groups about control over coercive mechanisms of the new state, political power and economic resources are linked to post-war stability (e.g., Hartzell 1999, Harzell and Hoddie 2003a). However, agreements are usually designed to benefit the ones that comply with the terms of the agreement but not necessarily include provisions about punishing the ones that violate the terms. Negotiated settlements without the credible threat of punishment are vulnerable to cheating or tactical ceasefire and lead to increased likelihood of renewed violence (Toft 2009). In addition, uncertainty-reducing agreements, such as the ones that include UN observer missions to monitor the implementation of agreements are more likely to reduce information asymmetries in the post-conflict environment and decrease the probability of conflict recurrence (e.g., Savun and Mattes 2010). Moving on the link between the agreement implementation and peace duration, Hartzell and Hoddie (2003b) show that the full implementation of military aspects of an agreement significantly improves the prospects for maintaining the peace since implementation is a costly signal for commitment to peace. Jarstad and Nilsson (2008) later extend the study of agreement implementation and focus on three different pacts (military, political and territorial) and the effect of implementation on peace duration. The authors argue that implementing military and territorial pacts are more costly compared to political pacts and the implementation of pacts that carry high costs enhance the prospects of peace (Jarstad and Nilsson 2008). Like Hartzell and Hoddie (2008, 2003a), they also apply the logic of costly signaling which suggests that parties, by engaging in costly concessions, reveal their good intentions of promoting peace (Jarstad and Nilsson 2008). Likewise, DeRouen et.al (2009) looks at the implementation of different power-sharing provisions and their effects on the duration of peace. They also find that implementation of military and territorial power-sharing provisions (e.g., autonomy) lead to longer peace. However, their 5

6 reasoning is different than what Jarstad and Nilsson (2008) suggest. DeRouen et.al (2009) argue that political-power sharing is costly and that such arrangements would create an incentive for the government to renegotiate or renege as they gain the upper hand in the future compared to less costly power-sharing agreements such as military power-sharing. Even though the government takes a risk by allowing former rebels into army, in the long run it will likely retain control of this important institution. More recently, Joshi et.al (2015) show that sequence of implementation also matters in explaining peace duration. Post-conflict elections preceded by high levels of implementation make conflict recurrence less likely, while post-conflict elections without high levels of implementation have destabilizing effects (Joshi, Melander and Quinn 2015). While these studies provide us a good understanding of the relationship between successful implementation of peace agreements and duration of peace, and several case studies (e.g., Stedman et.al 2002, Hampson 1999, Brosché 2008, Bekoe 2008) offer insights into the implementation of agreements, there are not many studies that explore the implementation process cross nationally. What makes the parties actually follow through on the promises made in these agreements? Why is it in some cases that governments successfully introduce political and economic reforms in the post-conflict period, while in other cases they make little or no progress? In other words, why some peace agreements are successfully implemented, while others fail? 3. Theory and Hypothesis When the main issue of implementation is the demobilization of forces, any move towards implementation becomes very risky for both parties so they need third party guarantees to enforce the agreement terms. Although the implementation process is risky and creates a lot of uncertainties about the compliance of parties, not every agreement requires the full demobilization and disarmament of groups. On the contrary, agreements usually envisage partial demobilization 6

7 and integration into a new national army where equal or proportional numbers of government and rebel forces are represented, and/or appointment of rebel leaders to high ranking positions in the military or government. In some cases, the ratio of rebel forces to government forces is set to be 50-50, while in other cases rebel and government forces are represented in the new army proportionally. The Bicesse Agreement signed between the government of Angola and UNITA, which called for the incorporation of 30,000 UNITA members into the Angolan Armed Forces, Navy and Air Force, or the General Peace Agreement (The Acordo Geral de Paz (AGP)) signed between the government of Mozambique and Renamo which called for the incorporation of 15,000 Renamo soldiers in to the army are a few good examples of such arrangements. These arrangements would enable the rebel group to enforce the terms of the agreement as well as make them more likely to sign an agreement since they will be less concerned about enforcement problems in the implementation stage. Additionally, even though peace agreements can be self-enforcing by having security measures as in above examples, conditions under which integration of ex-combatants to new army can be realized has yet to be addressed. Moving from agreeing on certain political and security arrangements for the post-conflict period, to putting these provision in place is a crucial step towards achieving peace. Moreover, previous studies only look at the implementation of powersharing provisions in a given agreement and its effect on peace duration, whereas agreements usually have numerous other reforms that do not explicitly require power-sharing. Therefore, implementation of such agreements remains to be explained. Theoretical framework presented in the rest of this section aims to address some of these gaps and provide an explanation of variation in the degree of peace agreement implementation. 7

8 I argue that reaching an agreement and implementing it are two different yet interdependent bargaining processes and expectations about the implementation stage affect how parties bargain and reach an agreement in the first stage. I make the assumption that the parties act rationally and would take into account the costs and benefits of achieving their desired goals through negotiation or fighting and make a decision on which one to use accordingly (Pillar 1983, Reiter 2009). The timing of negotiations, therefore, depends on the prospects for future military success or failure (Pillar 1983). In general, information revealed by fighting might eliminate the information asymmetry over capabilities that exist at the beginning of a conflict and parties are more likely to make concessions if they believe future fighting will be costly or the probability of success through fighting is low (Reiter 2009). I argue that the first stage of a peace process, negotiations between a government and a rebel group to reach an agreement, resembles war of attrition model presented by Fearon to explain the interstate cooperation (Fearon 1998). The model predicts that when the shadow of future is long (expectation of future gains and long term cooperation), both sides prefer to hold out and wait with the hope that the other side will make concessions first. However, as the parties hold out to reach a better deal, they suffer the costs of noncooperation. One implication from the model is that although eventually longer shadow of future makes agreements more enforceable, it also makes the time to reach an agreement longer. In the context of civil war termination, the cost of non-cooperation is the cost of continued fighting. Thus, as a general condition, I assume that the negotiations occur when both sides have no expectation that they might get a better outcome by continuing fighting. While it is likely that rational actors should always try to find ways to negotiate in order to avoid the costs of fighting, 8

9 the decision to initiate negotiations depend on each combatant s belief that future fighting is more costly compared to negotiating and ending conflict. Therefore, the process of negotiations is dependent on the expectations of the parties in the post-agreement period. Civil was combatants likely to value future payoffs from peace agreements differently. For example, rebel groups might value the provisions regulating future elections, integration into government and structure of political system more compared to other provisions. The reason is that having a favorable political setting once these provisions are implemented will provide gains not only in the immediate aftermath of the conflict but also in the in the long-run. If combatants believe that these agreements will be implemented, they are more likely to work towards reaching a better deal. As argued above, the implication is that time to reach an agreement becomes longer. When the parties value future gains, and when they do not expect implementation problems they will negotiate tougher and longer. In other words, if the parties expect to be able to enforce the agreement in the second stage, they will bargain more seriously in the first stage. Conversely, if the parties do not believe that the agreement is likely to be fully implemented or believe that once they settle they expect to get other benefits besides gains from the agreement, they might be inclined to reach a quick agreement and do not pay much attention to the details. In return, agreements that are a result of long and tough negotiation process are likely to have higher levels of implementation compared to agreement that are quickly concluded. H: The longer the negotiation process preceding an agreement, the higher the degree of implementation will be. However, parties might reach an agreement because of outside pressures during the negotiation process. Both in interstate and intrastate wars, third party involvement in any stage of conflict might affect outcomes especially the outcome of a negotiated settlement (e.g., Walter 2002, Fortna 2004, Regan and Aydin 2006, Quinn et. al 2007, Beardsley and Lo 2014). Various 9

10 studies show that, while third party involvement might create incentives to sign an agreement in the short run, the long term effects might be adverse if the agreement did not reflect expectations about conflict outcome had the parties continued fighting, and when third party mediators are biased and/or use various types of leverage (e.g., Werner and Yuen 2005, Svensson 2009, Beardsley 2011, Reid 2015). One potential scenario is when there is an external intervention is in favor of a given rebel group or when there is pressure on the government from the international community to end the conflict, which results in an agreement where the rebels gain more than what they would without any intervention. Similarly, third party presence in the negotiations might pressure parties into reaching a quick agreement with the hopes that they can alter it later. As the international pressure fades, the government will implement less than initially agreed upon. Therefore, the involvement of third parties during negotiation stage might create incentives to renegotiate, revise the terms of the agreement during the implementation stage, or renege on some or all of the terms of the agreement in the second stage. In these cases, the level of implementation will also be lower. In other words, involvement of third parties in some capacity during negotiations might influence the dependent variable (the degree of implementation) and the independent variable (the length of negotiations) separately and bias the statistical analysis. To account for this I control for the involvement of third parties during the negotiation process, which is discussed in more detail below. Lastly, long negotiations might be a product of many agenda items. If the parties are negotiating on a lot of different issue areas, then the negotiation stage might last longer, not necessarily because they are trying to reach a better agreement, but because there is too much 10

11 ground to cover. For this reason, I also use negotiation days per provision, total days of negotiations divided by the total number of provisions in the agreement, as an alternative measure. Next section describes the dataset and the main variables used in the statistical analysis. Further discussion of alternative explanations and confounding factors is provided in the control variables sub-section. 4. Data 4.1. Implementation of Peace Agreements Dataset (IPAD) In order to test the hypothesis presented above, I constructed an original dataset. Implementation of Peace Agreements Dataset (IPAD) identifies the degree of implementation following final peace agreements after civil wars. It builds on the Uppsala Conflict Data Program (UCDP) Peace Agreement dataset (v.2.0) (Hogbladh, 2011) and includes all final agreements regardless of their content, i.e. comprehensiveness and/or inclusiveness. It records 80 peace processes that took place between 65 government-rebel group pairs in 38 civil conflicts between the years 1989 and These processes result in 68 final agreements. The data is recorded at two levels. Process level data includes information on the start and end date of a peace process as well as the level of implementation of each provision at the end of a given process. The event level data includes each implementation event and the implementer (government, rebel group, or third parties) over the course of a given peace process, along with event type (single day event or summary event) and temporal precision codes, which is suitable for time series analysis. The current analysis uses the process level data. The need for creating an original dataset stemmed from the fact that currently available datasets do not provide adequate coverage of agreements and/or their provisions in order to answer 11

12 the questions raised in this paper. 3 One main limitation is sample selection. The inclusion of all final agreements and all provisions within these agreements, not only the power sharing pacts, is needed to improve measurement and statistical inference. Inclusion of all final agreements regardless of their content, i.e. comprehensiveness and/or inclusiveness is especially important to explain variation in implementation. Failing to do so may introduce selection bias. The main problem with selecting on the content of agreements is that more comprehensive agreements might have higher levels of implementation. There are two reasons why that might be the case. First, parties might design more comprehensive agreements when they think that implementation will be successful. Second, parties might have more incentive to push for successful implementation because they have a lot more to gain from these agreements. In sum, to answer the question of why some agreements get implemented while others fail, information on all final agreements, regardless of their success and content is needed. Starting from the UCDP peace agreement dataset, 80 peace processes across 65 dyads in 38 civil conflicts between 1989 and 2009 have been identified. A peace process is defined as continuous talks that lead to a final agreement signed between one state and one non-state group, in which the final agreement at least partially addresses the incompatibility in the conflict and lasts for at least one month. The UCDP Peace Agreement Dataset includes 183 peace agreements of various types signed within the timeframe of this analysis. Among these agreements, 68 final agreements are identified based on following criteria: 3 For example, the Peace Accords Matrix from the Kroc Institute for International Peace Studies at the University of Notre Dame, and the Implementation of Pacts (IMPACT) by Anna Jarstad, Desirée Nilsson & Ralph Sundberg or Power Sharing Event Dataset (PSED) by Ottmann and Vullers. 12

13 1) The agreement must be the final product of a peace process. Any agreement that initiates a process and/or references future agreements is not included. These agreements by nature start a process however, some do not lead to final agreements. Seven process agreements signed between Uganda and the Lord Resistance Army (LRA) between 2007 and 2008 as a part of the Juba Peace Process are good examples of this exclusion criteria. The Juba Peace Process was supposed to end by a final agreement, after which all other agreements would be implemented. After several delays in signing the final agreement, in September 2008, chief mediator Riek Machar disbanded the cessation of hostilities monitoring team and declared the peace process over. (UCPD Peace Agreement dataset). In other words, the second stage (implementation) never started. Similar examples of failed processes indicate that implementation does not start until the final agreement is signed. Since the purpose of the analysis is to explain variation in the degree of implementation, these types of agreements are excluded. However, all process agreements that lead to a final agreement, 74 in total, are reviewed and the provisions are accounted for if they are not already incorporated into final agreements. 2) An agreement must last at least one month. This criteria enables enough time to establish the reasonable start of a post-agreement process. One example is the Paris Accord signed between Air and Azawad Liberation Front (Front de libération de l'aïr et l'azaouad - FLAA) and the government of Niger. This agreement led to internal divisions within the movement and the main rebel faction denounced the agreement making it obsolete within a month after it was signed (UCDP Conflict Encyclopedia). 13

14 3) Thirteen agreements that are follow-ups or make modifications to final agreements are reviewed and their provisions are accounted for. However, these agreements are not treated as the start of a new process and are not included as standalone agreements. 4) Ceasefire agreements are also excluded. The reason is that the aim of ceasefire agreements is to end fighting for a certain period of time without any obligation to share power/resources etc. and/or introducing reforms to address main incompatibility. Therefore, the effort required in their implementation is not comparable to agreements that contain concessions, broad reforms and/or power sharing at various levels. Additionally, even though both types of agreements (final and ceasefire) can have tactical use, it is reasonable to expect that ceasefire agreements are more prone to be used for the purposes of gaining time and recouping capabilities since they impose less costs and obligations (e.g., the Tamil Tigers were accused of repeatedly signing ceasefires with the purpose of buying time to rebuild). Second, in order to accurately measure the degree of implementation, this dataset identifies the end of a peace process. Existing datasets use 5 or 10 years following an agreement as a cut-off point for the end of peace process (or if conflict recurs). However, both cut-off points might generate potential measurement problems for the purpose of this analysis. The first cut of point may be missing important implementation events that take longer than 5 years to implement, which leads to underestimation of the degree of implementation. A good example is the implementation of Chittagong Hill Tracts Peace Accord which was signed between the Government of Bangladesh and the rebel group JSS/SB (Parbattya Chattagram Jana Sanghati Samiti/Shanti Bahini; Chittagong Hill Tracts People's Coordination Association/Peace Force) in 1997 to end the conflict in the Chittagong Hill Tracts (CHT) districts. One of the main issues in the implementation process was 14

15 that the amendment of Land Reform Act of While the New Chittagong Hill Tracts Land Dispute Settlement Commission Act 2001 (Act 53 of 2001) passed on July 17, , the new version did not meet the criteria specified in the 1997 agreement. A new draft bill was approved by the Cabinet in June 2013 and sent to the Parliament s land reform commission for approval which is yet to legalize the amendments. 5 By contrast, a 10 year cut off might be counting many years during which implementation is already over. One good example of this situation is the implementation of Sudan Comprehensive Peace Agreement (2005). The implementation process ended after the January 2011 referendum as envisaged in the agreement, which resulted in the independence of South Sudan. Therefore, it is necessary to identify the duration and outcome of all peace processes. A number of empirical and theoretical criteria serve to justify why particular start and end times were used. In order to identify a start date I conducted research on negotiations preceding agreements. A peace process starts with the first round of negotiations that leads to the signing of a given agreement. While parties might make several attempts to negotiate, this criteria excludes unsuccessful attempts that do not lead to an agreement. Negotiations that are not direct, failed before signing a final agreement, or have been interrupted by at least two years of active conflict are excluded. The start date of the first direct round of negotiations is then recorded as peace process start date. This process generated 149 pre-agreement years. The end of a peace process is defined in two ways. A process is considered terminated when the last observed implementation event takes place and no other implementation (or reversal of implementation) event occurs for five consecutive years. The second definition for termination 4 Bangladesh Gazette The People s Republic of Bangladesh Extraordinary Published by Authority Commission-Act-2001.pdf, 5 The Daily Star 6_27_2013 Bangladesh: Amendment of CHT Land Commission Act 2001 needed. 15

16 is when conflict recurs between the state and the rebel in the peace dyad. When there are multiple signatories one government and multiple rebel groups each dyad may have different start and end dates. This generated a total of 332 post-agreement years. In addition, for each peace process type of termination recorded. A total of 30 out of 80 peace processes (37.5%) ended with conflict recurrence, 28 peace processes (35%) ended with successful implementation, and 17 peace processes (21.3%) resulted in only partial implementation, where there was no activity for 5 years since the last recorded activity and not all parts of the agreement were implemented. In addition, there are five peace processes (6.2%) ongoing as of December The ongoing peace processes are not included in the statistical analysis. Detailed information on the content of agreements can be found in the appendix Dependent Variable Level of Implementation Level of implementation for each provision recorded as a categorical variable. Limited implementation (1) means that the implementation of a given provision was initiated but was not carried out and/or did not go beyond a symbolic gesture. Partial implementation (2) means that the implementation of a given provision was initiated and several aspects of was implemented, but overall it fell short of full implementation as identified in the agreement. Full implementation (3) means that all aspects of the provision was implemented. No implementation (0) means that the parties did not start the implementation of a given provision. No implementation is recorded in two ways, either the peace process ended before the implementation of a given provision or there is supporting evidence (i.e. reference in NGO, IGO or news reports) that the implementation of a given provision never initiated. The aggregate level of implementation is then calculated as a ratio of total implementation score over maximum implementation score. For example, if an agreement has 10 provisions 16

17 maximum implementation score is 10 times 3 (maximum implementation level for each provision). If half the provisions implemented fully (5*3) and 3 of them implemented partially (3*2) and 2 of them were not implemented (2*0), the total implementation score is 21 and the aggregate level of implementation is 0.7 (21/30). The mean implementation level in the dataset is I primarily relied on open sources such as news reports accessed through Factiva database, various NGO reports, case analyses and secondary sources such as academic articles and books to record implementation of provisions. While news articles yield disaggregated information at the event level, a valid concern still remains about media bias. There can be two types of biases that might affect the statistical analysis based on data generated by media-reporting. One is that the news agencies only cover important events that they deem news worthy but there might be a lot of progress being made in implementation that does not get media attention. The second is that the original local source used by the international news agencies might be biased. In some cases international news agencies rely on reports from various local sources that may be state-owned news outlets, independent radio stations, and/or local as well as regional agencies (e.g., the South African Press Association SAPA). Some of these sources might have their own agenda and therefore might be biased in their reporting. State owned radio stations, for example, might try to put the blame on the rebel group for lack of progress or purposefully not cover progress made towards implementation by the rebel in order to cover up lack of progress on the government side. Alternatively, reports might exaggerate the progress made by the government. The data based on such reports, consequently, would systematically under or over report progress made one of the parties. In order to address these biases two strategies are being used. 17

18 Firstly, I make use of United Nations Secretary General Reports and General Assembly Reports. For the cases, where a UN report is not available, the following sources were used to validate media-based event collection: African Union Progress Reports. Conciliation Sources Accord Series, International Crisis Group Reports, and Yearbook on Peace Processes published by the School for a Culture of Peace, Peace Accords Matrix annual summaries, and Power Sharing Event Dataset (PSED) event summaries. In addition, source information was not limited to major news outlets, such as Reuters, Agence France Press (AFP) or British Broadcasting Corporation (BBC) but major regional and local news sources are included (e.g, Jakarta Post, Business World, Kyodo News). For the second type of bias, if a reference is made to a local source and/or stateown media outlet, or when the news article cites a rebel group or the government as a source, multiple additional sources are tracked down to cross-check the event coding Independent Variable Length of Negotiations Two main independent variables used in the analysis are the total number of negotiation rounds and the total number of negotiation days. The total number of negotiation days is created based on start and end date of each round. One potential measurement problem is that the parties might have few rounds that last long or have numerous rounds of negotiations, each last for a few days. Both examples might introduce measurement problems. Since the purpose is to identify attempts of long and serious negotiations; 2 rounds each lasting for 15 days and 10 rounds each lasting 3 days might be considered as equal for the purpose of measuring the length of negotiations. Therefore both variables are used in the statistical analysis. A round of negotiation (whether single day or multiple days) is defined based on the following criteria. Parties have to engage in direct negotiations to talk about issues pertaining to resolution of the conflict and the incompatibility. A direct negotiation takes place between the 18

19 warring parties represented by the leaders, representatives and/or negotiating teams and takes place in one location and a start and an end date can reasonably be identified. 6 Negotiations that set an agenda for peace talks, address the incompatibility and post-conflict political setting, conflict behavior, i.e. conditions for ceasefire, demilitarized zones, security issues such as disarmament of combatants, resettlement of refuges, or other topics that are included in the peace agreement are included. While talks about talks in which parties agree to hold negotiations without setting an agenda or addressing any substantive issue are excluded, since the main purpose is to capture the serious and tough bargaining process given the expectations about the implementation of an agreement. This criteria generates 157 pre-agreement years counting from the year in which the first round of negotiations occurred to the year of agreement. Once the first direct negotiations between the parties is identified, I used open sources, mainly above mentioned ones, to identify rounds of negotiations preceding the agreement. 536 rounds of negotiations are recorded. In cases where the start and end date was not clear, temporal precision codes are recorded in order to account for uncertainty. 94% of rounds were recorded at precision level 1; both start and end date of a given round is known at the day level. In addition, a separate variable records two types of negotiations. Single day and/or continuous negotiations for each day between start and end date are record as type (1), while a round with interrupted talks where the parties had negotiations on and off or cases in which it is not clear if parties are negotiating every day are recorded as type (2). 94% of the rounds are type (1). 6 In some occasions the parties meet in one city and then decide to meet in another city and/or the capital of the country very shortly after to continue talks. While these talks might be continuous with couple days and/or weeks of interruption, these two talks are recorded as two separate rounds. 19

20 4.4. Control variables Several control variables simultaneously or in separate models are included in the analysis. First one is an aggregate measure of the specificity of the provisions in a given agreement. One alternative explanation for higher levels of implementation is that the highly precise and specific agreements might be inherently easier to implement since every aspect of the post-agreement stage has been planned out and regulated. Alternatively, if the provisions regarding to reforms in the political structure, for example, are very detailed and specific, then the only job left for the implementation stage is to pass the laws and/or regulations as approved by the parties instead of designing new legislation. Therefore, Detailed Agreement variable is expected to be positively correlated with the degree of implementation. For each provision, IPAD records how detailed and specific a given provision is. A categorical variable that ranges from 1 to 3 (1 least detailed, 3 most detailed) is coded using the original text of the agreement and/or unofficial translation provided by UCDP. Out of 68 final agreements, 9 agreements are only available in summary format. Therefore, this variable for is not available for these agreements. The Detailed Agreement variable is then aggregated as a ratio of the sum of the detailed score to the maximum detailed score for the given agreement. For example, if half of the provisions in a 10 provision agreement are very detailed (5*3) and the other half is not detailed at all (5*1), the Detailed Agreement variable is 0.66 (20/30). A similar reasoning can be applied to the agreements with a timeline. If an implementation timeline is built into the agreement, the level of implementation might be higher for one of the two reasons: either timeline signals the commitment and willingness of parties to implement the agreement or the fact that parties included a timeline makes them stay on course and implement more of the agreement. Timeline variable is a binary variable that records whether or not a given 20

21 agreement includes an implementation timeline as a part of the agreement (in the main text or appendix). Timeline variable is expected to be positively correlated with the dependent variable. Similarly, the fact there are more gains to be made from implementation of an agreement, especially for the rebel group members, might incentivize them to press for higher levels of implantation. If a rebel group received major concession such as guaranteed positions in the government and/or share of revenues from natural sources, than it might be more likely to make sure that these provisions are implemented. Alternatively, more capable rebel groups might be more likely to reach such favorable agreements and receive more concessions from the government, compared to less capable groups (e.g., Gent 2011). In turn, more capable groups might also be better at enforcing the agreement, either through threating to return to violence or having capability to monitor the implementation process themselves. These agreements are then likely to have higher levels of implementation. In order to capture this dynamic, I construct Guaranteed Power Sharing variable. Guaranteed Power Sharing is a ratio of the guaranteed power sharing provisions regarding political, military, economic and territorial issues over the total number of provisions. Higher values of guaranteed power sharing is expected to be related with higher levels of implementation. Another alternative explanation for higher degrees of implementation might be inclusiveness of the agreement. The UCDP Peace Agreement Dataset records whether or not an agreement is comprehensive. Comprehensive agreements include all conflict dyads, compared to dyadic agreements where at least one of the warring parties in the conflict is excluded. Comprehensive Agreement is a binary variable that records whether or not a given agreement includes all dyads in a given conflict. The expectation is that the implementation level of comprehensive agreement might be higher due to the fact that the all parties to a conflict agreed 21

22 on a settlement, as opposed to cases in which a government signs an agreement with one group while fighting against another one. The implementation of an agreement in this scenario might suffer due to continued fighting that would occupy the government s attention and time. Another factor that might affect the level of implementation is the presence of third party peacekeepers in the implementation process. The involvement of peacekeepers for securing peace, establishing trust, and monitoring implementation is a crucial part of the post-conflict processes. As previous research suggests (e.g., Walter, 2002; Fortna, 2004, Hultman et al. 2013, 2014), the presence of peacekeepers makes duration of peace longer, probability of recurrence less likely and in cases where they are deployed in ongoing conflict situations, mitigate conflict intensity and civilian victimization. I used the data on personnel commitments to United Nations peacekeeping operations from the International Peace Institute s Peacekeeping Database, which records total uniformed personnel contributions by type and contributing country for each mission from in the post-cold war period. The number of deployed troops, police and military observers are coded at the monthly level for the duration of the mission. I aggregated the data at the process level and used the maximum number of total deployed personnel for a given peace process. I also run separate models with a binary variable recording peacekeeper presence and the mean number of peacekeepers throughout the peace process. Neither variation of the variable changed the results presented in the following section. As discussed above the presence of third parties during the negotiation stage might incentivize parties to reach a premature agreement, which in turn would likely to impact the level of implementation negatively. For each round of negotiations, I record whether or not the third 22

23 parties were present as mediators, facilitators or observers. 7 In cases of overlap, I cross-checked the third party presence coding with direct talks recorded in Managing Interstate Conflict Africa Dataset (v 2.1) that covers years between 1993 and 2007 (Melander and Uexkull 2011). Out of 536 total negotiation rounds, 166 rounds (%30) did not experience any third party presence, while rest (%70) had one or more third parties present. Since throughout a negotiation process, third parties might be present in one or many rounds, I include a variable Third Party Presence (Negotiation Stage) that records percentage of negotiation rounds in which third parties were present in some capacity in a given process. For example, a value of (1) means that in all rounds of negotiations, third parties were present, while (0.5) means that third parties were present in half of the rounds. The expectation is that while the presence of peace keepers in the post-agreement period is likely to have a positive effect, the presence of third parties in the negotiation stage is likely to have a negative effect. In addition to the third party peacekeeping and monitoring missions, in some cases parties also establish their own monitoring commissions. These commissions are composed of members from both sides, and in some cases from third party states or international organizations as observers. The tasks and composition of these domestic implementation commissions vary. In some cases, parties form several commissions and sub-commissions, each dealing with one aspect of implementation. IPAD records information on these commissions and their activities at the implementation stage. These commissions are grouped in two general categories based on their tasks; political and military. Political commissions are primarily tasked with monitoring of the 7 The data also records information on the identity of each third party, i.e. state, international organization, faith based or other type of organization. In line with the literature, I intend to expand the measure of third party involvement to account for different types of leverage, i.e. material (Beardsley 2011) and/or contextual/cultural (Reid 2015), third parties might have over combatants to accurately measure the potential effect third parties on the length of negotiations and the level of implementation. 23

24 overall implementation of the peace agreement and oversee other commissions when applicable, while military commissions are tasked with monitoring provisions regarding demobilization, disarmament, and integration of ex-rebels into army and/or reformation of existing army. Out of 80 peace processes, 51 of them had one or more type of domestic commissions. Various examples include Commission for the Consolidation of Peace (COPAZ) established by the Chapultepec Peace Agreement of 1992 signed between the government of El Salvador and FMLN or the Joint Ceasefire Commission established as a part of the peace process between the government of Burundi and CNDD-FDD. Commission variable records whether or not one of the two types of commissions was established as a part of the process, while Commission Active variable is a categorical one that records how active a given commission was throughout the implementation process. This variable takes a value of (0) if there was not any commission, or (1) if there was one or more commissions envisaged in the agreement and the commission is established but did not go beyond first meeting and/or never become functional, or (2) if commission is established and held sporadic meetings but, did not complete its specified mandate and (3) the commission is established and held regular meetings and completed its mandate. The expectation is that presence of domestic commissions and their activities would positively correlate with the level of implementation. Inclusion of binary commission variable, as opposed to the categorical version did not change the results presented in the next section. Therefore, only the model with the categorical version (Commission Active) is presented in the Table 2. Finally, the statistical analysis takes into account time. It is likely that the longer the peace process lasts, the more parts of the agreement will be implemented. However, it is also likely that if parties are willing to implement the agreement and have means to enforce and oversee it, they will be also be likely to complete the implementation process swiftly. In turn, cases in which parties 24

25 do not have the incentive to implement and/or do not face pressures internally and/or externally are more likely to last longer and result in lower levels of implementation. For these reasons both the Total Peace Years and a square of the variable is included in the statistical analysis. In addition to these control variables, I run separate analysis including variables that capture type and intensity of conflict. Type of incompatibility, i.e. over territory or government as defined and coded by UCDP Peace Agreement Dataset (Högbladh 2011) is included since, various studies show (e.g. Walter 2009, Cunningham 2011 and 2013) dynamics and resolution of territorial conflicts are different. Additionally, number of parties in a conflict and intensity of conflict (e.g., Cunningham 2006, Zartman 2001) are linked longer wars and less durable peace. I include the total number of dyads in an agreement and a dummy variable measuring the intensity of conflict, (0) for minor conflicts in which battle related deaths were recorded for each year of conflict, and (1) for major wars in which 1000 or more battle related deaths for at least one year of conflict was recorded by the UCDP dyadic dataset (Harbom, Melander and Wallensteen 2008). None of these variables altered the main results presented in the following section. 5. Results This section discusses the results from the statistical analysis and explains the substantive impact of several variables, including the main variable of interest, length of negotiations, on the level of implementation. Descriptive stats are presented in the Section 7, Table 1. Table 2 presents 6 models. Each model is an Ordinary Least Square regression and the dependent variable is level of implementation. Although the dependent variable is a proportional variable that ranges from 0 to 1 and the OLS might technically predict values that go above or below 0 to 1 range, the majority of the data falls in to the middle ( ) range and is normally distributed, therefore OLS does 25

26 not pose any estimation problems (Long 1997). The unit of analysis is peace process (final agreement dyad). Robust standard errors are clustered on dyad to account for non-independence across observations. [Table 2 Here] As hypothesized, the total number of negotiation rounds (Model 1) is a significant predictor of level of implementation. Similarly, the total number of negotiation days (Model 2) is statistically significant in the expected direction. These two variables are not only statistically significant, but also, as shown in Figure 1 and 2 substantively significant in increasing the level of implementation. Going from 1 round of negotiation (10 th percentile) to 17 rounds (90 th percentile), the level of implementation increases from 0.40 to 0.69 (72% increase), while all other explanatory variables are set to their observed values (Hanmer and Kalkan 2012). Also, going from a total of 3 day negotiations (10 th percentile) to 124 days (90 th percentile) increases the level of implementation by 71% ( ). Across all other models, different measures of negotiation lengths are statistically significant with p-values below conventional levels (p-value <0.01 or <0.05). This provides preliminary support for the main hypothesis presented in this paper. Turning to the discussion of control variables, comprehensive agreement is also statistically significant across all models. However, substantive significance seems to be slightly weaker compared to the pre-agreement factors, i.e. length of negotiations. Implementation level of a comprehensive agreement %38 higher than the implementation level of an agreement that is not comprehensive. Guaranteed Power Sharing provision is not statistically significant. Having more provisions regarding guaranteed power sharing does not seem to affect the level of implementation. Same is true for Detailed Agreements. More detailed agreements do not have higher levels of implementation. Presence of timeline is statistically significant in 4 out of 6 26

27 models. This variable is left out in Model 4 and 5, since the significance level dropped below conventional levels. While these findings do not suggest a strong and consistent relationship, the level of implementation of an agreement is %30 higher when a timeline is built into the agreement. While both the presence of third parties during negotiation and implementation stages are in the expected direction, only third party presence (negotiation stage) is significant across all models but Model (5). While there are various robust findings showing a strong relationship between peace keepers and duration of peace, it is not surprising that the presence of peace keepers does not directly affect the level of implementation. While peacekeepers prevent conflict recurrence and indirectly influence the peace process, the factors that make degree of implementation higher or lower are likely to be closely related to incentives and political will of the parties to the agreement, rather than the third parties. Additionally, peace keeper might be effective in enforcing certain parts of an agreement, such as demobilization and/or demilitarized zones, while their effect on the overall implementation level might not be present. Turning back to the third party presence (negotiation stage), the level of implementation when there were no parties present in the negotiation stage is 0.62, compared to 0.49 when there were third parties present in all negotiation rounds (26% decrease). While having an actively functioning domestic commission increases the level of implementation by 48% compared to not having a domestic commission and the variable is statistically significant at (p< 0.05) level as shown in Model (4). Model (5) includes Total Peace Years variable. As expected, time seems to have a curvilinear relation to the level of implantation. Opposite signs on coefficients of Total Peace Years and the squared term as well as the Figure 4 supports this expectation. This relation will be further examined in the future iterations. Addition of time, however, does not change the effect of the length of negotiations on the level of implementation. 27

28 Finally, Model (6) uses an alternative measure of negotiation length, days of negotiations per provision. As discussed above, absolute value of total negotiation days might be a product of higher number of agenda items, and does not necessarily capture serious negotiations between parties. When parties spend approximately one day per provision, the overall level of implementation is 0.47, compared to 0.62 when they spend 10 days per provision (30% increase). 6. Conclusion This paper argues that the pre-agreement factors, i.e. length of negotiations, is an important factor explaining the agreement implementation. The main arguments is that the expectations about the implementation stage would shape the way combatants negotiate. If the parties expect that the implementation of a given agreement will be successful, than they would spend more time negotiating to reach a better deal. This argument is tested using original data on 80 peace process between 1989 and IPAD data records the level of implementation for all provisions in a given final peace agreement. Looking at the total days of negotiations, as well as negotiation days per provision, statistical analysis shows that the degree of implementation will be higher for cases when parties spend more time negotiating. At the face value, a negotiation process might look unsuccessful because parties fail round after round to agree on important issues to end the conflict and reach an agreement, the argument and the preliminary results in the paper suggest that the end result might be positive. Several next steps will follow the current analysis. First one is to explore the link between level of implementation and peace process outcomes. When and why do parties sometimes return to fighting in the face of failed implementation and other times accept the gains from a partial 28

29 implementation and do not challenge? An additional question to be addressed is whether or not different types of third party involvement at the negotiation stage has a different impact on peace process outcomes and levels of implementation? Moreover, additional in-depth case study analysis is underway to support the findings presented in this paper that focuses on several negotiation processes and the content of negotiations. 29

30 7. References Beardsley, Kyle The Mediation Dilemma. Cornell Studies in Security Affairs. Ithaca, NY: Cornell University Press. Beardsley, Kyle, and Nigel Lo Third-Party Conflict Management and the Willingness to Make Concessions. Journal of Conflict Resolution 58 (2): Call, Charles T Assessing El Salvador s Transition from Civil War to Peace. In Stephen J. Stedman, Donald Rothchild and Elizabeth M. Cousens eds. Ending Civil Wars: The Implementation of Peace Agreements. Lynne Rienner. Cunningham, David E Veto Players and Civil War Duration. American Journal of Political Science 50 (4): Cunningham, Kathleen Gallagher Divide and Conquer or Divide and Concede: How Do States Respond to Internally Divided Separatists? American Political Science Review 105 (02): Cunningham, Kathleen Gallagher Actor Fragmentation and Civil War Bargaining: How Internal Divisions Generate Civil Conflict. American Journal of Political Science 57 (3): DeRouen, Karl, Mark J. Ferguson, Samuel Norton, Young Hwan Park, Jenna Lea and Ashley Streat-Bartlett Civil war peace agreement implementation and state capacity. Journal of Peace Research 47(3): Derouen, Karl JR., Jenna Lea and Peter Wallensteen The Duration of Civil War Peace Agreements. Conflict Management and Peace Science 26(4): Fearon, James Bargaining, Enforcement, and International Cooperation. International Organization 52(2): Fortna, Virginia Page Peace Time: Cease-Fire Agreements and the Durability of Peace. Princeton University Press. Gent, Stephen E Relative Rebel Strength and Power Sharing in Intrastate Conflicts. International Interactions 37(2): Hampson, Fen Osler Nurturing Peace: Why Peace Settlements Succeed or Fail. United States Institute of Peace Press. Hanmer, Michael J., and Kerem Ozan Kalkan Behind the Curve: Clarifying the Best Approach to Calculating Predicted Probabilities and Marginal Effects from Limited Dependent Variable Models. American Journal of Political Science 57 (1):

31 Harbom, Lotta, StinaHögbladh, and Peter Wallensteen Armed Conflict and Peace Agreements. Journal of Peace Research 43(5): Harbom, Lotta, Erik Melander and Peter Wallensteen Dyadic Dimensions of Armed Conflict, Journal of Peace Research 45(5): Hartzell, Caroline A Explaining the Stability of Negotiated Settlements to Intrastate Wars. Journal of Conflict Resolution 43(1): Hartzell, Caroline, and Matthew Hoddie. 2003a. Institutionalizing Peace: Power Sharing and Post-Civil War Conflict Management. American Journal of Political Science 47(2): Hartzell, Caroline, and Matthew Hoddie. 2003b. Civil War Settlements and Implementation of Military Power Sharing Arrangement. Journal Peace Research 40(3): Hartzell, Caroline A., and Matthew Hoddie Crafting Peace: Power Sharing and the Negotiated Settlement of Civil Wars. Penn State University Press. Hoddie, Matthew, and Caroline A. Hartzell Strengthening Peace in Post-Civil War States: Transforming Spoilers into Stakeholders. University Of Chicago Press. Högbladh, Stina, Peace agreements Updating the UCDP Peace Agreement dataset, in Pettersson Therése & Lotta Themnér (eds.), 2012, States in Armed Conflict 2011, Uppsala University: Department of Peace and Conflict Research Report 99. Jarstad, Anna K. and Desiree Nilsson From Words to Deeds: The Implementation of Power-Sharing Pacts in Peace Accords. Conflict Management and Peace Science 25: Joshi, Madhav, Erik Melander, and Jason Michael Quinn Sequencing the Peace How the Order of Peace Agreement Implementation Can Reduce the Destabilizing Effects of Post-Accord Elections. Journal of Conflict Resolution, March, Joshi, Madhav, Jason Michael Quinn, and Patrick M. Regan Annualized Implementation Data on Comprehensive Intrastate Peace Accords, Journal of Peace Research, June, Long, J.S Regression Models for Categorical and Limited Dependent Variables. Sage Publishing. Mattes, Michaela, and Burcu Savun Information, Agreement Design, and the Durability of Civil War Settlements. American Journal of Political Science 54(2): Ottmann, Martin, and Johannes Vüllers The Power-Sharing Event Dataset (PSED): A New Dataset on the Promises and Practices of Power-Sharing in Post-Conflict Countries. Conflict Management and Peace Science, July,

32 Pillar, Paul. R Negotiating Peace: War Termination as a Bargaining Process. Princeton: Princeton University Press. Quinn, J. Michael, David T. Mason and Mehmet Gurses Sustaining the Peace: Determinants of Civil War Recurrence. International Interactions 33(2): Regan, Patrick M., and Aysegul Aydin Diplomacy and Other Forms of Intervention in Civil Wars. Journal of Conflict Resolution 50 (5): Reid, Lindsay Finding a Peace That Lasts Mediator Leverage and the Durable Resolution of Civil Wars. Journal of Conflict Resolution, October, Reiter, Dan How Wars End. Princeton. Princeton University Press. Rothchild, Donald Settlement Terms and Post Agreement Stability. In Stedman, Stephen John, Donald Rothchild, and Elizabeth M. Cousens Eds. Ending Civil Wars: The Implementation of Peace Agreements. Reprint. Lynne Rienner. Stedman, Stephen John, Donald Rothchild, and Elizabeth M. Cousens Ending Civil Wars: The Implementation of Peace Agreements. Reprint. Lynne Rienner,. Svensson, Isak Who Brings Which Peace? Neutral versus Biased Mediation and Institutional Peace Arrangements in Civil Wars. Journal of Conflict Resolution 53 (3): Toft, Monica Duffy Securing the Peace: The Durable Settlement of Civil Wars. Princeton University Press. Walter, Barbara F Designing Transitions from Civil War: Demobilization, Democratization, and Commitments to Peace. International Security 24 (1): Walter, Barbara F Committing to Peace the Successful Settlement of Civil Wars. Princeton, N.J: Princeton University Press. Werner, Suzanne, and Amy Yuen Making and Keeping Peace. International Organization 59 (02): Werner, Suzanne The Precarious Nature of Peace: Resolving the Issues, Enforcing the Settlement, and Renegotiating American Journal of Political Science 43 (3): 912. Zartman, I. William Ripe for Resolution: Conflict and Intervention in Africa. Oxford University Press. Zartman, I. William The Timing of Peace Initiatives: Hurting Stalemates and Ripe Moments. Global Review of Ethnopolitics 1 (1):

33 33

34 8. Tables and Figures Table 1 Descriptive Statistics Statistic N Mean St. Dev. Min Max Level of Implementation (ratio) Total Number of Negotiation Rounds Total Number of Negotiation Days Days of Negotiations per Provision Guaranteed Power Sharing (ratio) Comprehensive Agreement Timeline Third Party Presence (% of rounds) (Negotiation Stage) Max Number of Peacekeepers (Implementation Stage) 75 3, , ,278 Detailed Agreement (ratio) Domestic Commission (categorical) Total Peace Years

35 8.2. Table 2 Main Results OLS Regression Dependent Variable: Level of Implementation (1) (2) (3) (4) (5) (6) Negotiation Rounds (logged) *** 0.115*** (0.039) (0.042) Negotiation Days (logged) *** * ** (0.026) (0.028) (0.019) Days of Negotiations per Provision (logged) ** (0.038) Guaranteed Power Sharing (ratio) Detailed Agreement (0.203) (0.202) (0.214) (0.281) Domestic Commission (categorical) ** (0.028) Total Peace Years (logged) *** (0.232) Total Peace Years Square (logged) (0.076) Comprehensive Agreement ** ** 0.143* ** *** ** (0.076) (0.075) (0.081) (0.071) (0.053) (0.080) Timeline ** * 0.158** ** (0.075) (0.078) (0.080) (0.076) Third Party Presence (Negotiation Stage) * ** ** * * Max Number of Peacekeepers (Implementation Stage) (logged) (0.085) (0.085) (0.097) (0.085) (0.065) (0.090) (0.011) (0.011) (0.011) (0.011) (0.008) (0.010) Constant ** ** 0.334* *** *** *** (0.099) (0.105) (0.185) (0.087) (0.152) (0.092) N R Adjusted R Robust standard errors are clustered on dyad * p <.1; ** p <.05; *** p <.01 35

36 8.3. Figures 36

37 37