Abstract
In addition to providing crucial insights, the rebel-to-party literature exhibits an unacknowledged conceptual tension: despite remarkable agreement on what ‘rebel-to-party transition’ should capture, there are nearly as many definitions and measures as there are studies of it. I demonstrate that conceptual imprecision has an analytic ripple effect—compromising the validity of the concept, the quality of the measure, the validity of inclusion criteria, and the results of analyses. Across four existing rebel-to-party variables, scholars only agree with regard to eight transitions (out of 161) and five failures (out of hundreds). To address these limitations, I propose a novel conceptualization and measure of rebel-to-party transition—distinguishing between failures, nominal participants (the conventional benchmark for transition), and seated participants. I demonstrate that some definitions of ‘failure’ induce selection effects into samples, and that minimalist indicators of ‘transition’ introduce problematic heterogeneity into ‘successes’. My analyses reveal that nominal participants are statistically indistinguishable from failures on key traits predicting transition and, moreover, seated participants consistently drive results. As such, the new conceptual framework advances the literature on conceptual and empirical grounds.
Introduction
On the heels of a 50-year conflict, scholars, policymakers, participants, and victims turned their eyes to Colombia’s 2018 elections as the (former) Revolutionary Armed Forces of Columbia (FARC) rebels began competing as the (new) FARC party. While allowing FARC to try their hand at politics was contentious, the literature on rebel-to-party transition leaves room for optimism. Evidence suggests that integrating former militants into post-conflict politics creates a tenable path toward peace, stability, and democratization (Hartzell & Hoodie, 2015; Marshall & Ishiyama, 2016; Matanock, 2017; Norris, 2008). Yet, when it comes to evaluating the success (or failure) of FARC’s attempted rebel-to-party transition, a more fundamental question looms large: how will we know it when we see it?
Notwithstanding path-breaking contributions, the rebel-to-party literature got ahead of itself on a crucial front: scholars are exploring the causes and consequences of the phenomenon of rebel-to-party transition without agreement on (or debate about) what it actually is. Yet, this is not just a story of discord. The literature exhibits an unacknowledged conceptual tension that simultaneously demands and paves the way for reconciliation.
On the one hand, scholars display a remarkable convergence on the core meaning (the ‘background concept’) and the stakes of rebel-to-party transition. 1 They agree that one outcome is paramount: rebels’ integration into legal, competitive politics. 2 Political integration, they argue, enables (former) rebels to express preferences and grievances through legal channels, thereby obviating the use of violence (Marshall & Ishiyama, 2016: 1010). This stark agreement on what ‘rebel-to-party transition’ should capture suggests a unified conceptualization is both possible and desirable.
On the other hand, the literature reveals nearly as many different definitions of rebel-to-party transition as there are studies of it. Implicit conceptual disagreements may inadvertently yield serious operational disparities. Across four cross-national datasets to test rebel-to-party theories, researchers identified 161 unique transitions, but unanimously agreed on only eight cases. 3 Thus, scholars are not only using the same concept to refer to different outcomes, they are also testing theories on very different – and differently coded – sets of cases.
Of course, conceptual disagreement is not inherently problematic; different choices may be tailored to answer different questions. However, discord can become problematic when it goes unacknowledged. If every rebel-to-party scholar silently uses a different definition, translates it into a different coding rule, and identifies a different universe of cases, then the literature is built on shifting sands. Without guidelines to adjudicate among varying approaches, scholars new to the literature will find themselves in the dark. Or worse still, when taken alongside the overwhelming agreement as to what ‘rebel-to-party transition’ should capture, new readers are liable to mistake silence for consensus.
The article is situated within the burgeoning field of conflict methodology. In recent years, conflict scholars have brought mixed-methods expertise to interrogate disconnects between the concepts central to this field, and the conventional ways those concepts have been (narrowly) defined (Gutiérrez-Sanín & Wood, 2017), (inappropriately) measured (Krüger et al., 2013), and (mis)counted (Dawkins, 2021; Hoover Green & Ball, 2019). Studying conflict often entails doing the best we can with what we have. At a certain point, however, pushing the field forward demands interrogating, testing, and resolving the conceptual discord. I show how implicit conceptual disagreements cascade through analyses, affecting theory, measurement, data collection, and inferences—irrespective of whether we use qualitative or quantitative tools. Thus, the insights derived here are relevant both within and beyond the conflict scholarship.
I begin by evaluating the conceptual and methodological trends in the rebel-to-party literature and outlining the questions left unanswered. To address these limitations, I propose a novel conceptualization of rebel-to-party transition that distinguishes between (failed) political aspirants, nominal participants, and seated participants. Building on the field’s conceptual foundations, the new framework places critical scope conditions on ‘failure’, adds nuance to ‘success’, and explicitly distinguishes between ‘transition’ and ‘transformation’ to specify the relevant boundary between rebels and parties. Next, I derive new measures and guidelines for data collection. In the empirical section, I test the implications of existing operational disparities and demonstrate the utility of the new framework. I conclude with a discussion of limitations and advantages for rebel-to-party scholarship.
Rebel-to-party transition in the literature
I evaluate the literature in terms of the micro- and macro-level conceptual stakes. Micro-level stakes refer to the conceptual content: How do scholars define positive cases, negative cases, and the boundary between them? 4 Macro-level stakes address the broader implications: Do existing definitions capture what they intend to (concept validity)? Do they translate into empirical rules for identifying an appropriate universe of cases and sorting them into informative categories (construct validity)? Would a reasonable scholar come away from the literature equipped to adjudicate among approaches and make informed decisions?
Early scholarship in the field focused on developing rich theories of the rebel-to-party process in what are now the archetypal cases (De Zeeuw, 2008; Manning, 2002; Manning, 2004; Söderberg Kovacs, 2007). As post-conflict transitions became increasingly common, scholars tackled prodigious data collection efforts in search of cross-national trends (Acosta, 2014; Manning & Smith, 2016; Matanock, 2016; Söderberg Kovacs & Hatz, 2016). Assembling large datasets, however, forced researchers to specify not only a universe of cases, but also reliable indicators of transition in less clear-cut instances. Notwithstanding valuable advances in the field, researchers take different approaches at every conceptual stage. I evaluate these approaches in turn, starting with the very nature of the transition process.
The rebel-to-party process (i.e. the boundary between ‘failure’ and ‘success’) is conceptually slippery because it has more than one locus of change. Drawing on the organizational change literature, we can distinguish rebels’ internal, organizational transformation from their external, environmental transition (Hannan & Freeman, 1989). Transformation captures the structural changes needed to forge a party organization (Allison, 2006; De Zeeuw, 2008; Ishiyama & Batta, 2011; Sindre, 2016b). Transition is the shift toward operating in a new environment (Manning & Smith, 2016; Söderberg Kovacs & Hatz, 2016). Organizational transformations and environmental transitions are related, yet they capture different processes, entail different challenges, and have different observable indicators (Barnett & Carroll, 1995). Although implicitly identified in some scholarship, 5 this distinction is never directly engaged. More often than not, the concepts and their indicators are used interchangeably, 6 undermining both clarity and concept validity. Thus, evaluating how positive and negative cases are defined first demands identifying which process scholars are engaging.
Where the goal is to capture rebels’ integration into politics, the implied focus is on their transition into the electoral system. This distinction is mirrored in the political parties literature, in which the parallel is party formation as opposed to party consolidation (Levitsky, 2003; Olson, 1998). Distinguishing transition from transformation enables future analyses to speak more directly to the political parties literature and allows scholars to examine the interaction between the functional move into politics and the organizational changes that promote (or inhibit) it.
To define positive cases of transition, most scholars use indicators of political participation. But, how much participation is enough for a transition to ‘count’, — to bring about the favorable outcomes we associate with success? Most rely on minimalist benchmarks such as party registration or electoral participation (Acosta, 2014; Manning & Smith, 2016; Matanock, 2016; Söderberg Kovacs & Hatz, 2016). Minimalist indicators have the benefit of observability across cases, but the trade-offs are significant. While both steps are important components of political integration, neither fully captures the definition of ‘parties’ on which scholars uniformly rely: ‘any political group identified by an official label that presents at elections […] and is capable of placing through elections, candidates for public office’ (Sartori, 1976: 54). 7 Additionally, minimalist definitions may introduce unexplored heterogeneity into positive cases, resulting in conceptual stretching (Sartori, 1991). Registration alone generates a set of transitions comprising rebels that registered but never ran in elections, and rebel parties with longstanding careers in office. Thus, minimalist definitions raise questions about whether groups with varying levels of integration differ systematically from each other.
Rebel-to-party scholars take varied approaches to defining ‘non-transitions’ or failures. 8 One tack, which is increasingly common among those using multipurpose datasets, involves borrowing the dataset’s (broad) inclusion criteria to serve as the de facto scope conditions for the transition variable (Acosta, 2014; Acosta, 2019; Matanock, 2016). 9 Researchers score transitions (however conceived) as 1, then fill in 0s for all remaining observations in the dataset. Thus, for Acosta and Matanock, the implied definition of ‘failure’ is ‘any militant group that did not participate in electoral politics’. This approach, however, sometimes results in comparing successes with non-viable contenders: groups lacking the intent or opportunity to transition. Groups like the Legion of Doom or the Animal Liberation Front are systematically different from those who tried to form parties and succeeded and those who tried and failed. Consequently, this conceptual decision may produce empirical selection effects.
Unfortunately, more specific definitions of failure can introduce the opposite problem: omitting viable contenders. Defining failures as ‘groups prohibited from governing’ (Van Engeland & Rudolph, 2008: 7) introduces unsubstantiated theoretical assumptions, thereby excluding highly informative comparisons. Although institutional barriers are contextually important, this definition falsely assumes failure is always externally imposed. The stakes of adequately defining ‘failure’ cannot be overstated. When we ask why some groups can transition whereas others cannot, the distinguishing traits we identify will be driven largely by how researchers define the ‘others’. Moving forward, the literature would benefit from an explicit discussion of the trade-offs involved in defining the comparison set.
A final conceptual issue concerns the role of disarmament. Although some distinguish ‘disarmament’ from ‘party formation’ to explore their relationship as a salient line of inquiry (Matanock, 2016; Matanock & Staniland, 2018), others embed ‘disarmament’ in the definition of ‘rebel-to-party transition’ (De Zeeuw, 2008; Klapdor, 2009; Söderberg Kovacs & Hatz, 2016). According to this logic, these hybrid organizations can use violence to influence or circumvent the political process, making the transition incomplete. Compelling as the argument may be, translating this hypothesis into a conceptual rule creates two problems. First, an armed wing does not prevent an organization from (also) functioning as a political party. Many parties operate with overt paramilitary wings. Therefore, this criterion imposes a double standard on political organizations without justifying how rebel parties and paramilitary parties represent differences in kind.
Second, treating ‘failure to disarm’ synonymously with ‘failure to transition’ creates ambiguous measures and prevents scholars from asking crucial questions. 10 Many rebel groups go on to have longstanding, electorally-successful careers while maintaining an armed wing. Omitting these groups from analyses prevents us from asking whether political integration affects patterns of violence. This conflation further highlights the importance of distinguishing between internal transformations (of which disarmament is one) and transition into electoral politics.
Rebel-to-party scholars have laid a crucial foundation in a high-stakes field. Yet, the overwhelming agreement on what ‘rebel-to-party transition’ should capture would lead a reasonable scholar to expect more theoretical and operational convergence than we find. Beneath the consensus, anyone looking to break into the field would find multiple definitions of the outcome and four dissimilar datasets suited to analysis; however, they would find little guidance on identifying the trade-offs or adjudicating among their options.
Before moving on, I want to clarify what is not at stake, or at least what is not under attack. Although I partially attribute conceptual shortcomings to the prioritization of observable coding rules, this argument is not a tired rehashing of the quantitative–qualitative debate. I argue, as others have, that rigorous operationalization requires thorough conceptualization (Adcock & Collier, 2001: 530). Proper conceptualization vis-á-vis measurement is not a story of tradeoffs; it’s a story of sequence.
(Re)-conceptualizing rebel-to-party transition
Broadly, ‘rebel-to-party’ transition occurs when a militant organization qualifies as a legal, functioning political party. Yet, a thorough conceptualization involves many conceptual decisions along the way: What is a rebel? A party? A transition? I propose a conceptual framework and set of measurement strategies which optimize across the theoretical stakes (capturing political integration) and the empirical stakes (reliably identifying cases and sorting them into meaningful categories).
This task is not without challenges. On the one hand, the definitions of ‘rebels’ and ‘parties’ must be general enough to cover the variety of conflicts and political systems in which the outcome occurs. On the other hand, their definitions must be specific enough to preserve the phenomenon. Some definitions of political parties, e.g. ‘non-governmental political institutions’ (Aldrich, 1995), are broad enough to erase the line between rebel groups and parties entirely, thus making it impossible to demarcate the transition from one to the other.
Functionalist conceptual approaches are especially well-suited to meeting this challenge. Motivated by Panebianco’s argument that political parties ‘can be distinguished [from other organizations] by referring to the specific environment in which they carry out a specific activity’ (Panebianco, 1988: 6), I define rebels and parties in terms of their environments and functions. This approach maintains an observable line between the start and end points, it captures the nature of transition between them, and is consistent with the literature on which this study builds. 11 Moreover, a functionalist approach comports with and justifies the focus on rebel-to-party transition (the move into and operation in the electoral environment).
Conceptualizing ‘rebels’ and ‘parties’
From a functionalist standpoint, ‘rebel groups’ operate outside of the legal electoral environment by using armed force to influence the outcome of a political incompatibility. 12 Viable contenders for rebel-to-party transition cannot operate as a party contemporaneously during the conflict. 13 Although this scope condition is not required for a broader understanding of militant organizations, it is critical to preserving the outcome: the ability to transition from the conflict environment into party politics.
Functionalist conceptions of ‘parties’ have two common threads: the uniqueness of operating in the electoral environment; and the attempt or capacity to place candidates in office through elections (Epstein, 1980; Key, 1942; LaPalombara, 1974; Sartori, 1976). Sartori’s conception of parties represents an optimal choice for rebel-to-party analyses: ‘any political group identified by an official label that presents at elections (free or not free), and is capable of placing through elections, candidates for public office’ (1976: 54). This definition enhances conceptual specificity by forcing us to consider holding office as a key component of the background concept: integration into competitive politics. It also directly engages the rebel-to-party literature on its own terms. Finally, its agnosticism with regard to electoral openness makes it applicable to the range of regimes in which rebel-to-party transitions occur.
A new rebel-to-party framework
A sound concept must answer three questions (Goertz, 2006): What is it? What is not it? What makes up the boundary between negative and positive? Moreover, it should translate into clear operational guidelines for sorting empirical cases (Adcock & Collier, 2001; Sartori, 1970). Drawing on Sartori’s definition of parties and the collective aim of capturing political integration, I offer a functionalist conceptualization of rebel-to-party transition according to observable distinctions in transition functions and the environments in which they are performed.
Failed transition
Since a core goal of conceptualization is to facilitate meaningful comparisons (Sartori, 1970: 58), the first conceptual category explicitly defines failed transitions. Rebel-to-party failure occurs when a militant group engaged in a localized conflict has the desire and legal opportunity to enter electoral politics, but does not achieve even the minimal benchmarks of party status. 14 I identify three criteria that characterize failure, which guide theory and data collection alike.
First, militants must be engaged in a localized conflict. Becoming a party requires a target government in which to participate, and one which is at least somewhat geographically constrained. This criterion restricts the pool of rebel organizations to those for whom a within-state political solution is conceivable. As such, it filters out transnational organizations seeking to expand their influence across many states, such as Al-Qaeda or Boko Haram, both of which have appeared as failures in rebel-to-party analyses (Acosta, 2014). 15 Unless a transnational group expresses interest in transitioning within a single state, they should not be compared with groups that do. Moreover, specifying where groups would become parties (or fail to do so) is crucial for assessing the legality and obstacles to transition.
The second and third criteria address rebels’ willingness and opportunity to transition. Requiring party aspirations filters out groups that would never attempt transition in the first place. If one goal in the field is to assess why some groups succeed at becoming parties whereas others fail, scholars will only get meaningful answers by comparing groups that tried but had divergent results. The third criterion is the institutional complement to the second: the legal opportunity to enter party politics. To truly fail, the group must have attempted transition in a place where party formation was possible. 16 After all, ‘rebel-to-party status’ cannot be treated as a variable if the outcome cannot vary.
Nominal participation
The next conceptual distinction captures minimal benchmarks for party status, and I call this nominal participation. Groups in this category register as parties and sometimes appear on a ballot, but fail to win any seats. Thus, they transition into the electoral environment, but fall short of political integration (functioning in agoverning environment). Following the literature, I agree that taking logistical steps toward integration likely sets nominal participants apart from those who never even register. Yet, the factors associated with running are likely different from those associated with winning, and their implications for post-conflict dynamics may be different as well. Accounting for the heterogeneity in rebel-to-party outcomes requires conceptual differentiation at consequential junctures.
Seated participation
The final conceptual category comprises groups that have won at least some seats in post-conflict elections. ‘Seated participation’ captures a more comprehensive scope of functions that accompany political integration, aligns much better with Sartori’s definition of parties, and facilitates testing the mechanism by which rebel integration promotes peace. Specifically, if peace hinges on gaining access to legal channels for dissent, our conceptualization should differentiate between groups with qualitatively different levels of access (those who register versus those who take office). Although winning seats is not a requirement of party status, it is a salient conceptual category for theorizing the causes, implications, and mechanisms of transition.
Data collection and measurement
Keeping in mind Sartori’s instruction that concepts are not just theoretical components, but ‘data containers’ (1970: 62), conceptual frameworks are only valuable insofar as they inform inclusion criteria (who we should sort) and measurement (how we should sort them). The new rebel-to-party conceptualization provides clear guidelines for both tasks.
The framework’s explicit definition of failure guides data collection by identifying an appropriate universe of cases, which is especially important as quantitative analyses gain traction in the field. When assembling datasets and variables to test theories of transition, scholars face a trade-off between degrees of freedom and case relevance. Restricting the sample to feasible transitions promotes more valid comparisons, but limits statistical power. Adding more observations as zeroes decreases standard errors at the cost of potentially introducing selection effects by including irrelevant cases. From a case selection standpoint, every negative case in an analysis should be a defensible choice for a comparative case study. Thus, a conceptually-informed universe of cases should be limited to groups for whom a within-state political solution is a desired and plausible outcome, irrespective of the statistical costs.
As far as measurement is concerned, the framework is theoretically bounded, yet operationally flexible. I present three potential measurement strategies to accommodate a variety of questions, limitations, and goals. The first directly translates the conceptual categories into a three-stage measure: (0) failed transitions, (1) nominal participants, and (2) seated participants. 17 This measure allows scholars to assess differences in the causes and consequences of fine-grained levels of political participation.
The second measurement strategy dichotomizes transition by highlighting the different meaning of ‘success’ when militant groups win either legislative or executive seats. Because most binary variables will exhibit heterogeneity within categories, the operational task is to define a boundary that ensures the inevitable heterogeneity is least problematic. If the goal is capturing political integration, I argue that grouping nominal participants with failures is theoretically and empirically justified which I test in the following section.
The third measure builds on theoretical and empirical insights to justify further disaggregating seat winners into short-term and long-term participants (Manning & Smith, 2019; Marshall, 2019; Söderström, 2013). Specifically, the data reveal that although many rebels-turned-parties become persistent contenders in the electoral system, a significant number lack that same staying power. In this second group, rebel successor parties win some seats in the first post-conflict election but, ultimately, fail to pass muster in the long term. By the second or third election, their vote share drops below the threshold for winning seats and never recovers. Placing these ‘one-hit wonders’ in a distinct category enables scholars to test whether they exhibit systematic differences from groups that never make it beyond the ballot as well as groups that exhibit electoral persistence.
18
The measurement strategy is enumerated below: 0) Failures: groups with the opportunity and desire to transition, but who fail to register; 1) Nominal participants: registered as parties or appeared on a ballot, but failed to win seats; 2) Short-term seat holders: groups that won seats, but dropped out within three election cycles;
19
3) Persistent contenders: groups that won seats in three or more elections.
The choice to distinguish between temporary and persistent success is motivated by the stakes of the analysis. If a core goal is to elucidate how rebels’ political participation affects prospects for peace and democratization, then we should want to test not just what makes parties form, but what makes parties stick. The observable heterogeneity of rebel-to-party outcomes suggests that a more fine-grained measure of transition could reward us with more specific inferences about its causes and implications.
Empirical tests and analytic implications
The increasing frequency with which former rebels are allowed to participate in electoral politics has enabled scholars to employ statistical analyses to explore this phenomenon. Seizing the opportunity, researchers have assembled or expanded four cross-national datasets with variables suited to testing the causes and consequences of rebel-to-party transition: Acosta’s (2019) Revolutionary and Militant Organizations Dataset (REVMOD), Söderberg Kovacs and Hatz’s (2016) expansion of the Uppsala Conflict Data Programme (UCDP) Peace Agreement data (hereafter, SK&H), Manning and Smith’s (2016) expansion of the UCDP Conflict Termination data (hereafter, M&S), and Matanock’s Militant Group Electoral Participation (MGEP) data. 20 The conceptual debates above, taken alongside this wealth of new data, raise three critical questions about the coding and composition of rebel-to-party transition variables. First, do the codified differences in definitions capture the full scope of disparities across rebel-to-party data? In other words, are new scholars in a position to adjudicate among rebel-to-party variables? Second, do broad definitions of ‘failure’ introduce selection effects by including irrelevant cases? Third, do the theoretical propositions motiving the new conceptualization bear out empirically? This section assesses each question in turn.
Comparing transition variables
This article was motivated by a concern that uncodified conceptual disagreements would manifest in the data in elusive ways. Of course, different researchers with different analytic goals will justifiably take distinct approaches to conceptualization, measurement, and data collection. However, by not accounting for how or why they depart from one another, the literature may not equip new scholars to make informed decisions about which data or measures best suit their own objectives. The question is, are the disparities between transition variables predictable and justified?
The different conceptual and operational rules should give rise to three types of predictable disparities across the datasets. First, scholars using minimalist indicators of transition (e.g. party registration) should count more instances of the outcome (for a given scope and timespan) than those using stricter benchmarks (e.g. running in elections, disarming). Second, transition variables assembled using broad inclusion criteria should count more failures (and potentially more transitions) than those assembled according to more restrictive scope conditions. 21 Third, the respective sets of rebel-to-party transitions should be additive: the strictest coding rules and narrowest scope conditions should identify a core set of cases, and as the coding and scope become more permissive, scholars should count additional transitions on top of the core set.
Overview of existing rebel-to-party (RtP) variables
a Soderberg Kovacs & Hatz’s (2016) expansion of the UCDP Peace Agreement data.
c Manning & Smith’s (2016) expansion of the UCDP Conflict Termination data.
In the light of the unexpected counts, I examined each variable’s sample and found that the cases are not even remotely additive: for example, REVMOD’s 69 transitions are not the 33 cases in SK&H, plus 36 more that only registered or retained arms. 22 Rather, their respective sets of transitions only have nine cases in common. Similarly, REVMOD only has 29 transitions in common with MGEP and a mere 15 in common with M&S (despite using the same coding rule). Across the four samples, the authors only unanimously agree on eight transitions (out of 161) and five failures (out of hundreds). Often, cases coded as transitions in one dataset are not even included in others, even in instances in which the scope conditions overlap. For example, REVMOD counts 54 transitions that do not appear in M&S (during the dataset’s timeframe); conversely, M&S counts 52 transitions that do not appear in REVMOD (even though REVMOD covers this era and has broader inclusion criteria). Indeed, 94 cases are fully disjoint: one author codes the case as a transition and the others either code it as a failure or omit it entirely.
The rebel-to-party variables exhibit myriad additional irregularities that remain unexplained by scholars’ coding rules or objectives. Every pairwise comparison of transition variables reveals at least one anomaly (and as many as 17): cases coded as ‘successes’ according to stricter rules are coded as ‘failures’ according to more permissive ones. For example, MGEP counts 17 electoral participants that REVMOD codes as failures, and SK&H counts eight transitions coded as non-participants in MGEP. Probing further, the anomalies manifest themselves even in cases in which they appear explainable, for example, MGEP counts 12 transitions that SK&H codes as failures. Although we might assume those groups participated without disarming, MGEP only codes three of them as ‘violent participants’, a designation meant to capture this dynamic—a designation meant to capture armed status—leaving nine cases unexplained.
Moreover, both MGEP’s and REVMOD’s transition variables exhibit double counting. Since both datasets are multipurpose and cover long time spans, they include militant groups that eventually coalesced into umbrella organizations (e.g. the five component groups of El Salvador’s Farabundo Martí National Liberation Front (FMLN)). While their inclusion in the dataset is useful for other analyses, both datasets code the united front and its components as having transitioned, even though, in this example, FMLN became only one party. Crucially, the composite groups may warrant inclusion in the dataset, but they should not be coded as separate cases for this outcome variable, because they neither succeeded nor failed.
Different questions or analytic goals should give rise to corresponding differences in conceptualization, measurement, and sampling. However, many coding and inclusion disparities are beyond or at odds with expectations derived from the literature. Even if the measures capture precisely what scholars need, the extent of unpredictable discord in coding and case composition leaves new researchers ill-equipped to make informed decisions when adjudicating among the datasets.
Empirical strategy
These unexpected disparities introduce complications for testing the remaining questions motivating this section: whether overly broad definitions of failure introduce selection effects, and whether the new measure is empirically justified. Ideally, I would run the same model predicting rebel-to-party transition on each outcome variable and subset of cases, then compare results. However, each dataset contains different covariates and different sets of cases, and not all are publicly available. As such, both merging and replication analyses are hindered.
Given these limitations, I take each outcome variable and sample (including my own), merge it into an existing dataset, and accept the inevitable dropped cases resulting from disparate inclusion criteria. Three considerations drove my choice of a target dataset. First, because REVMOD and MGEP use the broadest inclusion criteria, only these datasets would allow me to test my hypothesis about selection effects. Moreover, a larger dataset should result in fewer losses when merging. Second, I needed a dataset with relevant covariates of transition to run and compare analyses. Third, I wanted to minimize the likelihood that dropped cases were systematically different from those included.
To optimize across these criteria, I merge the outcome variables into REVMOD’s (2019) update. REVMOD has the following advantages: it is broad enough to test for selection effects; it contains the explanatory variables of Acosta’s (2014) model predicting rebel-to-party transition; and the cases included are chosen randomly from a longer list of militant organizations (2019: 727), thereby reducing the likelihood of systematic differences between included and dropped cases. However, this choice is not without limitations. Merging revealed that REVMOD is missing numerous high-profile transitions, including 30 groups coded as transitioning in at least two other datasets. 23 Consequently, the respective samples are much smaller after merging.
I use Acosta’s (2014) model predicting transition as the basis for my analyses. He identifies three variables positively associated with transition: the group’s number of state sponsors, their size (whether they exceed 1,000 members), and whether they partially achieved their goal during the conflict. Due to the attenuation of cases from merging, all but one sample (MGEP) were too small to replicate a multivariate logit. Thus, I assess the impact of coding and dataset composition using difference-in-means tests to compare average covariate values and predicted probabilities across the samples in their entirety and subsetted by their transition status.
Irrelevance bias
The conceptual section raised theoretical questions stemming from broadly defining failure. Here, I test the corresponding empirical expectation: these definitions may induce what I call irrelevance bias. Irrelevance bias describes the changes in effect size estimates that result from including irrelevant negative cases. The origin of the bias likely occurs when scholars assemble multipurpose datasets (e.g. REVMOD and MGEP) and treat the inclusion criteria for the dataset (all militant groups) as equally-appropriate inclusion criteria for each constituent variable. However, ‘all militant groups’ includes organizations that are unlikely, unwilling, or unable to attempt transition. 24 Thus, by coding transition as 1 for every participant and 0 for all other groups in the dataset, the zeros comprise two distinct populations: true failures and irrelevant failures (illustrated in Figure 1).
The problem compounds if, as I expect, the groups least likely to become parties also exhibit systematically lower values on traits positively associated with transition. A Y = 0 sample would, consequently, introduce an implicit selection on X (King & Zeng, 2001: 701), further undermining the ability to identify traits separating transitions from true failures. Indeed, if REVMOD’s and MGEP’s transition variables resemble Figure 1(a), two problematic models may achieve significance: a model separating the true successes (however conceived) from the true and irrelevant failures (represented by the line in Figure 1(b)); or a model separating true success and true failures from irrelevant failures (represented by the line in Figure 1(c)). The first model makes good predictions, but likely has biased point estimates, because the irrelevant failures artificially attenuate the mean covariate values among the zeros. The second model returns not only biased point estimates, but inaccurate predictions as well.
Although difficult to test directly (especially given the data limitations), I expect REVMOD’s and MGEP’s samples to exhibit lower mean covariate values and lower mean predicted probabilities than the other samples, irrespective of coding rules.
25
Further, I anticipate irrelevance bias to manifest most severely in the (binary) size variable. Rather than having a true effect on the outcome, I expect some baseline membership is required for groups to even survive long enough to have the Depicting irrelevance bias
I first conduct two-tailed t-tests comparing the average covariate values and predicted probabilities across the samples. The results, which are illustrated in Figure 2, Differences in covariate values and predictions across samples
Next, I disaggregate each sample by its respective transition coding to test whether these attenuated values are clustered among the zeros. Depicted in Figure 3, t -tests reveal that REVMOD’s and MGEP’s failures take on significantly lower variable and predicted probability values than the other three samples, with only one exception. 26 Moreover, Figure 3(b) confirms my expectations about the true role of group size. Size is only significantly different between transitions and failures in REVMOD’s and MGEP’s samples, in which the failures substantially attenuate the proportion of groups under 1,000 members.
These results demonstrate that explicitly defining the negative outcome is as important for conceptualization as it is for sampling. The inclusion of non-viable groups changes the distribution of traits across the samples and artificially inflates the gaps between the positive and negative cases, making some variables appear more substantively relevant in explaining transition than they actually are.
Assessing the new measure
The question remaining is whether the theoretical propositions motivating my conceptualization and measure bear out empirically. First, if the goal is to capture political integration with a binary measure, I expect nominal participants to look more like failures than seat winners. Differences in covariate values and predictions by transition status
Coding nominal participants as failures
Although nominal participation (registration or ballot appearance) is a useful distinction for some questions, I argue it is insufficient to capture ‘political integration’: the unifying conceptual thread across the studies I engage. Thus, I code nominal participants as failures in the binary measure. If this decision is justified, my ‘failures’ (comprising 11 nominal participants and 12 failures after merging) should be indistinguishable from others (except REVMOD and MGEP, in which the zeros exhibit irrelevance bias). Otherwise, this subset should boast substantially greater values in relation to traits predicting transition, because it comprises groups that have hit minimal benchmarks of participation.
Difference in mean covariate values for seated and unseated participants
† p < 0.10, *p < 0.05, **p < 0.01 in a two-tailed t-test.
From the victors the spoils
I argue that demarcating transition at [ital] seat-winning better captures political integration, reduces unjustified heterogeneity among ‘successes’, and isolates the groups that are driving results. Here, I test whether this expectation bears out empirically by comparing nominal and seated participants across samples. Disaggregating the transitions reveals that the distributions of seat holders vary widely across samples: 39% of REVMOD’s; 62% of MGEP’s; 52% of M&S’s; and 82% of SK&H’s transitions hold seats in their respective datasets. After merging, only the proportion in M&S changes substantially, jumping to 79%. 28 However, the small number of nominal participants in the merged UCDP-based samples (two and four) means even descriptive statistics will be unreliable. Thus, to ensure robust comparisons, I also collect nominal participants across the samples into a single group (N = 46) 29 and compare seated groups’ traits with both the aggregate and in-sample nominal participants.
The differences are stark and pervasive. Difference-in-means tests (Table II) confirm that seated parties boast significantly more sponsors, a higher rate of partial goal achievement, and greater predicted probabilities of transition than their unseated counterparts. Figure 4 illustrates the magnitude by which predicted probabilities differ within the set of transitions for each coding of the outcome.
The results suggest that existing operationalizations of transition give rise to statistically and substantively significant heterogeneity among ‘success’ cases. They strongly support my expectation that seated participants, (the groups fully integrated into the political system) are driving results, thereby lending empirical credence to my conceptual framework. Moreover, the widely varying distribution of seated participants across datasets is another crucial (yet uncodified) consideration when adjudicating among them.
Empirical discussion
Three questions motivated this section: First, I ask whether the literature equips scholars to adjudicate among existing measures. A systematic comparison of transition variables reveals unpredictable, yet sizable disparities, even where we should least expect them. For every difference attributable to the posited measurement strategy, I find an anomaly in coding or inclusion for which the codified choices do not account, which compromises researchers’ abilities to make informed methodological decisions.
Second, I ask whether broad definitions of ‘failure’ induce selection effects, particularly for REVMOD’s and MGEP’s variables. Tests comparing the full and disaggregated outcome variables across samples reveal that both REVMOD and MGEP are drawn from fundamentally different populations than other outcomes. Irrelevant failures are responsible for attenuating the average values for every covariate and inflating gaps between negative and positive cases.
Third, I ask whether my conceptual reasoning is empirically justified. Two complementary sets of tests support the conceptual framework. First, I demonstrate that nominal participants are almost always statistically indistinguishable from failures—except in cases in which failures exhibit irrelevance bias—thereby justifying their coding in the binary measure. Second, tests reveal that Predicted probabilities for failed, nominal, and seated groups
Advantages and limitations for rebel-to-party research
The new conceptualization of rebel-to-party transition advances the literature along several dimensions: conceptual; theoretical; and empirical. Building on the same conceptual foundations as previous work, the framework restores consistency between the outcome that many aim to capture in theory, and the disparate approaches to measuring transition in practice. By enhancing conceptual specificity and validity, the new framework allows researchers to ask more nuanced questions and facilitates more precise empirical comparisons. Here, I outline the framework’s core contributions and discuss the trade-offs and limitations.
Grounding rebel-to-party transition in a functionalist approach enhances conceptual specificity in two ways. First, by defining ‘rebels’ and ‘parties’ in terms of the functions they perform in a given environment, this approach specifies the boundary between negative and positive cases and provides a theoretically-grounded distinction between transition and transformation. Moving forward, scholars can theorize separately about rebels’ functional participation in electoral politics and the internal transformation that may (or may not) accompany it. Does organizational transformation precede or follow transition? Does success in one predict success in the other?
Second, this approach supports decoupling ‘transition into politics’ from ‘disarmament’. Separating these distinct outcomes paints a more accurate picture of conflict (Manning & Smith, 2016; Matanock & Staniland, 2018), and allows us to ask important questions central to understanding post-conflict dynamics. Under what conditions do armed parties employ violence? How do repertoires of violence change after attaining office? Does violence coincide with the electoral cycle? Relatedly, measures derived from the new conceptualization allow scholars to empirically test the purported mechanism by which rebel-to-party transition facilitates post-conflict stability: assessing how far into transition rebel groups must progress before the peaceful outcomes associated with transition become a reality.
Moreover, by drawing on the conceptual foundations of the party literature, the framework builds a stronger theoretical bridge between conflict studies on the one hand, and studies of democratization and party formation on the other. This bridge allows us to ask questions at the intersection of these research areas. For example, do different levels of participation affect the likelihood that democracy takes root? How do electoral institutions and party system characteristics affect new parties’ strategies and likelihood of transitioning?
On the empirical side, the new framework informs clear and flexible strategies for measurement and data collection. Disaggregating transition into salient levels of participation captures the heterogeneity of rebel-to-party outcomes worldwide and enables scholars to assess more concretely which level is apposite to their question. Although scholars acknowledge variation in rebels’ electoral performance (Allison, 2010; Manning & Smith, 2019), these outcomes are treated separately from transition, which is still defined in minimalist terms. Moving forward, scholars can use the insights into different participatory benchmarks to separate or collapse stages in pursuit of more tailored and rigorous comparisons.
Finally, the explicit definition of failure provides theoretical and empirical boundaries that guide data collection. Bounding failed transitions by geographic conceivability, desire to transition, and legality, pushes the field forward by ensuring that negative cases are valid and informative comparisons, irrespective of the type of analysis. This approach translates into theoretically-informed inclusion criteria for assembling large-N datasets and rigorous case-selection guidelines for comparative case studies. As such, the new framework can help resolve some of the unjustified inclusion disparities across existing rebel-to-party samples, which will make for more comparable analyses. 30
Of course, all conceptual and measurement strategies come with limitations and trade-offs. The most potentially consequential limitation of this conceptualization is its election-centric view of political parties, particularly in the indicators. Notwithstanding their central role in government, parties do more than just run candidates in elections. Rallies and other local activities are crucial off-ballot functions that parties perform. As such, rebel groups that become active political organizations, yet are institutionally limited in their ballot appearance or performance, may fly under the radar.
Although losing relevant data is undesirable, this decision is not without justification. First, holding seats after elections is a reliable way of capturing both party status and political integration (LeBas, 2011: 26). Second, scholars of political parties (Panebianco, 1988) and rebel-to-party transition (Shugart, 1992; Söderberg Kovacs & Hatz, 2016; Van Engeland & Rudolph, 2008) argue that operating in the electoral environment is what distinguishes parties from other political organizations. Third, the distinction between transition and transformation lays the groundwork for examining groups that have rebuilt themselves as (peaceful) political organizations, but have not transitioned into electoral politics. We can then ask whether their causes and implications differ from those that achieve electoral success.
Relatedly, although I make no assumptions about the quality of elections, the measure functions best when elections occur with some regularity. In many contexts, however, elections are inconsistent, which can create ambiguous cases. Hamas, for example, participated in the 2006 Palestinian Legislative Council elections, won a majority of seats, which they still hold, yet there have been no elections since. Consequently, long-term party status is not always dependent on election cycles. One option is to code parties as persistent contenders if they have held seats through what would have been multiple election cycles. They could also be coded by the book, dropped entirely or, as I prefer, ambiguous cases can be denoted separately and dropped (or included) in robustness checks.
Conclusion
‘Rebel-to-party transition’ is a conceptual minefield. One can find disagreement over every component word; however, explicit conceptual debates are notably absent from the rebel-to-party literature. No individual scholar is at fault, yet the literature is quickly but quietly fragmenting. Although a field can be cohesive without unanimity, cohesion is not possible without direct engagement. Explicit debates bridge analyses, allowing new scholars to trace conceptual progress and make informed methodological decisions. In short, the conversation is as important as the conceptualization.
This article demonstrates that conceptual choices have critical downstream effects—directly influencing the quality of theories, measures, data collection strategies, and analyses—irrespective of whether we use qualitative or quantitative tools. To restore consistency between how many scholars use the concept in theory and the disparate approaches to capturing it in practice, I reconceptualize rebel-to-party transition and develop measures that enhance precision and capture meaningful heterogeneity in post-conflict outcomes.
The stakes of conceptual precision have implications for scholarship and policy alike. Making provision for rebels’ electoral participation has become the name of the game in conflict resolution (Matanock, 2018). On the ground, these policies are shrouded in controversy as governments grant amnesty to groups that perpetrated violent and unlawful activities throughout the conflict. Thus, for policymakers, the stakes are a matter of whether the ends justify the means, something that cannot be evaluated without an understanding of the associated risks and challenges as militants forge a path into the party system.
Although many studies pave the way for optimism, their validity is contingent on whether they are measuring what they intended to. If we get it right, studies can reveal the processes and institutions needed to create lasting resolutions to conflicts. They can shed light on new questions that bridge the gap between conflict dynamics and democratization. If we get it wrong, however, conceptual and empirical imprecision may compromise the validity of inferences. In turn, those inferences put policymakers at risk of hastily concluding that rebel-to-party transitions are complete, effective, or desirable. Prematurely withdrawing post-conflict support could create opportunities for spoilers to instigate a return to violence. In short, if rebel-to-party analyses fail to capture what they purport to, both the scholarship and the policies that follow are based on a literal false sense of security.
Footnotes
Replication data
Acknowledgments
This article benefited from rigorous feedback from a myriad of wonderful colleagues. Many thanks to Amelia Hoover Green, Laura Jakli, David Kang, Meredith Loken, Hilary Matfess, Gerardo Munck, Sarah Orsborn, Evan Ramzipoor, Paul Staniland, Hillel Soifer, and William Wohlforth. I am also grateful for lively and incisive discussions from Stanford’s CISAC working group and the University of Southern California’s CP workshop. In addition, the incisive comments from the editor and three anonymous reviewers were very helpful.
Funding
The author received no financial support for the research, authorship, and/or publication of this article.
