In Chapter 3, the main hypotheses were derived from the theory. The models predict when rebels are likely to use child soldiers or when they will refrain. I now describe how the explanatory variables identified earlier are measured.
Secession
Hypothesis 1 predicts that where the rebel goal is secession, child soldiers are less likely to be used. The variable secession is used to capture the primary goal of the
rebellion. To operationalize this concept, I code the type of rebellion using the START Terrorist Organization Profiles (Terror Knowledge Database 2010), para-military profiles from GlobalSecurity.org (2010b), and additional secondary sources.23 I recoded the data from the EACD in order to operationalize goals, rather than tactics. This was coded based on the stated goals of the rebellion or an analysis of their objectives. I used the values coded directly from the START database where researchers identified the goals explicitly. In the GlobalSecurity profiles, I coded the rebellions based on the relevant descriptions. Secession is a dichotomous variable that is recorded as "1" if the rebellions evaluated goals were either “Nationalist/Separatist” or “Secessionist,” and "0" otherwise. Even if other goals were included (e.g., religious, communist/socialist, leftist), rebellions were recorded as secessionist. Nearly half, 41.75% of the rebellions in my sample are coded as secessionist. The expected direction of the relationship is negative: Secessionist
23
I did not use the type of conflict recorded in the original EACD data. Here, there were several categories that would have captured the type of rebellion. However, the inclusion of terrorists as a type indicates that in some cases, tactics were used instead of goals and would be an inappropriate proxy for the concept of group goals.
rebellions should be less likely to break the norm against child soldiers. Table 2 offers a summary of the primary independent variables included in the dataset. Figure 5
summarizes the number of groups coded as separatist. --Table 2 here-- --Figure 5 here--
Rebel goals may change. Some may call for a change in policy and later a change in government. Frustrated, they might then call for independence. While conceivable, goals are unlikely to have changed in the 10 year period that I have data on child soldier use. In fact, they do not for any of the 103 rebellions in the dataset according to the secondary source material. My sample does capture change in behavior, but I treated this as imprecision in the source material. In other words, for the sake of thoroughness, I attributed holes in the recruitment record not as evidence of child protection, rather as limited information in the period. This was only the case if there was no evidence in a single report but child soldiering evident in one or both of the other reports. Thus, rebel goals are treated as invariant for the large-n analysis. I confront the issue of changing goals in Part Three of the project where I look at case studies of individual rebellions in detail.
Opportunism
Hypothesis 2 predicts that rebellions that are opportunistic will be more likely to use child soldiers. To determine if illicit funds were used, I used four sources of
information. First, I consulted the data on illicit funds collected by Fearon (2004). Unfortunately, this only covered rebellions up to 1999. I also used funding information
from Civil Wars of the World (CWW) (DeRouen and Heo 2007). With both sources, a significant problem was that the information contained therein referred to the civil war state rather than individual rebellions. Once again, I consulted the START database and GlobalSecurity.org. Between these four sources, I was able to rule out contraband financing for approximately 65% of rebellions in the dataset. I let START and GlobalSecurity.org take precedence in the coding. That is, if Fearon (2004) or CWW recorded illicit financing, but both START and GlobalSecurity.org failed to provide evidence, I changed the coding to "0". This strategy is used because while each source is valuable, only the GlobalSecurity and START profiles contain individual rebel
information. Both CWW and Fearon (2004) often include two or more rebellions active during the conflict. The only time I can specify which rebellion is using contraband financing using these sources is if there is only one active rebellion. In the absence of alternative information, however, Fearon (2004) and CWW is preferred.
Scholars have met considerable difficulty in arriving at a consensus regarding funding and civil conflict in previous literature. Many estimates involve the ratio of primary commodity exports to GDP (Collier and Hoeffler 2004) or some variation of this measure. The type of resource is important, though, and may account for the mixed results in the literature. For example, Ross (2004a) suggests that for resources to be influential, they must be both portable and profitable. In other words, lootable resources should matter, whereas non-lootables may serve as an incentive, but are not likely to substantively influence the current funding profile of the rebellion. Another approach is to examine the geographic distribution of resources and the location of rebels (Lujala 2010). This too is useful at explaining motivations, but it does not help fully understand
the impact of funding on behavior. For example, while active in the heart of a region known for drug-trafficking, the Shan State Army in Burma is not known to be involved in the illicit trade. In fact, they are even known to cooperate with Thai anti-drug taskforces to combat the cross-border trade in heroin (GlobalSecurity.org 2010). The same cannot be said of its contemporaries like the United Wa State Army and the Mong Tai Army. Likewise, though the Khmer Rouge was known to use illegal logging of forests in its controlled territory to fund its rebellion, the NDFB in India is said to erect signs to combat smuggling of trees in the Manas forest reserve (GlobalSecurity.org 2010). Therefore, state-level measures and geographic data were not precise enough to determine which rebellions were actually using lootable contraband financing to fund their rebellions. My approach offers such information.
I consider rebel financing specifically. It is not enough to be in an area that
happens to have the potential for illicit economic activity, and it is certainly not sufficient to be in a country with high value lootables. Instead, by recording the evidence for or against contraband financing of the operation, I am able to include reliable measures on economic activity. The conceptual definition, illicit economic activity, and the empirical definition I employ are more closely aligned than other approaches. Descriptive statistics are presented in Figure 6.
--Figure 6 here--
Duration
Hypothesis 3 predicts that longer rebellions will be more likely to use child soldiers. I use the natural log of duration to capture the number of years the rebellion has
lasted from its foundation.24 To measure the length of the rebellion, it was necessary to consider when the rebellion began, beyond the 1998 beginning date of the sample. I used activity dates from the START terrorist database and GlobalSecurity.org. I also used the World Bank Violent Conflict Database (World Bank 2010a) and the Upsala Conflict Data Program (Gleditsch et al. 2002) as a cross reference for active dates. This combined approach allowed me to avoid intermittent lulls in hostility. To estimate the conditional effects of secession on duration, I use an interactive term duration*secession which is log duration times the dichotomous variable secession.
General Drawbacks
There are some shortcomings with my general data gathering approach. First, it may be the case that reporting bias limits information on non-Western or otherwise smaller rebellions. While this may be the case, because my temporal period is recent guards against this major concern. My initial rebel list comes from a dataset already widely used and contains detailed information. Rebellions explored here already have detailed information that was used to compile the initial list. Likewise, the secondary sources consulted appear to have little regional bias. That is, they report as much
information on the Middle East and Africa as they do with South American rebellions. I suspect there could be bias due to the size of the rebellions or where there are many active rebels in one country. In GlobalSecurity.org’s profiles, in at least two cases, the Democratic Republic of the Congo and the Burmese conflicts, the website does not have specific profiles for the rebellions involved. However, in both of these cases,
GlobalSecurity maintains a detailed report about the overall conflict, including discussion of the behavior and funding characteristics of parties involved. Therefore, while the individual profiles are not there, the relevant information is still available. Likewise, that I use two or more primary sources for explanatory variables, I am able to rule out
substantial systematic reporting bias.