Peace Negotiations in Civil Conflicts: A New Dataset

This article introduces the Peace Negotiations in Civil Conflicts (PNCC) dataset, which identifies whether a state-party and rival non-state armed group is in the formal negotiation phase of a peace process. The PNCC goes beyond recording instances of peace talks by offering a conceptual framework to identify when a government-rebel group dyad is at risk of formal negotiations. It explicitly considers issues central in event-history modelling, including censoring and the observation period. The PNCC also provides detailed information on negotiations, including the date and location of peace talks, and whether negotiations were bilateral or through mediation. The PNCC is the first source to distinguish mediated and non-mediated civil conflict negotiations under a single framework. Structured over the UCDP/PRIO Armed Conflict Dataset with global coverage for 1975-2013, the PNCC is integrable to commonly used civil war datasets.


Introduction
Unlike interstate warfare, civil conflict takes place among asymmetrical units; one side is a state-party whereas the other is a non-state armed entity. Such an asymmetry emerges as a major obstacle against the peaceful resolution of conflict by predisposing state-parties to categorically refuse to negotiate with internal armed challengers, even though negotiations are necessary for peaceful resolution of conflict (Darby and Mac Ginty 2000). Unwilling to grant recognition and legitimacy, governments often brand non-state armed groups as mere bandits, criminals or terrorists, and denounce negotiations (Bapat 2005;Kirkpatrick 2018;Asal et al. 2019). This unwillingness indicates that engaging in explicit bargaining requires overcoming important obstacles. For example, during the very first negotiation process between Colombia and the FARC rebels, the government interlocutors asserted that holding peace talks were already an extraordinary step: "the most significant concession that [the President] is making is to negotiate with an organization such as yours that has taken up arms against the state. This [until now] has been the principal obstacle, and which in any other country or under any other president would make an official agreement, or even the hint of a negotiation impossible (quoted in Chernick 1988, 70)." After considering the costs of holding peace talks in civil wars, Kaplow (2016, 45) proposes that there "may be some situations in which warring parties would agree to a peace deal if only they could overcome barriers to negotiation". Yet, the predominant research paradigm conceptualizes civil conflict as a bargaining failure (see Walter 2009;Cederman and Vogt 2017) without recognizing the fact that the belligerents often do not even engage in any direct communication at all. Such a conceptualization brings an assumption that a bargaining process is always ongoing between the conflict parties. This assumption is problematic because Schelling's (1960) crucial distinction between explicit and implicit bargaining is lost. Explicit bargaining refers to negotiations, i.e. verbal communication to make a joint decision through compromise, whereas implicit bargaining does not involve communication through language, but takes place via move and counter move (Schelling 1960;Pruitt 1981). Negotiations have a profound impact on civil war processes. They alter incentives and opportunities for violence (Höglund 2008;Sisk 2009;Ruhe 2021); interact with other conflict management instruments including peacekeeping and third-party intervention (DeRouen and Chowdhury 2018; Beardsley et al. 2019;Clayton and Dorussen 2022;Owsiak et al. 2021); and influence the cohesiveness of rebel groups and decrease the barriers to entry (Lounsbery and Cook 2011;Fjelde and Nilsson 2018;Duursma and Fliervoet 2021). Yet, the systematic study of peace talks remains underdeveloped (Cunningham and Sawyer 2019).
This article aims to address such an important gap by introducing the Peace Negotiations in Civil Conflicts (PNCC) dataset. Structured over the dyadic UCDP/PRIO Armed Conflict Dataset (Melander et al. 2016) and with global coverage of 1975-2013, the PNCC offers both a disaggregated (negotiation event) and an aggregated (dyadyear) version. The event version records instances of peace talks between a state-party and a non-state armed group. It collates detailed information on negotiations, including their exact date and location, whether a codified document issued after talks, and the identity of mediator(s) when a third-party is involved. Using the information on instances of negotiations, the dyad-year version captures whether a state-party and a rival non-state armed group are holding peace talks in a given calendar year.
Distinguishing mediated and non-mediated (i.e. bilateral) peace talks in a single dataset is a unique feature of the PNCC. This comprehensive approach to record both types of peace talks under a unified framework is an important step forward to model the data generation processes implied by the leading theories of conflict resolution. As Bercovitch (2006, 290) argues, "mediation is, in essence, a form of assisted negotiations." Yet, the literature has overwhelmingly focused on mediation at the expense of bilateral negotiations (Lounsbery and Cook 2015). Further research can be expanded in two interrelated areas. First, exclusively focusing on one type of negotiations while neglecting the other may generate sample selection issues and lead to an incomplete analysis of civil conflict resolution. For example, some countries such as India, Bangladesh and Myanmar are historically sceptical towards third-party mediators when they negotiate with non-state armed challengers, mostly due to sovereignty concerns. These examples further illustrate the need to divorce the willingness to negotiate from the willingness to accept mediation (Melin and Svensson 2009). Considering that the determinants of negotiations is a central question (Greig and Regan 2008;Melin et al. 2013;Clayton and Gleditsch 2014;Keels and Greig 2019;Arı 2022), a logical next step is to differentiate cases that denounce negotiation altogether from cases that are willing to negotiate but reject third-party mediators.
Second, an emerging literature emphasizes that conflict management attempts -including mediation, bilateral negotiations, and peacekeeping-are temporally and spatially interdependent (DeRouen and Chowdhury 2018; Beardsley et al. 2019;Clayton and Dorussen 2022;Owsiak et al. 2021). Their occurrence and effectiveness depend not only on their temporal sequence, but also on the interactions between them. Although different conflict resolution instruments have often been studied in isolation so far (Diehl and Regan 2015), this growing literature underscores the need to expand our efforts towards examining conflict management trajectories as a whole to uncover such crucial interdependencies . Indeed, different instruments to manage conflict may be used concurrently, or follow each other in a temporal sequence. For example, bilateral negotiations may act as a precursor to mediation (or vice versa). Similarly, mediation may pave the way for peacekeeping. The PNCC contributes to this emerging debate by providing information that expands our understanding of conflict management trajectories. As shown later in this article, around 90% of UN peacekeeping operations are preceded either by bilateral or mediated peace talks between the belligerents.
Beyond recording peace talks, the PNCC also offers a conceptual framework to identify when a government-rebel group dyad is at risk of formal negotiations. This conceptual framework explicitly considers issues central in event history modelling, namely case-entry, exit and re-entry; censoring; and the observation period (see Box-Steffensmeier and Jones 2004;Cleves et al. 2008). This article presents the procedure with which the PNCC framework distinguishes different types of no recorded negotiations. It also unpacks how such an approach overcomes sample selection problems that have not been addressed in existing research yet.
The rest of the article is structured as follows. We start with discussing existing data sources to explain the rationale behind a new dataset. Distinguishing features of the PNCC are highlighted before continuing with a comprehensive discussion of the underlying conceptual framework and data structure. After illustrating spatial and temporal trends in negotiations, we also show that peace talks and the deployment of UN peacekeeping missions are interrelated, as the latter often follows the former. To demonstrate the usefulness of the data, two exploratory applications are presented. In conclusion, how the PNCC can help bridging the divide between peace and conflict research is discussed.

Why a New Dataset?
Although earlier studies made significant advancements in collecting data on civil war peace processes, these efforts were either limited in their spatial and temporal coverage, or remained largely ad hoc in their approaches, designed to address a particular research question without an explicit aim to construct a comprehensive dataset. Walter (2002) is an early effort to conceptualize and measure a peace process through stages, but this source is at an aggregate level with limited temporal coverage and its integration to recent and disaggregated civil war data structures is not attainable with ease. Cunningham and Sawyer (2019) and Bapat (2005) rely on the UCDP Conflict Encyclopedia and the Minorities of Risk narratives, respectively, but these secondary sources are not systematic in their reporting of negotiations. Ogutcu-Fu's (2016) study covers countries with legislative bodies, but inevitably excludes cases that are outside of its scope condition. Thomas (2014) is limited to African conflict episodes with at least 25 battle deaths, whereas the Steps to Peace (Pearson et al. 2011) focuses exclusively on Asia-Pacific. The Civil War Mediation Dataset, hereafter the CWM (DeRouen et al. 2011), is the most comprehensive and widely used source to date, but bilateral peace talks without the involvement of a third-party mediator are out of its scope. 1 As these examples demonstrate, a comprehensive dataset that builds upon earlier studies and binds them together under a common conceptual framework while using a systematic and rigorous data collection procedure is very much needed. We highlight four main benefits of the PNCC before unpacking its features in detail.
The standard approach in studying civil conflict places violent episodes at the centre of its scope. A violent episode is identified based on a predefined casualty threshold. As Staniland (2017) argues, such a focus on violence is problematic for analysing interactions between an armed non-state actor and a state-party because key features, including negotiations, may take place during periods in which such a casualty threshold is not reached. Peace and conflict often emerge as interdependent stages of a single process (Chiba et al. 2015). This interdependence is particularly pertinent to studying conflict resolution because violence may cease or escalate precisely because of ongoing peace talks (Höglund 2008;Sisk 2009).
Following this line of thought, the PNCC explicitly considers the data-generation process behind peace negotiations by relying on theories of civil conflict and its resolution (Darby and Mac Ginty 2000;Sisk 2009). Recognizing that the level of violence is often a function of ongoing conflict resolution attempts, the PNCC framework first decouples negotiations and violent episodes, and then follows a government-rebel dyad for its full lifespan. As a result, the PNCC identifies the observation period in a theoretically driven manner with careful consideration for caseentry, exit and re-entry. Here, the observation period refers to the duration within which a government-rebel group dyad remains at risk of negotiations. Identifying when the units of a study are at risk of the event of interest -negotiations in this case-is crucial for analysis (see Cleves et al. 2008, 29-36). The PNCC is unambiguous regarding censoring and when a dyad is at risk of negotiations at a given period because caseentry, exit and re-entry are defined and coded by separate rules.
Through an example of the conflict between Senegal and the non-state armed actor MFDC, Figure 1 illustrates why the observation period is so crucial for studying civil war peace processes and how the PNCC minimizes the danger of selection bias. The Senegal-MFDC dyad enters the UCDP Armed Conflict Dataset (hereafter the ACD) in 1990. Note that the ACD uses a violence threshold to identify cases in the sample, and due to a low level of violence, the Senegal-MFDC dyad leaves the ACD six times until the belligerents sign a comprehensive peace deal in 2004. 2 As shown in Figure 1, only the PNCC covers the 1990-2004 period in full, whereas other sources systematically fall short on capturing negotiations that took place in years below the ACD's 25 battle-related deaths threshold, such as the years of 1991, 1996, and 2002. Note that the issue is beyond a simple disagreement in coding among different datasets; it is also about the sample coverage because observations are excluded and left undefined. Such an omission might not be problematic for substantive findings of the respective studies as they are designed to address particular research questions. However, this exclusion introduces a sample selection issue, which is a problem for studying conflict resolution in a systematic manner because the level of violence tends to correlate with negotiations. In contrast, the PNCC is unambiguous about the period in which peace talks could take place, and then provides full coverage about whether the parties negotiated or not. 3 Second, the PNCC allows distinguishing non-mediated (i.e. bilateral) and mediated peace talks. This unique feature of the PNCC opens new avenues for research. The rapidly growing mediation literature expanded our understanding of civil conflict resolution (Wallensteen and Svensson 2014), but this inevitably shifted the focus away from non-mediated negotiations (Lounsbery and Cook 2015). Following Bercovitch and Houston's (1996, 12) assertion that "mediation must in essence be seen as an extension of the negotiation process", the PNCC adopts a comprehensive approach by including both mediated and non-mediated peace talks. Although bilateral negotiation and mediation are both integrative (rather than distributive) in their approach, they are distinct instruments in the menu of conflict management strategies, and should be studied in separation, but not in isolation (Owsiak et al. 2021). Note that such a distinction does not indicate independence. Negotiation and mediation often emerge as interdependent phases in a temporal sequence. One may pave the way for the other. This is an important issue for studying conflict resolution. For example, if negotiation processes that progressed to an advanced level are more likely to invite third-party mediators, we might overestimate the effect of mediation. Conversely, if mediators are invited only if the belligerents fail to make common ground, then we might underestimate the role of mediators. Similarly, there might be factors that influence the selection of mediated or bilateral talks when the parties decide to negotiate. In this sense, the PNCC presents an important step forward towards addressing sample selection issues in studying conflict resolution. After introducing the dataset in more detail, the usefulness of making such a distinction will be demonstrated through an application.
Third, constructing a complete history of negotiations by considering historical inter-rebel linkages is crucial because conflict resolution attempts form a pathdependent trajectory in which preceding negotiations not only influence succeeding ones, but also affect the internal cohesion of the rebels. The PNCC pays special attention to such inter-rebel linkages by identifying rebel group splintering and merging. It explicitly links preceding rebel groups with succeeding ones. Note that this feature complements the other novelties of the PNCC and renders the dataset well-suited to contribute to the recent debates about the impact of peace processes on rebel group fragmentation (Lounsbery and Cook 2011;Fjelde and Nilsson 2018;Duursma and Fliervoet 2021). For example, do mediated and non-mediated peace talks differently influence the prospects of fragmentation? Can UN peacekeeping mitigate the pressures of fragmentation and deter splintering during peace processes? The PNCC can be used to address such questions and advance the literature on the negotiation-fragmentation nexus.
Lastly, the PNCC provides detailed information on negotiations, including their date and location, and whether the talks produced a codified document or not. Going back to the example in Figure 1, thanks to the information available in the event version of the PNCC, we know that Senegal and MFDC held peace talks at the Palm Grove Hotel, Banjul on 25-26 March 2002 with the mediation of The Gambia and Guinea Bissau. The availability of such fine-grained information improves the confidence of measurement, which is an important step forward because earlier studies occasionally fail to overlap, recording conflicting accounts of the very same observation. The PNCC first triangulates data from multiple sources and then implements a rigorous data collection procedure over this baseline to construct the most accurate and comprehensive source on peace talks to date.

Conceptualization and Data Structure
In essence, the PNCC records negotiations between a state-party and a rival non-state armed group. A non-state armed group is defined as an organized entity that operates outside a state's command structure and challenges the state's claim on the monopoly of the legitimate use of force to pursue political goals (San Akca 2009). A non-state armed group forms a dyad with the opponent state party.
The PNCC comes with two complementary versions, named according to their respective units of analysis: negotiation event and dyad-year. The event data records instances of negotiations between a government-rebel group dyad. It is disaggregated at the exact date and location level. 4 Such fine-grained disaggregation allows distinguishing multiple instances of negotiations in the same calendar year. The dyad-year version is aggregated to specify whether peace talks are taking place or not in a given calendar year. In other words, the dyad-year sample also contains information on lack of negotiations.
The two versions are complementary because the same procedure is applied to specify the universe of cases and temporal coverage. To identify the universe of government-rebel group dyads, the PNCC relies on the dyadic UCDP/PRIO Armed Conflict Dataset (ACD) (Melander et al. 2016). After specifying the universe of cases, the PNCC implements a theoretically-driven approach to identify when a case is at risk of transition into and out of negotiations. The PNCC, therefore, follows the dyad for the full lifespan of a non-state armed group or until a comprehensive peace agreement is reached.

Conceptualizing Negotiations as a Stage
Achieving a negotiated settlement in civil conflicts intrinsically requires a series of interdependent phases that span a period of time. This very quality renders the stage conception suitable for studying conflict resolution (Darby and Mac Ginty 2000;Walter 2002;Pearson et al. 2011;Findley 2013). The PNCC follows this approach and identifies five distinct stages, as shown in Table 1. The objective of the PNCC is to determine whether a conflict dyad is in the Formal Negotiation stage (N).
Negotiation is defined as "a process by which a joint decision is made by two or more parties. The parties first verbalize contradictory demands and then move toward agreement by a process of concession making or search for new alternatives" (Pruitt 1981, 1). A central feature of negotiation is explicit bargaining through verbal communication, which is different from "tacit bargaining in which communication is incomplete or impossible" (Schelling 1960, 53). During tacit bargaining, which is assumed to be always ongoing according to bargaining models of war, parties communicate nonverbally, "by a process involving move and countermove" (Pruitt 1981, 4). The PNCC captures the distinction between tacit and explicit bargaining.
The PNCC operationalizes Formal Negotiation as an act of explicit bargaining between a state-party and a non-state armed group that is recognized by both parties and that is held to resolve the conflict fully or partially through verbal communication. This definition puts emphasis on recognition to distinguish Secret/Backchannel Talks from Formal Negotiation. As Darby and Mac Ginty (2000) highlight, it is common in civil conflicts that the belligerents hold some kind of communication through secret or unofficial channels before moving towards more serious negotiations. Such a stage is often needed because at least one of the belligerents hold an official position not to recognize the other as a legitimate bargaining partner with which pursuing a joint decision is possible. Negotiations that are acknowledged by both parties are significant events in civil conflicts because they entail recognition and a shared flow of legitimacy among the belligerents. As put by Bayer (2010, 536), "recognition of the other is an important milestone in relations," and "when two sides do not recognize each other's existence, reconciliation is clearly not part of the picture." Table 1 shows that No Contact (C) and Secret/Backchannel Talks (S) can lead to Formal Negotiations (i.e. C/S → N transition). Once a conflict enters into Formal Negotiations, talks can either fail (N → C/S) or succeed to produce a peace agreement (i.e. N → P). The main objective of the PNCC is to record whether a dyad is in the

Transitions into Formal Negotiation
The procedure to distinguish Formal Negotiation relies on the acknowledgement of verbal communication criteria. Note that in secret talks parties do not recognize that any sort of contacts are taking place. For example, in a speech in the House of Commons, the British Prime Minister John Major declared that "[negotiating with the provisional IRA or Sinn Fein] would turn my stomach …we will not do it" (quoted in Reiss 2010, 70). However, information surfacing years later revealed that the British government in fact was talking to the provisional IRA at the time of the speech. In stark contrast, the admission of the Sinn Fein to peace talks on 15 September 1997 or the following meeting between Tony Blair and the Sinn Fein leadership on 13 October are historical events broadcasted on television. Most formal negotiations are scheduled in advance and reported while they are taking place. It is often possible to observe the exact date and venue of negotiations and even to identify the interlocutors involved. In contrast, Secret/Backchannel Talks are reported ex-post and with a significant delay, even if any reports surface at all. Detailed information such as interlocutor identity and the date and venue of negotiations are seldom available. In some cases, even the interlocutors are not entirely sure of the identity of their negotiating partner. For example, the first known contact between Spain and the ETA took place in 1975 through a special envoy of King Juan Carlos I, but the ETA negotiators were not sure whether their counterpart was really representing the King (Clark 1990). Such clandestine contacts without clear acknowledgement from negotiating partners are distinct from Formal Negotiations. A mutual recognition criterion is therefore necessary for identifying formal negotiations as a distinctive phase in a conflict resolution process with great accuracy across cases. Although the PNCC framework conceptually distinguishes No Contact (C) and Secret/Backchannel Talks (S) as separate phases, observationally separating them for measurement purposes is not achievable in a systematic manner. As a result of this limitation, the PNCC identifies transitions into and out of Formal Negotiation (N) stage only. Collecting data on transitions to and from Secret/Backchannel Talks (S) would be ideal, but such transitions are rarely observable. More importantly, observing secret talks is dependent on systematic factors such as journalistic coverage or rules of secret services regarding the release of historical documents, which greatly vary across cases. Distinguishing Secret/Backchannel Talks from Formal Negotiation thus addresses a potential measurement error.

Transitions out of Formal Negotiation
A conflict within the Formal Negotiation stage can transition out through two mutually exclusive outcomes; negotiation failure (i.e. N → C/S transition) or success (i.e. N → P transition). Negotiation failure happens when parties continue to exist but stop holding peace talks without reaching a comprehensive peace agreement. 5 The level of violence during peace talks is not a condition to define a transition. As such, a dyad does not necessarily transition out of Formal Negotiation if violence erupts between the parties. This approach is different from earlier conceptualizations that required parties to negotiate in good faith or not to use systematic violence during the negotiations (Darby and Mac Ginty 2000;Walter 2002). To have a rigorous definition of a negotiation phase that can be coded across cases with minimal discretion, the PNCC deviates from such an operationalization. Note that measuring negotiations conditional on the level of violence would be misleading because some peace talks continue and progress despite ongoing violent episodes. Moreover, the two are correlated: conflict parties may use violence to push their rivals to talks and gain leverage at the negotiating table. However, users can incorporate violence and such other variables to generate a user-defined negotiations phase.
If parties reach a comprehensive peace agreement, the process moves towards the Implementation phase, which entails an N → P transition. Signing of a comprehensive accord does not mean that the conflict is settled, but rather shows that the peace process progressed out of the Formal Negotiation stage and entered into Implementation. When such an N → P transition takes place, the dyad leaves the PNCC sample. For example, revisiting Figure 1, Senegal and MFDC signed a comprehensive peace deal in 2004 and entered into the Implementation Phase (Joshi et al. 2015), thus exiting the PNCC.
Comprehensive peace agreements may collapse and parties may return to the battlefield. Similarly, any peace agreement can be re-negotiated. If a peace accord fails and the dyad becomes active in the ACD after the agreement, it is conceptualized as a re-entry to the PNCC. We next turn to the observation period to discuss case entry, exit and re-entry.

Observation Period: Entry, Exit and Re-entry
To identify actors in an internal armed conflict, the PNCC relies on the dyadic version of the ACD, but the observation period is defined separately, and based on whether a dyad is at risk of transition into or out of Formal Negotiation. 6 After non-state armed groups are identified, the dyads are followed until either (1) the rebel group is ceased to exist as a non-state armed group, or (2) a comprehensive peace agreement is signed between the parties. When a dyad exits the study, the reason is coded as a separate variable. Distinguishing the exit type allows users to identify why the case is no longer in the sample and whether the observation is censored.
As highlighted in the previous section, this structure is an important feature separating the PNCC from other data sources. To further unpack this structure, borrowing from the event history modelling terminology is helpful. Note that units in the PNCC sample are considered to be in one of the two types of risk sets. These two type of risks are (1) transition into Formal Negotiation (i.e C/S → N) and (2) transition out of it (either through N → C/S or N → P). If a dyad is no longer at risk of such a transition, it exits the study. If a dyad exits the study while being at risk of a transition, this is understood as censoring. We next turn to each of these scenarios.

Case Exit
A dyad can leave the study if the non-state armed group ceases to exist. This can happen by victory, defeat or merging with another non-state actor. The definitions of rebel victory and transformation to another non-state actor overlap with the UCDP Conflict Termination dataset (Kreutz 2010). A rebel group is defeated if there is no indication that the non-state actor continues to exist as an armed entity. If there is an indication that a non-state armed group continues to exist despite the dyadic conflict is below the violence threshold to be included in the ACD, the PNCC keeps following the dyad and the case remains in the study. In this respect, the PNCC decouples violence from the observation period. As discussed, using a conflict intensity threshold to force cases out of the study may generate sample selection bias because violence and peace talks are interdependent. Some dyads might be in conflict lulls precisely because of ongoing negotiations. The structure of the PNCC handles this hindrance.
A dyad can also exit the PNCC with a comprehensive peace agreement. Once a comprehensive peace agreement is coded, the dyad is understood to be in the Implementation stage, and thus no longer at risk of peace talks. 7 This conceptualization requires distinguishing interim agreements from comprehensive accords that cause transition into Implementation, as the latter prompts case exit whereas the former does not. The PNCC relies on the UCDP Peace Agreements Dataset (Högbladh 2012) and the Peace Accords Matrix (PAM) (Joshi et al. 2015) to identify comprehensive peace agreements. When these two sources do not overlap, the guiding criteria is that if the accord includes a provision to disarm the non-state actor or subject the group under state control, such as transformation into a local security force or integration into the military, the agreement is understood to be a comprehensive deal. Interim agreements short of a comprehensive accord and joint communication releases are coded separately to indicate that negotiations produced a codified document. This presents strong evidence that the dyad is indeed in Formal Negotiation.
Finally, a dyad can be still at risk of transition into or out of Formal Negotiation outside of the PNCC's temporal range. This is understood as censoring and identified in the PNCC. Although censoring is central in the event history analysis, the PNCC is the first data source on negotiations to unambiguously distinguish censored observations from other types of case exit.

Case Re-entry
A dyad that had exited the study may re-enter if the non-state actor re-emerges as an armed entity challenging the state-party. There are two main paths for re-entry. First, comprehensive peace agreements can be fully abandoned, and parties may return to the battlefield. In such cases, the dyad re-enters the PNCC with a new entry date. An example of this is the Mozambique and RENAMO conflict. The belligerents signed a comprehensive peace agreement and exited the PNCC in 1992, but the case re-entered after parties fell back into civil conflict in 2013 and the dyad became at risk of Formal Negotiations.
Second, a non-state armed group that had acquired state power through victory can be deposed and return to be classified as a non-state actor challenging the state. A prominent example of this pattern is the Taliban, which had acquired state power through victory in 1996, forcing the Afghanistan-Taliban dyad out of the PNCC. The Taliban was eventually deposed and returned to be a non-state actor in 2001, prompting case re-entry.

Procedure for Data Collection
Data collection is carried out through three steps: (1) mapping other data sources and the case literature; (2) relying on copies of primary sources such as documents that are released after negotiations (e.g. interim peace agreements and codified joint declarations); (3) a systematic search on the Nexis database.
The data collection starts with information available in the UCDP Conflict Encyclopaedia and other secondary sources such as the Accord Series by the Conflict Resources. These sources are employed to map out the conflict. Reported negotiations are noted but not coded in this step. They are used as a guideline for the Nexis search in step three. Information available in other important datasets are also considered (Kreutz 2010;DeRouen et al. 2011;Pearson et al. 2011;Högbladh 2012;Thomas 2014;Ogutcu-Fu 2016). These datasets differ in their structure as well as their geographical and temporal coverage, but the information they contain is integrated into the PNCC structure when possible. Note that a positive coding in another dataset is not sufficient for a positive coding in the PNCC, as verification is needed in the second and third steps.
In the second step, the UN Peacemaker database is used. This database includes digital copies of primary sources such as various agreements and press releases after negotiations. Each of the 633 documents in this database is checked for signatories. If a document is successfully matched with a dyad (i.e. signatories include both state-party and non-state armed organization), then negotiations are coded. A successful match is considered as "smoking gun" evidence. A similar procedure is also carried out for other peace accord datasets, including PA-X (Bell and Badanjak 2019), PAM (Joshi et al. 2015), and UCDP (Högbladh 2012).
In the final step, the PNCC systematically search for the Nexis database using the following structure: ("Country Name" OR "Alternative Name") w/25 ("Non-state Actor Name" OR "Alternative Name") w/25 (negot! OR "peace process" OR talk! OR dialogue OR barg! OR meet! OR met OR mediat!) Search terms are modified based on the information gathered from the first and second steps. Names of important individuals can also be incorporated. If a party rejects the reports of peace talks, the dyad is not considered to be in Formal Negotiation. Peace talks must be between representatives of the central/federal government and non-state armed group to be formal. Negotiations involving local governments are not coded if the central government is not represented. Similarly, negotiations involving individuals from the legislative branch are not considered as peace talks if these individuals are not appointed by the central government for the purpose of conducting negotiations. Negotiations for hostage-taking situations are not considered as peace talks.
The PNCC registers the location and date of peace talks as well as whether an external party is involved to mediate. The main objective is to distinguish mediated talks from bilateral negotiations without the involvement of an external party. When multiple mediators are involved, the list of mediators is not exhaustive, but special attention is directed towards the UN. If the UN is involved in assisting peace talks, it is recorded separately. Finally, if a state-party offers negotiations but the non-state actor refuses to participate, this is also registered in a separate variable.

Presenting the Data
Starting with the basic properties of the PNCC, there are 79 countries and 305 unique non-state actors in the sample, with 22 re-entries. Victory by one side is the most common reason for case exit whereas 99 peace agreements carried dyads to the Implementation stage. 18 of these peace agreements eventually collapsed, prompting re-entry to the PNCC, which makes the accord failure the most frequent cause for reentry. The remaining four re-entries are due to rebels who had acquired state power losing governmental control and becoming a non-state entity once again.
There are 2183 instances of negotiations identified in the event sample. 63 out of 79 countries engaged with peace talks with a rebel group at some point in time. Yet, the overall rate of negotiation is 27% in the dyad-year sample, which corresponds to 688 out of 2556 observations. This low rate indicates that on average negotiations are not common and the conflict proceeds without explicit bargaining.
However, a large variation exists among cases and across regions and time periods. Figure 2 further unpacks the rate of negotiation by region and type. Europe has the highest rate of peace talks with nearly half of the cases on average remaining in the Formal Negotiation stage during their lifespan, whereas talks are the least likely in the Middle East. Russian mediation in the former Soviet Union is a driving factor behind the high proportion of non-UN mediated talks in Europe.
Turning to time trends, Figure 3 shows that the end of the Cold War brought an increase in conflict resolution attempts. The rate of dyads in Formal Negotiation peaks in 1992, when nearly half of the cases (48%) held peace talks. However, this peak is succeeded by a declining trend, and by 2010 the rate drops as low as 14%. Such a trend overlaps with the findings of Lundgren and Svensson (2020), which expose a mediation gap in the last two decades despite an increase in the supply of mediators. Note that our  ability to consider bilateral negotiations also reveals that this mediation gap is not due to an increase in non-mediated peace talks, but rather a more persistent trend in lack of conflict resolution attempts is discernable.
UN peacekeeping operations are a focal point of interest for peace and conflict research, but estimating their effectiveness remains a major hurdle due to their nonrandom assignment. To enable causal inference, researchers are often interested in uncovering selection processes that lead to the deployment of UN missions. The PNCC becomes relevant in this respect because peace talks, external mediation, and UN mission assignment are interrelated processes (Beardsley et al. 2019;Kathman and Benson 2019;Clayton and Dorussen 2022;Owsiak et al. 2021). A growing literature reveals that an overwhelming majority of conflict management attempts come as part of a package (see Greig et al. 2019). Indeed, multiple instruments are often pursued in tandem or follow each other in a sequence. These temporally linked instruments form a conflict management trajectory. Moreover, different conflict management instruments within the larger package interact with each other (Owsiak et al. 2021). Their selection and effectiveness thus depend on their sequence within the trajectory and with respect to preceding and succeeding resolution attempts.
Although earlier research tended to investigate different conflict management instruments such as mediation and peacekeeping in isolation, recent studies increasingly turned towards uncovering interactions between them (DeRouen and Chowdhury 2018; Beardsley et al. 2019;Clayton and Dorussen 2022). This is a promising area of research because conflict management attempts are interdependent in time and space (Diehl and Regan 2015). Therefore, questions such as the determinants of selecting a conflict management instrument from a larger menu or estimating their effectiveness should ideally be investigated with respect to the whole trajectory while considering sequences and interactions (Owsiak et al. 2021).
The PNCC is well-positioned to contribute to these emerging debates. Formal negotiations, especially those that are mediated by the UN, are likely precursors to UN missions. By providing data on peace processes, the PNCC helps us to uncover the temporal sequence of mediation and the deployment of UN peacekeeping missions. Figure 4 shows that 28 countries in the PNCC sample received a UN peacekeeping or political mission. Apart from Lebanon, Syria and Haiti, all UN missions preceded by negotiations involving at least one non-state armed group. However, there are different trajectories for mission deployment depending on how the UN interacts with a peace process. For example, a recurrent trajectory is that preceding UN mediation is followed by mission deployment (e.g. Burundi, Cambodia, Morocco and Nicaragua). Conversely, parties may first negotiate extensively and then invite the UN at the final stage, as happened in Mozambique (Fortna 2008). Finally, a UN-led peace process in one country can spill over to another. UN mediation in Nicaragua facilitated a peace agreement and the deployment of the MONUCA, whose mandate also included El Salvador and Guatemala. The MONUCA then enabled peace processes through UN mediation in these respective countries and paved the way for the eventual deployment of the ONUSAL (in El Salvador) and the MINUGUA (in Guatemala). The PNCC renders all these trajectories discernable by providing data on different types of peace talks, as illustrated in Figure 4. In this respect, the PNCC contributes to an emerging area of research that aims to uncover temporal and spatial interdependencies in conflict management trajectories (Owsiak et al. 2021).
Lastly, mapping the geographical distribution of peace talk venues is possible by using the information available in the negotiation event sample. As shown in Figure 5, Geneva and Nairobi are the most frequent hosts, followed by Moscow, Paris, New York, and New Delhi. Nairobi sharing the lead with Geneva as the busiest negotiation hub is an interesting finding, which overlaps with findings from an emerging literature underscoring the importance of African mediation in African conflicts (see Duursma 2020). Indeed, African states do frequently mediate and host peace negotiations for other African countries, as shown in the PNCC event data.

Illustrative Applications
Next, two applications using each version of the PNCC are presented below. The first application takes the dyad-year data to model the likelihood of peace negotiations. The second application incorporates information on negotiation locations from the event sample and explores correlations between the probability of a peace agreement and the minimum distance of negotiation venue from the capital.

Separating Bilateral and Mediated Talks
Scholars proposed several predictors of peace talks and the PNCC framework enables revisiting earlier findings by using a more comprehensive dataset that addresses previous sample selection issues. Most notably, it is possible to consider both bilateral negotiations and mediated peace talks under a single framework. As proposed variables may have divergent impacts depending on the type of negotiations (Bercovitch and Jackson 2001), the following application demonstrates the usefulness of separating mediated and non-mediated peace talks.
Three hypotheses in the literature are revisited. First, the perception of a mutually hurting stalemate is the predominant explanation of peace negotiations in civil conflicts (Zartman 1993(Zartman , 2003. According to this theory, if one or more parties expect that victory is around the corner, there is little room for compromise. However, if none of the parties anticipate advancing militarily and the continuation of conflict is costly for all, a mutually hurting stalemate emerge, which paves the way for negotiations. For operationalizing mutually hurting stalemate, a common approach is to rely on the duration and intensity of conflict (Zartman 2003;Bapat 2005). A prolonged conflict reflects a stalemate by revealing that neither party has been able to secure military victory despite years of fighting, whereas the intensity reflects the costs of conflict. Following Beardsley et al. (2019), we calculate a conflict intensity rate until the observation-year by dividing the cumulative intensity by the elapsed time.
Second, Clayton (2013) theorizes that governments have little incentives to offer negotiations to weak rebels. Testing this hypothesis on mediation data, Clayton (2013) shows that the stronger the rebels, the more likely to have peace talks. How does the rebel strength influence the likelihood of non-mediated peace talks? This is an important question due to two interrelated reasons. First and foremost, it helps us to uncover the underlying dynamics of negotiations in terms of how explicit bargaining unfolds. Second, and relatedly, it is crucial for anticipating mediation (Greig and Regan 2008;Clayton and Gleditsch 2014;Ruhe 2015). During a peace process, non-state actors are often more vulnerable than the government, especially when a third-party is not involved (Walter 1997). We can thus expect non-state actors to prefer the involvement of external actors, such as mediators and peacekeepers, whereas state-parties to repudiate when they can (Melin and Svensson 2009). In an analysis of the Chittagong Hill Tracts peace process in Bangladesh, Fortna (2008) identifies the relative weakness of the rebels for their failure to bring in external actors both during the negotiation and the implementation phases. This qualitative finding suggests that the rebel strength might have a divergent impact on the likelihood of non-mediated peace talks, as strong rebels might be more likely to attract external mediation while weak rebels settle for bilateral negotiations. For the relative rebel strength measure, we rely on the Non-State Actors in Armed Conflict Dataset (Cunningham et al. 2013) to construct a variable with three categories; (1) weaker or much weaker rebels; (2) rebels at parity; (3) stronger or much stronger rebels.
Lastly, we replicate the main finding of Keels and Greig (2019), which reveals that negotiations are positively correlated with the number of ethnic groups, but negatively with the number of excluded groups. In their main analysis, Keels and Greig (2019) estimate the likelihood of mediation in all civil wars from 1990 to 2005. We expand this analysis to 1975-2013 and separate mediated and non-mediated peace talks as different outcomes.
All models also control for population and real GDP per capita (Gleditsch 2002), the number of rebel groups in the country, the level of democracy using the V-Dem's Polyarchy index (Gleditsch 2002), the type (i.e. territorial/governmental) and intensity of conflict (Melander et al. 2016), whether a rebel group is associated with an ethnic group (Vogt et al. 2015), and cubic polynomials of the time since the last negotiation to account for temporal dependency. Table 2 presents the results. Model 1 is estimated through a logistic regression on negotiation without separating bilateral and mediated talks. Results from Model 1 is in line with all three hypotheses. For the mutually hurting stalemate indicators, both duration and intensity rate are strong predictors of negotiations. As expected, peace talks become more likely as the conflict prolongs. Similarly, a higher intensity rate is associated with an increased willingness to negotiate.

Results
In line with Clayton (2013), rebel strength is associated with peace talks. Compared to relatively weaker rebels, governments are more likely to negotiate with stronger rebels and rebels that are at parity. Regarding the total and excluded number of ethnic groups, the results also overlap with Keels and Greig (2019): the higher the number of total ethnic groups, the more likely the negotiations whereas the higher the number of excluded ethnic groups, the less likely the negotiations.
We continue by separating bilateral and mediated negotiations as different outcomes and implement a multinomial logistic regression. Our interest lies in whether the results from Model 1 change once such a distinction is introduced in Model 2. Substantive interpretation remains the same for the duration of conflict, the level of democracy, and the total and excluded number of ethnic groups. Intensity rate is still a strong predictor of mediation, but it loses significance for bilateral talks, albeit the direction of the relationship remains the same.
For the rebel strength, however, the results indicate a heterogeneous impact. As formulated by Clayton (2013), the rebel strength is positively associated with mediation, but the likelihood of bilateral peace talks decreases as the rebel strength increases, indicating an opposite impact. Figure 6 illustrates the quantities of interest. This analysis shows that the rebel strength is important not only for whether peace talks will take place, but also for the type of negotiations. In this sense, the analysis goes beyond replicating the main finding of Clayton (2013) and provides preliminary evidence that Fortna's (2008) qualitative argument for Bangladesh can be expanded to other cases.
To further unpack the divide between bilateral talks and mediation, Model 3 adopts an alternative approach and estimates the likelihood of mediation given negotiations. Model 3 can be considered as the second stage of a two-part model, in which Model 1 is the first stage. 8 In this formulation, parties first agree to hold peace talks and then decide to conduct negotiations with our without mediation. In line with the findings from Model 2, Model 3 indicates that the rebel strength is positively associated with mediation; the stronger the rebels, the more likely the negotiations involve a third-party mediator. Mutually hurting stalemate indicators are also positively associated with mediation; as the duration and intensity increase, mediation becomes more likely. On the other hand, ethnic group variables and democracy are not meaningful to explain whether the belligerents choose to negotiate with or without mediation. Finally, the results also indicate that mediation is more likely when the conflict is over territory or when the number of rebel groups is high. This exercise demonstrates that some predictors of peace talks such as the number of excluded ethnic groups (Keels and Greig 2019) have a similar impact on all type of negotiations (i.e. whether mediated or not), whereas other predictors such as the rebel strength (Clayton 2013) may have a heterogeneous impact, increasing the likelihood of one type at the expense of the other. The PNCC can be used to reveal such important differences and expand our understanding of civil war peace processes.

Negotiation Location and Peace Agreements
In the second application, we leverage the event sample to explore correlations between the location of negotiations and the probability of reaching a peace agreement. Earlier research demonstrated that geography and conflict resolution is indeed correlated; the likelihood of mediation changes when the location of fighting approaches towards the capital (Greig 2015;Ruhe 2015).
Availability of data on negotiation locations opens new avenues for research. For example, one might ask whether talks in intercontinental locations are more (or less) likely to produce peace agreements. Intuitively, the national capital is the natural venue for negotiating a comprehensive deal, especially for conflicts over governmental power. However, several important peace agreements are signed in cities thousands of kilometres away, such as the 1991 Paris Peace Agreement to end the Cambodian civil war. Talks held in particularly distant locations are likely to have high salience, with closer scrutiny and interest from the international community. On the other hand, the national capital might not be the first choice for peace talks in conflicts over secession or autonomy of a territory. For example, Senegal and the MFDC, which is a territorial conflict over the Casamance region, negotiated a peace agreement in Zinguinchor, which is the main town in Casamance.
Following this conjecture and using the PNCC event data, it is possible to explore correlations between the location of talks and the likelihood of reaching a peace agreement. Figure 7 illustrates the findings from a basic application using logistics regression. The minimum distance of negotiation location to capital is correlated for governmental conflicts with a U-shaped relationship, but no such trend is discernible for territorial conflicts. A relatively low predicted probability with a narrow confidence interval indicates that negotiations in the capital are consistently unlikely to produce a peace agreement for territorial conflicts. However, this consistency starts to disappear, and talks become more likely to produce an accord once the location of negotiations moves further away from the capital, as indicated by a steeply increasing confidence interval. This basic application is to illustrate how the event data and corresponding information on geographical location of negotiation venues can be used.
Further applications may involve analysing the relationship between conflict geographies and negotiation locations. A large number of negotiation locations available in the PNCC is in close proximity to battlefield locations and not in major conflict resolution hubs. Indeed, 374 out of 1975 negotiation locations are unique, with 154 of them appearing only once in the sample. In this sense, the PNCC is suitable to be used in conjunction with other geographically informed conflict datasets, such as the ACLED (Raleigh et al. 2010) and the UCDP GED (Sundberg and Melander 2013).

Conclusion
Although peace talks are crucial events that take place only after significant barriers are overcome, a strand of civil war literature assumes that a bargaining process is always ongoing. This inevitably generates a conceptual mismatch. On the one hand, studies on conflict resolution and peacemaking recognize negotiations as a significant step that is rarely achieved. On the other hand, studies on civil war tend to consider bargaining as an intrinsic feature of conflict that is embedded in the very act of war-making. Such a mismatch overlaps with a larger pattern of separation between the study of peace and conflict as two distinct domains. As Gleditsch et al. (2014) and Bright and Gledhill (2018) caution, a divergence without sufficient links and communication between the two domains is taking place, and the study of violence and conflict is becoming more dominant at the expense of peace and peacemaking. In line with this cautionary note, Staniland (2017) argues that the standard approach in studying civil conflict processes focuses on episodes of violence, and therefore is not appropriate for analysing the interactions between conflict actors, including peacemaking efforts.
The PNCC aims to bridge this divide by conceptualizing peace and conflict as an interrelated process. As highlighted, the PNCC does not focus exclusively on violent episodes, which is an important step forward both in conceptualization and measurement. The PNCC provides disaggregated data on peace negotiations during periods both with and without violence, and whether a third-party is involved or not. These features address possible selection issues arising from exclusively splitting peace and conflict processes (Chiba et al. 2015), and divorcing mediation and bilateral talks (Bercovitch and Houston 1996).
Researchers can use the PNCC data both as a predictor and an outcome variable to further unpack the complex interdependencies between peace-and war-making. This article suggested various ways to do so, but deriving an exhaustive account is not attainable because infinite combinations exist. Further research will keep uncovering such interactions, especially when the study of peace and conflict is explicitly linked. The PNCC aims to serve this endeavour. 5. Failure is used not in its substantive meaning, but rather refers to the termination of talks without reaching an agreement. Many talks that ended without a comprehensive agreement were vital in managing violence and saving lives. 6. Cases such as interstate wars and coups attempts that do not fall under the non-state armed group definition are excluded from the sample. 7. Negotiations in this stage are conceptualized as implementation negotiations and are out of the PNCC's scope. 8. Two part models are widely used in conflict research because of their relatively easy application and interpretation compared to alternatives such as nested logit and Heckmanselection (see Bartusevičius and Gleditsch 2019). We consider this approach as an alternative to the multinomial logit in Model 2.