Transnational Transmitters: Ethnic Kinship Ties and Conflict Contagion 1946–2009

Abstract

Previous research has proposed that ethnic conflict may spread across borders. Although the importance of transnational ethnic groups is often emphasized, the processes through which contagion may take place remain unspecified. The present study presents a context for more precise analysis of contagion. Further, it identifies distinct processes through which contagion is likely to occur within this context. It is argued that when an ethnic group engages in violent conflict in one state, kin in a nearby state may be inspired to rebel because the outbreak of conflict renders ethnic bonds and similar conditions salient. These bonds and similarities become even more salient when the kin group has opportunities and willingness to mobilize for rebellion. Statistical analysis employing unique global data covering 1946–2009 supports this argument. These results indicate that kinship ties matter for contagion and identify some of the conditions which amplify the effects such ties have for contagion.

KEYWORDS:

• civil war
• contagion
• diffusion
• ethnic conflict
• ethnic kinship ties
• transnational ethnic groups

Notes

¹ See, for instance, Ayres and Saideman (2000), Buhaug and Gleditsch (2008), Cederman, Girardin, and Gleditsch (2009), Forsberg (2008), Gleditsch (2007), and Saideman and Ayres (2000).

² Viewing contagion as a process rather than an outcome follows, for instance, Elkins and Simmons (2005).

³ I use the term state A to refer to the state with conflict and state B to refer to the nearby state considered at risk of contagion. This follows the terminology used in other studies using the same conceptualization of contagion (for example, Black 2012; Forsberg 2008).

⁴ Fearon’s inclusion of descent is informative in categorizing which religious communities that could be classified as “ethnic,” such as Catholics in Northern Ireland.

⁵ For instance, the measure used by Buhaug and Gleditsch (2008) includes kinship ties between a given state and people living in a nearby conflict area. It is not clear to what extent these groups are actually involved in the conflict.

⁶ In addition, with a monadic country-year setup, each kinship tie cannot be assessed separately (that is, for each dyadic relationship).

⁷ In a monadic country-year setup, the focus had instead been on, for example, Macedonia, which would have been coded as having ethnic kin involved in conflict somewhere in the neighborhood.

⁸ This analogy is used to illustrate the approach only. It is not meant to preclude human agency and portray contagion as a deterministic process.

⁹ The notion that units that are to some extent connected and similar to each other makes political contagion processes more likely has also been suggested by Walt (2000).

¹⁰ Another possibility is that group members in state B choose to cross border and support their kin involved in conflict rather than being inspired to initiate their own rebellion. However, since this is outside of the definition of contagion it will be left for future research.

¹¹ This is also a reason why the argument is explored in the context of countries, rather than ethnic groups. It further allows for a more straightforward comparison to existing research, which is dominated by country-level analyses.

¹² Indeed, conflict involving Touaregs erupted in Niger in 1994. See the UCDP Conflict Encyclopedia (UCDP) for further information about the Touareg rebellions.

¹³ Estimates vary but generally report the Touareg population of Burkina Faso to constitute less than 1% of the total population.

¹⁴ A review of all groups involved in ethnic conflict shows that a majority has significant populations of kin members living in other states but primarily so in proximate states. Most exceptions consist of ethnic groups which have links to diaspora scattered elsewhere due to recent migration. However, as diaspora groups are mainly in opposition to the regime of their homeland, not their host state, they are less likely to be linked to contagion processes as defined here.

¹⁵ Alternative specifications of proximity are discussed and analyzed in the online appendix.

¹⁶ Alternative specifications of one and three years have also been tested and do not substantially alter the main findings. These results are reported in the online appendix.

¹⁷ The function equals 2ˆ(-y/2) where y is the number of years since the last onset of ethnic conflict in the country (or since the country is first observed) and 2 is the half-life selected based on log-likelihood.

¹⁸ Hence, to evaluate this particular type of transnational kinship links it does not suffice to determine whether or not two states share an ethnic group, using data from the Minorities at Risk Project (2005) or the EPR-ETH data set (Cederman et al. 2010). One is required to link ethnic groups to rebel organizations involved in armed conflict.

¹⁹ A subset of this data collection was introduced in Forsberg (2008). The temporal domain of that study is 1989 to 2004. For the present study, that data has been expanded to cover the period from 1946 to 2009.

²⁰ This follows Forsberg (2013), who examines a suggested domino effect of granting concessions to separatists. Alternative half-lives have been estimated; five years performed best in terms of log likelihood.

²¹ A fuller description of the coding procedure is located in the online appendix.

²² Since the MAR data set covers a smaller sample of ethnic groups than the EPR-ETH data set, data is missing information for some of the coded kin groups.

²³ Alternative specifications of this variable are discussed in the results section.

²⁴ All four specifications of kinship are modeled as decay functions in the same way as the Ethnic kin variable.

²⁵ For both GDP/capita and Population size I used the epolate command in Stata for linearly generating information to fill in gaps in time-series (by interpolation) and to extend the data-series up to 2009 and back to 1946 (by extrapolation). Countries with no available data for any of the years in the time-series are left blank.

²⁶ Alternative clustering strategies are discussed in the online appendix.

²⁷ The variables relating to the type of kinship cannot be estimated in the same model, since they are related to each other by definition (Lee Ray 2003). The kin variables pertaining to different characteristics of the kin group are all coded from the same subset—those assigned “1” on Ethnic kin.

²⁸ This finding is robust to alternative specifications of “large” meaning 20% or 30%. If 40% or larger, the covariate is not significant.

²⁹ Dropping these two variables from the models does not significantly alter the results.

³⁰ This supports the notion that state capacity may counteract the demonstration effect. Alternatively, state capacity can capture state B’s repressive capacity of putting down challengers early and “fortify” itself against demonstration effects. Such efforts may on the other hand in itself create additional incentives to rebel. As an alternative specification of a state’s repressive capacity, I reestimated the models with state B’s military expenditure (as part of GDP or the total government spending) added, using data from SIPRI (Stockholm International Peace Research Institute). This variable is significant in most of the models but is instead shown to increase conflict risk.

³¹ To assess the strength of the reported results, a number of robustness tests have been performed. These include reestimations of the models using rare events logit, alternatives for dealing with spatial and temporal autocorrelation, alternative operationalizations, additional sampling techniques, and models with additional control variables. For more information, consult the online appendix.