Child Soldiers in Armed Conflict
A quantitative study concerning the effect of education on child soldier recruitment.
Julia Carlbäcker Peace and Conflict Studies Bachelor Thesis, 15hp, Fall of 2017
Supervisor: Ralph Sundberg Department of Peace and Conflict Research
Uppsala University, Sweden
1
Table of Context
List of figures and tables ... 2
1. Introduction ... 3
1.1 Introduction ... 3
1.2 Background to Child Soldering ... 4
1.3 Previous Research ... 5
1.4 Research gap and contribution ... 7
2.Theoretical Framework ... 8
3. Research Design ... 11
3.1 Method and Data ... 11
3.2 Operationalization of variables ... 13
3.2.1 Dependent variable ... 13
3.2.2 Independent variable ... 14
3.2.3 Multicollinearity and Index variables ... 16
3.3 Source criticism ... 19
3.4 Scope conditions ... 20
4. Results and Analysis ... 20
4.1 Descriptive Statistics ... 21
4.2 Results ... 23
4.3 Analysis ... 26
4.4 Alternative explanations ... 30
5. Conclusion ... 31
6. List of References ... 33
Appendix I. Codebook and operationalization of variables ... 36
Appendix II. Robustness checks ... 37
2
List of figures and tables
Figure 1. Theoretical claim ... 11
Figure 2. Distribution of Index variables ... 22
Figure 3. Predicted Probability (%) of Independent variables ... 26
Table 1. VIF test ... 17
Table 2. Descriptive statistics ... 21
Table 3. Results from the logit regression ... 25
3
1. Introduction
1.1 Introduction
The use of child soldiers in armed conflict is violating international law and is, according to the UN security council, one of six grave violations against children in armed conflict (OSRSG- CAAC, n.d.). Still, children are recruited in armed conflicts all around the world. Child soldiers are present in some rebel groups but not in others and in some countries’ military forces but not in neighboring countries’ forces. This variation imposes a puzzle for the international community, NGOs and academic researchers. If we could understand under which circumstances children are recruited in armed conflicts, we could have a better chance of developing effective policy implications in order to protect children from the risk of being recruited or associated with armed struggle.
Previous research concerning the causes of child soldering has found different explanations for the puzzle. Although, in qualitative studies, educational opportunities have been highlighted as a main important factor decreasing the recruitment of children. Several qualitative studies have stressed the importance of the combination of quality of the education provided as well as access and availability of education (Brett and Specht, 2004; Cohn and Goodwin-Gill, 1994; Machel, 1996; Maclure and Denov, 2006). Despite these findings in qualitative research, quantitative research has only looked at access to education and its importance for child soldier recruitment and therefore missed out on a crucial aspect of the quality of the education in the analysis of the effect of education (Tynes and Early, 2015; Vargas and Restrepo-Jaramillo, 2016). According to my knowledge, no quantitative study has been conducted which matches the theoretical argumentation from qualitative studies and therefore, I argue, that the research area could be considered inadequate.1 A research gap is therefore identified with the lack of an adequate quantitative study in the area of the effect of education on child soldier recruitment.
In order to fill this research gap, I will conduct a quantitative research which includes both aspects of quality of the education itself and access to education, in order to evaluate education's effect on child soldier recruitment. From now on, I will use the term quality of education as a concept including both quality of the education itself and access and availability factors. I will build on existing theories from qualitative research in the field with the aim of testing the validity of these theories in a quantitative study and to test education’s effect on child soldier recruitment in a more adequate way concerning the theories developed in previous
1 This argument is also supported by Ames, 2010.
4 qualitative studies. By doing so, I will try to answer the research question: How does quality of education effect the likelihood of child soldier recruitment? If a relationship could be observed on a global level between the quality of education and child soldier recruitment, it would be an important contribution to the policy work protecting children from involvement in armed conflict. I argue that increased quality of education decreases the risk of child soldier recruitment in armed groups because increased quality of education leads to increased educational opportunities which keep children occupied and they are therefore less vulnerable for or attracted to recruitment in armed groups.
The structure of this research paper will proceed as follows. First, I will summarize previous research on the causes of child soldering and thereafter look more closely on the previous research concerning education. Second, I will explain my theory and methodological choices that I have made in order to fill the research gap in the best way possible. Following the methodological section, I will show the results of my systematic research and provide an analysis. The results of the research conducted for this paper, concerning quality of education’s effect on child recruitment, cannot confirm the relationship between the variables. In order to establish the true effect of education on child soldier recruitment, more research is needed. This will be discussed at the end of the paper where I will reflect on my methodological choices and their impact on the results as well as discuss alternative explanations and areas of further research.
1.2 Background to Child Soldering
International rules against child soldiers are present in the 1948 Universal Declaration of Human Rights and in the additional protocols to the Geneva Convention which were adopted in 1977, prohibiting military recruitment of children below 15 (Child Soldiers International, n.d.). Rules prohibiting child soldier recruitment is also present in the 1989 Convention of the right of the Child (Singer 2010:94). Although, it was not until 1996 that child soldiers received special attention and the UN asked for a comprehensive report of children’s situation in armed conflict. The report which was conducted by Graça Machel, shed new light on child soldiers and has been an important study in the field of children and armed conflict. The report led to the establishment of the Cape Town Principles in 1997, a meeting where experts were brought together to develop strategies to prevent child recruitment (UNICEF, 1997). In 1999 the UN security council adopted the first resolution on children and armed conflict and identified six grave violations that were priority for the Council (OSRSG-CAAC, n.d.). Building on the Cape Town Principles, the Paris commitments brought the issue up on the agenda again in 2007 and
5 governments agreed to act against recruitment of children in armed forces and rebel groups (Gates and Reich, 2010:3; UN, 2007). Even though the issue of child soldering has gained more international attention since the end of the cold war, children are still recruited on all sides in armed conflicts and even an increase has been observed (Singer 2010:93). It is impossible to know how many children that are a part of modern warfare but the globalization and technical advancement of small arms could be one factor explaining the observed increase (Gates and Reich, 2010: 11).
1.3 Previous Research
Previous research concerning the causes of child soldering has mainly been conducted by NGOs, civil society organizations and advocacy groups. Relatively little academic attention has been brought to the research field (Gates and Reich, 2010: VII,14). The research of advocacy groups is important but often lacks the time and money to conduct statistical analysis and theoretical models needed in order to reach a more precise understanding of why children join or are recruited into armed groups (Ames, 2010). Although, children’s situation in armed conflict has gained more attention recently and the research field is growing.
The available research of causes of child soldiers can be summarized into three different categories. One group focuses on strategic explanations for child soldier recruitment and the second group argues that child soldiering is a consequence of the conflict setting. The last branch of research highlights the importance of structural conditions and their effect on the likelihood of child soldier recruitment.
First, the strategic explanations for why children are recruited in armed conflict include lower cost of recruitment, tactical benefits and the easiness of manipulation (Beber and Blattman, 2013; Maclure and Denov, 2006; Singer, 2006). Children are easy to recruit and therefore a cheap way for armed groups to generate force according to Singer (2006). The use of children in battle might also offer tactical benefits for rebel groups because it can induce moral dilemmas for the opposite side (ibid.). Beber and Blattman (2013) studied rebel recruitment in Uganda and found that children are attractive recruits even if their strength and fighting abilities were less than adults because children can be manipulated at a lower cost, are cheaper to indoctrinate and easier to misinform (ibid:68-69). Maclure and Denov (2006) studied boy soldiers in Sierra Leone and found that when the conflict disrupts societal norms, children are easily manipulated and socialized into new norms of violent behavior. In the case of Sierra Leone, alcohol and hallucinatory drugs and the availability of small arms also facilitated the manipulation of children (ibid:124–131).
6 In contrast, Tynes and Early (2015) argue that child soldiers are recruited as a consequence of the conflict setting. The authors argue that when disputants become more desperate, child recruitments are more likely. The authors highlight the importance of looking at conflict duration and intensity as well as conditions that effect the nature of the conflict, such as level of militarization and presence of democratic institutions. Tynes and Early conclude that duration and intensity are significant factors explaining child soldier recruitment due to increased desperation from the fighting parties (Tynes and Early, 2015:107–108). According to the findings of the research, democratic governance decreases the likelihood of child soldier employment (ibid.). Tynes and Early (2015:85) explain this correlation by arguing that more democratic regimes follow the norms of human rights to a greater extent than authoritarian and repressive regimes.
The last and most comprehensive part of causes of child soldier research provides analysis on a number of structural explanations for when children are recruited in armed conflict, including existence of IDP (Internally displaced people) camps, poverty, employment opportunities and education (Achvarina and Reich, 2005; Brett and Specht, 2004; Cohn and Goodwin-Gill, 1994; Machel, 1996). First, Achvarina and Reich look at the relationship between IDP camps in a country and the existence of child soldiers but do not find a strong relationship (Achvarina and Reich, 2005). Other structural factors which have gained more attention are the effect of poverty and employment opportunities. Although several studies conclude that poverty is an important factor, it fails to explain variation because many communities are poor, especially during conflict, but not all of them recruit children (Achvarina and Reich, 2006; Brett and Specht, 2004; Vargas and Restrepo-Jaramillo, 2016). Brett and Specht, (2004) argue that poverty creates a general vulnerability for recruitment and they find that most child soldiers come from impoverished backgrounds. However, poverty cannot be the only factor because many poor children do not become child soldiers (ibid.).
Instead, several authors emphasis that access to education is a more crucial factor effecting child soldier recruitment (Brett and Specht, 2004; Cohn and Goodwin-Gill, 1994;
Machel, 1996; Maclure and Denov, 2006; Vargas and Restrepo-Jaramillo, 2016). If children are not involved in education or work, they need to find another activity to secure their economic survival or to find something to do with their time. In that case, according to Brett and Specht (2004:126) they could be more prone to join an armed group as an alternative economic activity. However, even if the authors find a strong relationship between education and child soldiers, they argue that their research fails to systematically explain the variation between variables because of the qualitative nature of their research (Brett and Specht,
7 2004:147). The report by Machel to the UN in 1996 also shows that when educational opportunities are limited, recruits tend to become younger and younger (Machel, 1996:17).
Improved economic status, access to basic health and education decrease the possibility of armed groups to talk children into running away and join the armed struggle. Good education opportunities might then prepare them to resist the offers from armed groups (Vargas-Barón 2010:203). Inadequate education can, on the other hand, make rebel groups seem like a viable option (Cohn and Goodwin-Gill, 1994: 34).
1.4 Research gap and contribution
The importance of education has been mentioned in several previous studies, however, the majority of them are qualitative studies based on interviews with former child soldiers (Brett and Specht, 2004; Cohn and Goodwin-Gill, 1994; Machel, 1996; Maclure and Denov, 2006).
Despite the importance showed in qualitative research, only two quantitative articles address the issue of education and both of them focus only on access to education. According to Brett and Specht, access to education is important but also the quality of the education itself; “It is, however, not only the real, effective access to school that is important, but also the quality of the educational experience, and its relevance to perceived future or even present economic prospects” (Brett and Specht, 2004: 131). Therefore, I argue that the quantitative research that has been conducted in the field has measured education in an inadequate way to match the theories developed by previous qualitative research.
The two previous quantitative studies which address education, have been conducted by Vargas and Restrepo-Jaramillo (2016) and Tynes and Early (2015). Vargas and Restrepo- Jaramillo (2016) conducted a quantitative study in Colombia and focus on poverty’s effect on child soldier recruitment. The authors include education as a variable and measure it by school attendance rates and average year of schooling but lack theoretical motivation for the relationship as well as fail to include the important aspect of the quality of the education itself.
The study is also limited to Colombia and lack inclusion of regional and global variance. The only global quantitative analysis which has included education as a variable, has been conducted by Tynes and Early (2015). However, in their article, education is only measured as expected year of schooling which is a measurement concerning how many years of schooling a child could expect to receive. This variable does not include present enrolment rates which indicates access and availability to education or the quality of the education provided. These factors have been highlighted by theories in previous qualitative research and argued to have a strong relationship with children´s educational opportunities and in turn their level of exposed
8 vulnerability for recruitment (Brett and Specht, 2004; Cohn and Goodwin-Gill, 1994; Maclure and Denov, 2006).
Concerning the importance of educational opportunities found in previous qualitative research, and the scarce and inadequate way in which quantitative research has followed up the theories developed in the field, I argue that a research gap is identified. There is a lack of a comprehensive quantitative analysis of education’s effect on child soldier recruitment, including both availability aspects as well as aspects of quality of the education itself. To fill this research gap, I will conduct a quantitative research on education’s effect on child soldier recruitment and develop the measurements of education further in order to include both aspects of education. By doing this research, I will try to answer the research question: How does quality of education effect the likelihood of child soldier recruitment? But why would quality of education matter for child soldier recruitment? This will be discussed in the following section.
2.Theoretical Framework
In this section an elaboration of the theoretical explanation for the relationship between the variables will be provided. Theoretical concepts used in this research will be specified and the section will conclude with the theoretical claim for this paper followed by a flow chart explaining the causal relationship between the variables and this paper’s testable implication in the form of a hypothesis.
2.1 Theory
In order to fill the research gap, I will expand the measurements of education to correspond to the theory provided in this section. In this paper, I argue that increased quality of education decreases the risk of child soldier recruitment in armed groups because increased quality of education leads to increased educational opportunities which keep children motivated and occupied. Children are therefore less likely to drop out and join the armed struggle as well as less vulnerable for forced recruitment.
This argument is built upon previous qualitative research where opportunities have been highlighted as one of the most important mechanism in explaining the causal relationship between the variables (Brett and Specht, 2004; Maclure and Denov, 2006). The most comprehensive analysis of the causal chain has been developed by Brett and Specht (2004). In sum, the authors argue that lack of educational opportunities lead to that children search for alternative opportunities and are therefore more attracted to employment in armed groups as a valid alternative to education or more vulnerable for forced recruitment. The authors highlight two main important factors that effects educational opportunities. First, motivation for
9 education is one important factor that increases children’s willingness to stay in school or to search for alternative opportunities (Brett and Specht, 2004). Lack of motivation can include poor quality of the education provided or lack of relevance (ibid.). If education is inadequate or unlikely to lead to employment, going to school might seem pointless and children are therefore more likely to drop out (ibid.). Second, availability and access to education is crucial.
According to Brett and Specht, availability can consist of economic availability, meaning that poor children might not have the opportunity to attend school or access to a physical school.
Access and availability also includes the equality and inclusion of education, for example equal access and availability between gender (ibid.).
Further, the authors argue that if some or all of these factors are lacking, children may not be in school and have nothing to do with their time. The need of activity, status, or income might influence the choice of joining the rebellion or armed groups voluntarily during armed conflict. Or, they are more vulnerable for forced recruitment because they have no place to be, are looking for other things to do and are exposed to different violent groups in society (Brett and Specht, 2004:44). Former child soldiers have emphasised this relationship and the importance of opportunity in the causal explanation; The main cause of going there was unemployment, I think. I had nothing to do here so I went there. If you have some business or you are studying, then you do not think about taking part in Jihad. Aziz, Pakistan (Brett and Specht, 2004:23).
Brett and Specht explain the relationship between the variables by arguing that both the quality of the education itself and the access to it is important for children’s educational opportunities. Therefore, I argue that the measurement of school life expectancy which has been used in previous quantitative studies is not a sufficient measurement in order to evaluate the importance of education for child recruitment. To contribute to the research field, the definition of the concept of quality of education developed in this paper aims to include a broader meaning of education, so that it would correspond to factors that affect children´s educational opportunities. Hence, I believe that this definition captures the phenomenon of interest and together with the description in the operationalization section, the concept is also possible to measure and compare across different cases.
To fully understand the theoretical argumentation in this paper the concept of child soldiers need to be specified. Several different assumptions exist regarding the concept of child soldiers and child soldier recruitment and it is therefore crucial to address these in order to explain how the concepts are used in this research paper. The definition of a child soldier varies but this paper will use the most common definition which is stated in the Paris principles from
10 2007. In the Paris Principles, a child soldier refers to “a person below 18 years of age who is or who has been recruited or used by any armed force or armed group in any capacity” (UN, 2007: 7). This definition changed after 2002 and increased the age from 15 to 18 years old (Child Soldiers International, n.d.). Even if the term soldier is used it can be misleading because children often work as spies, cooks, porters or sex slaves and are not necessarily wearing a gun in order to count as a child soldier (Gates and Reich, 2010: 3; UN, 2007: 7). In recent years a new term has been developed in the international community to include this variation in the concept. The more accurate term is Children Associated with Armed Forces and Groups (War Child, 2015). However, with this knowledge this research paper will still refer to the term Child Soldier due to the frequent use of the term by most official reports and academic research. This definition is also used by the dataset for this research on child soldier existence (Haer and Böhmelt, 2016: 160) which is why I have chosen to use the same definition for my research.
Another important aspect of the definition of child soldier recruitment is the difference between forced and voluntary recruitment. The international community has previously focused mainly on forced recruitment of young children. Although, in reality we see a high degree of children and adolescents who voluntarily join armed groups (Brett and Specht, 2004). Little research has focused on voluntary recruitment due to lack of valid information concerning why children join armed groups (Ames, 2010). Brett and Specht aim to fill the research gap and try to capture the concept of voluntary recruitment but at the same time they problematize it. How voluntary is voluntary recruitment? The authors mean that even if children answer that they voluntarily joined the armed group, they might not have had different options: “I didn’t choose this situation. You know that we are in a country at war, and then you don’t have much choice.
You can run away or fight”, Christine DRC (Brett and Specht, 2004:10). Although, even if considered difficult to measure, the concept of voluntary recruitment needs to be considered and therefore I choose to include a causal explanation for both voluntary and forced recruitment in my analysis. The data on child soldiers has not separated these two categories and therefore an assumption is made that both forced and voluntary recruited children are present in the data.
So, in this research paper, the term recruitment refers to “compulsory, forced and voluntary conscription or enlistment of children into any kind of armed forces or groups” (UN, 2007: 7).
Building on the theories developed in previous research with a focus on the theory developed by Brett and Specht and upon the conceptual definitions specified, I argue that quality of education, including both motivational factors and access and availability factors, are crucial in explaining child soldier recruitment. Decreased quality of education might on one hand lead to decreased motivation to complete the studies, or on the other hand, lead to
11 decreased availability or access to education. Both of these factors might in turn lead to a search for alternative opportunities which might increase the attractiveness of joining armed groups.
As well as increasing the vulnerability for forced recruitments due to increased levels of out of school children with low levels of educational background. At the end, this might lead to increased likelihood of child soldiers in armed groups. An illustration over the causal relationship can be observed in Figure 1.
Figure 1. Theoretical claim
Through this research, I will test the following hypothesis: Decreased quality of education increases the likelihood of child soldier recruitment in armed conflict. How the research will be conducted and details concerning the method used to test the hypothesis, will be explained in the next section.
3. Research Design
The following section will explain the methodological choices made in this research and the data used to conduct the analysis followed by an elaboration of the operationalization of the variables. Finally, a section discussing source criticism will be provided.
3.1 Method and Data
The aim of this paper is to provide a comprehensive quantitative analysis of quality of education and the relationship to child soldier recruitment. This will be conducted in order to fill the previously mentioned research gap by expanding on the analysis of education’s effect on child recruitment. In order to do so, I will build my research on the Child Soldier dataset provided by Haer and Böhmelt (2016)2. The data includes information concerning rebel and government recruitment of children between 1989-2003 and consequently, I will use the same time period for my analysis. However, my research will be limited to recruitment by rebel groups because
2 The original dataset is available at: https://www.polver.uni-konstanz.de/data/
Decreased quality of education
Search for alternative opportunities Decreased
availability and access Decreased motivation
Increase attractiveness to join armed groups
Increased physical and mental vulnerability of forced recruitment
Increased likelihood of child soldiers in
armed groups
12 the authors have argued that the government data is unreliable, and they advise not to use the data in any research (Haer, 2017, personal communication [email], 20 November). This limitation does not affect my study except for decreased variation in the dataset and a limited scope. The level of analysis is dyad-year and the dataset provides the most detailed information available in the field. This analysis is limited to the available observations from the Haer and Böhmelt dataset. However, the sample provided gives a broad understanding of the whole population of cases and the spread over regions increases the generalizability of the sample.
The total number of observations is 781 and the proportion of observation in each region is as follows: Africa 37%, Asia 29%, Middle East 18%, Latin America 10% and Europe 6%.
Another available global dataset on child soldier recruitment is conducted by Tynes and Early (2015). They analyze child soldier existence or no existence in rebel groups and government forces, but are not specifying the result towards specific rebel groups. I have chosen to build my analysis on the dataset by Haer and Böhmelt due to more specified information and increased number of countries included in the dataset. Consequently, this choice increases the number of observations for my sample with the intention to create more generalizable results which hopefully correspond well to the overall population.
Concerning the independent variable, I will conduct my own dataset with variables on relevant aspects of education. Expanded discussion about this can be found in the operationalization section below. The data for the educational variables will be taken from UNESCO Institute of Statistics (UIS, n.d.). This source is chosen because UNESCO provides the most comprehensive data on education and is the main source of educational statistics used by other international actors, for example the World Bank’s educational dataset (see World Bank, n.d.). The created education dataset will then correspond to the same time period as the information concerning child soldiers.
However, some adjustments to the original data have been made. The theory of the paper emphasis the importance of available education for the children in risk of recruitment.
Therefore, a dyad between USA and Al-Quida in 2002 was deleted from the dataset because the location of the conflict were USA and child soldier recruitment was observed in Al-Quida.
I argue that this dyad is outside of the theory of this paper because with my method of analyzing, the child soldier observation would be analyzed together with educational statistics from USA which would not be an accurate way of measuring according to my theory. Dyads where the conflict is carried out in another country and where the exact origin of the rebel group is difficult to trace should therefore not be included in the analysis. This is because in these cases, as the case of Al-Quida, accurate educational statistics could not be obtained and matched to the
13 corresponding dyad and therefore the research would lack validity. In this paper, this dyad-year was the only one that matched this criterion and therefore was the only dyad excluded from the analysis.
3.2 Operationalization of variables 3.2.1 Dependent variable
The dependent variable of this analysis is child soldier recruitment in armed groups. According to the definition provided by the Paris principles and by Haer and Böhmelt (2016) a child soldier is defined as a person below 18 years of age which has been recruited or used by an armed group. In their dataset, each dyad-year which contained child soldiers was assigned a code of 0 if no information could confirm child soldier usage in the armed group in that specific year and 1 if child soldier usage could be confirmed. Due to the dichotomous character of my dependent variable I will conduct a binominal logit regression analysis in order to test my hypothesis.
The definition of the dependent variable is consistent with the international understanding of the definition of a child soldier and the operationalization follows the same definition. Child soldier recruitment is coded if the group used child soldiers in their armed group regardless if the children were forcibly recruited or joined voluntarily or what task they performed within the fighting group. This operationalization corresponds to the definition explained in this paper which means that the operationalization of the dependent variable is measuring what it is supposed to measure and therefore is valid to a great extent.
However, before 2002 child soldiers were defined as children below 15 years old. The data traces information back to 1989 and uses several international organizations as sources for child soldier recruitment. There is a valid chance that these organizations used the previous definition of child soldiers before 2002 since that was the international standard at the time. This means that the information concerning these cases is only measuring children below 15 years of age.
The coding for cases before 2002 might therefore be built on sources which have only reported if a group used children below 15 and are therefore not corresponding to the definition provided by Haer and Böhmelt as well as the definition used in this paper. However, child soldiers are still measured both before and after 2002 even if children between 15-18 years old might not have been included in the earlier period. Therefore, I argue that this is not a problem for the results of the analysis, but this difference is important to acknowledge.
14 According reliability of the operationalization, I argue that the operationalization is reliable in the sense that the same data is available for download which makes it easy to test my results or run a different analysis on the same data as I have used for this research.3
3.2.2 Independent variable
The theory in this paper, which is built from previous qualitative studies, focuses on educational opportunities and highlights both motivational factors and the availability and access to education. Previous quantitative research has not included both categories and therefore failed to correspond to the theory concerning relevant and available educational opportunities.
Therefore, to match the theory, the independent variable used in this paper, quality of education, is a variable which aims to include several different variables which in turn captures both motivational factors as well as access and availability. There are no international standards on how to measure the expanded concept of quality of education, and it is done in a variety of ways dependent on what it is supposed to measure. Although, UNESCO, as a major actor of educational policies at the international stage, published a report in 2015 on achievements and challenges concerning the goal of education for all as an analysis as a preparation for the Sustainable Development Goals (UNESCO, 2015). The report addresses important factors which affect the availability, quality and access of education which corresponds closely to the definition of quality of education used for this paper. I have chosen measurements that correspond to the phenomenon explained in the theory and at the same time have been used in international organizations to measure similar phenomena.
In order to operationalize quality of education to correspond to the theory I have chosen to divide the concept of quality of education into two subgroups: Quality and Relevance of education and Access and Availability. Thereafter, I have included three measurements in each group which covers different aspects of the two subgroups. I will operationalize as follow:
Quality and Relevance of education will be measured by 1. Governments’ expenditure on education as a percentage of total expenditure, 2. Pupil/teacher ratio and 3. Adult literacy rate in the country. I have chosen these measurements because they capture important aspects of the theory by effecting the motivation for education, as explained by Brett and Specht (2004) and these are also used in UNESCO´s report concerning quality of education (UNESCO, 2015). In that report government prioritization in education, measured by the governments’ expenditure on education as a percentage of total government expenditure is used as well as adult literacy
3 The dataset for child soldier recruitment is available at: https://www.polver.uni-konstanz.de/data/
15 rates. UNESCO also highlights the importance of teachers in achieving quality of education and the pupil/teacher ratio is used as a measure of quality (UNESCO, 2015), which is why I included it as a variable. Access and Availability will be operationalized by 1. Net enrolment rates, 2. School life expectancy and 3. Gender Parity Index of enrolment rates4. This choice is also built on theory where school life expectancy has been used as a measurement of access (Tynes and Early, 2015). This operationalization also corresponds nicely to UNESCO´s report which used net enrolment rates as a measurement of access and availability of universal education as well as inequality rates measured by the parity index in enrolment rates (UNESCO, 2015: xii, 80-83).
This operationalization is more inclusive in the understanding of the diversity of the concept of education, compared to previous research in the field. I argue that these variables correspond more precisely to the theory and the causal relationship provided. If the theory points to the right relationship, increased levels of quality and relevance as well as access, availability and equality will correspond to decreased likelihood of child soldier recruitment.
According to the theory, this is because these factors increase the educational opportunities and decrease the search for alternative opportunities.
The unit of analysis for my research is dyad-year and the education statistics, available for country level, will be matched to the location of the conflict in the child soldier dataset. By doing so, accurate statistics on the quality of education provided in the country where the conflict is ongoing will be used. When information on education about specific years is missing, the closest prior value to the missing one will be used in order to receive the most accurate data available which relates to the opportunities provided for the children in the country. When conflict breaks out, educational statistics are often less available. However, the quality of education available before the outbreak of the conflict still provides an understanding of the level of quality of the education available for the exposed children. Increased intensity also affects both the quality of the education available as well as child soldier recruitment and will therefore be controlled for in my analysis in order to receive better accuracy of the educational data.
By my operationalization I have tried to include the most relevant variables which affect children’s motivation for and access to education. These are selected upon previous research on education´s effect on child soldier recruitment (Brett and Specht, 2004; Maclure and Denov,
4 In order to be measured together with the other variables, Gender parity index have been changed to include only inequality between gender and not if it favourable for boys or girls. Consequently, the variable can take a value between 0-1 were 1 is equality.
16 2006; Vargas and Restrepo-Jaramillo, 2016) as well as what have been highlighted by policy makers in the international community (UNESCO, 2015; UNICEF, 2016). Therefore, upon my knowledge and understanding, the operationalization is valid because it is measuring what is important for the theory in this research in a way that is accepted and commonly used in large international organizations. A full explanation of the variables used and the sources for them can be found in Appendix I.
However, even though the variables are closely selected upon these criteria and analyzed to be important both by scholars and policy makers, they are not perfect. Due to inconsistency in measurements of quality of education, alternative variables may be of importance in order to fully grasp the concept of quality of education. Education can be measured in a variety of ways and the indicators are available in several different formats, measuring slightly different aspects of a concept. For example, enrolment rates can be measured by net enrolment rate, adjusted net enrolment rate or gross enrolment rate which measures slightly different aspects of enrolment rates. I have, upon the best of my knowledge and building on previous research and reports from policy makers, tried to include the most relevant indicators. Although, this might need to be developed in future studies.
The operationalization is reliable due to available information on UNESCO´s website on the precise variables and indicators used for measuring the independent variable in the dataset for this paper. If my dataset on education will be used in future research, one will have access to the exact same figures on education for country and year which increases the reliability of the data.
3.2.3 Multicollinearity and Index variables
To measure education in a more inclusive way and not only in a different way, calls for a need of several independent variables which might relate to each other due to the fact that they are measuring different aspects of education. Before conducting the regression, a need to conclude if multicollinearity exists among the independent variables, or if they might be run together in the regression, is crucial. If a multicollinearity problem exists among the variables this might in turn affect the results (Kellstedt and Whitten, 2009:238-244). One approach to identify collinearity is to use a test of variance inflation factors (vif), where a higher value indicates higher collinearity (beckmw, 2013). The results from the test in this paper can be observed in Table 1. A value between 1-5 generally indicates low multicollinearity and a value between 5- 10 generally indicates high multicollinearity (beckmw, 2013). If a variable takes a value over 10 the regression coefficient would be poorly estimated due to multicollinearity (ibid.).
17 Table 1. VIF test
In Table 1, one can observe that both enrolment and school life expectancy have high multicollinearity with other variables and literacy rate almost reaches the same level. Although, none of the variables reach a level over 10. In order to follow the line of the theory and to answer the research question presented in this paper, I argue that all variables are needed in order to explain the variation in education. Therefore, I will keep all six variables but introduce them separately in the regression models due to the risk of high multicollinearity.
According to the theory, these variables are interacting and capturing different important aspects of education and the relationship between the total sum of all variables measured against the dependent variable would therefore still be interesting to observe. To do so and at the same time avoiding the risk of multicollinearity, I will create three index variables. All independent variables are standardized in the way that the mean takes a value of 0 and the standard deviation takes a value of 1. By doing so, all variables could be compared even if they initially were measured in different scales. For example, variables measured in percentage could be compared to variables measured in quantity. These standardized values could thereafter be added together and divided by the number of variables to receive an average value of the variables included. I will create a variable of an average of the total of all variables as well as an index variable including the three variables concerning access and another one with the variables concerning quality of the education itself. By doing so, the total score of all education variables could be measured against the dependent variable. The importance of the two subgroups, access and motivational factors could also be measured in the same manner, to evaluate if one subgroup is more related to child recruitment than the other.
3.2.4 Control Variables
As discussed in the previous research section, several different explanations exist regarding what causes child soldier recruitment. In order to evaluate the relationship between education’s effect on recruitment of children, three other important variables from previous research have been chosen to be included as control variables in this study; poverty, democratic governance and intensity.
In previous research, poverty has been discussed to be a key factor of causes of child soldering (Achvarina and Reich, 2005; Brett and Specht, 2004; Vargas and Restrepo-Jaramillo, 2016). However, even if poverty might have an impact, many poor countries do not recruit
Enrolment GPI in
enrolment Literacy rate Expenditure
on education Teacher/pupil
ratio School life expectancy
7.10 3.93 4.99 1.09 1.33 8.07
18 children and therefore the variable has failed to explain the variation (Achvarina and Reich, 2006: 5). Poverty is included as a control variable due to the attention that poverty has gained in previous research and because it might have an effect on both quality of education and child soldier recruitment. I have followed the same operationalization of the control variables as Tynes and Early (2015), because their article is a well-established and well cited article in the research field and they have all three variables included in their study. Poverty has been calculated using GDP per capita for each country and year using statistics from the World Bank (World Bank, 2017). Achvarina and Reich (2005) have measured poverty in a more inclusive way including several different measurements which gasp more of the concept of poverty.
However, in the scope of this article, the inclusive approach is not possible to include and therefore I will use the operationalization conducted by Tunes and Early.
Second, level of democratic governance will be measured by the Polity IV index as done by Tynes and Early5. Even if other variables are available in order to measure governance, the polity scale is used in order for this article to be more comparative to other research regarding level of democracy as well as correspond to the work by Tynes and Early (2015). The authors find correlation between the level of democratic governance and recruitment of children. Level of democratic governance might also have an impact on quality of education because governments’ policies are directly affecting the countries’ educational standards. Therefore, the variable is included as a control variable in this research.
The third and last variable is intensity and will be measured by battle related deaths per year in a country using data from UCDP, Uppsala Conflict Data Program6. Tynes and Early (2015) argue that high intensity increases the desperation of the combatant groups and therefore increases the likelihood of child soldier recruitment. Intensity might also affect education opportunities due to increased closure and detriment of schools (GCPEA, 2017) and is therefore important to include as a variable in order to control for the effect of intensity on child recruitment. Tynes and Early coded battle related death as a dichotomous variable using 0 for conflicts with 25-999 battle deaths per year and 1 for those conflicts reaching over 1000 (Tynes and Early, 2015: 94). In this research I have decided to use the actual numbers of deaths per year to include more variation in the variable. However, intensity data has been aggregated to country and year level and not at dyad level. This is because of coding difficulties due to inconsequent use of coding id and dyad names which made it difficult to merge intensity with the rest of the data in the limited scope of this research. I also argue that the level of intensity
5 The Polity IV is available for download at: http://www.systemicpeace.org/inscrdata.html
6 Data available at: http://ucdp.uu.se/downloads/
19 in the whole country is important because it might influence the feeling of desperation in several dyads which make the coding relevant in my case. Although, this might be done differently in future research to test the significance of the actual intensity level of the specific dyads.
3.3 Source criticism
The validity and reliability of the data have been discussed but several aspects and shortcomings concerning the sources of the data are necessary to highlight and specify at this point. Due to the limitation of the scope of this research, I have been unable to collect new data and therefore I had to rely on available data in the field. As mentioned before, quantitative studies of child soldiers are limited, and data availability is therefore restricted which have limited the choice of data used in this research. Hence, even if data has been available it is not perfect and therefore some faults need to be brought up in order to increase the transparency of the research as well as highlight some source criticism that might have an impact on the research outcome.
Concerning the data on child soldiers, it has been conducted from Hear and Böhmelt (2016). The authors used both advocacy groups, news reports and academic articles in order to receive information concerning child soldier usage in rebel groups (Haer and Böhmelt, 2016, Appendix II7). However, these sources could include biases. For example, organizations might have incentives to exaggerate the facts on child soldiers to receive attention to their organization and current mission. On the other hand, rebel groups, when asked by reporters, academics or staff from advocacy groups, might have incentives to downplay the use of children in order to avoid punishment (ibid.). The sources used when conducting the data might have had potential biases and the data could therefore suffer of reliability issues. However, this is a well- established methodological problem in studies of child soldering due to great fear in post- conflict societies to be honest about the facts, both from former child soldiers, rebel leaders and governments (Gates and Reich, 2010).
The process of choosing cases to conduct data from might also suffer from potential bias in the case of the data from Hear and Böhmelt. There is a vast majority of incidents reported, compared to non-incidents. 666 observations include child soldier usage while only 115 include non-usage in the original dataset (including also 153 NA:s). This might be representing the overall population, however, the variation in the dependent variable decreases, which might influence the analysis in the research. There is also no information concerning why these countries where chosen and in turn no discussion of sample representation is provided.
7 The Appendix II was received from Haer upon email request and were not available online.
20 There are also some important facts to correct from the article of Haer and Böhmelt. In their description of the data they state that the data includes dyads between 1989-2010.
However, in the dataset, only dyads between 1989-2003 are included. The time span is therefore shortened with almost 1/3 of what was thought to be available, which also limits the representation and variation for my research. Although, the dataset accomplishes to include a variety of cases from different regions, including South America, Africa, Asia and Europe.
Concerning the independent variable, UNESCO builds its data on a majority of governmental data which could sometimes suffer from bias. However, they also conduct household surveys which might widen the picture and together it might be quite close to reality (UIS, n.d.). A problem occurs under circumstances when this information is difficult to reach.
For example, during conflict. Data as close as possible to the actual date has been used, but it might be invalid due to that the data might in reality be very different during the conflict period.
This is important to have in mind, but the use of the educational data is done in the best way possible according to my knowledge as well as corresponding to previous research.
3.4 Scope conditions
According to Haer, the data on government use of child soldiers is unreliable (Haer, 2017, personal communication [email], 20 November), and therefore it was not included in the research. The results of the research are therefore limited to rebel use of child soldiers which in turn then might not explain the use of children in government forces. The results of this paper might be applicable even for those cases, but the scope of the research is limited to rebel groups.
The data includes countries from South America, Africa, Europe, Asia and the Middle East and therefore I argue that the scope of the analysis is not restricted to any substantial geographical limitations. However, the data collected for this research spans from 1989-2003 and a limitation in the temporal dimension exists. As mentioned in the introduction, child soldier usage has changed after the cold war and because of the limited time frame I argue that the research is mostly applicable to armed conflicts after the end of the cold war. However, the theory of this research does not have to be restricted to these circumstances even if the result of the analysis might be restricted. If the theory is applicable to a broader domain should be tested in further research.
4. Results and Analysis
In this section, the results from the logit regressions will be presented followed by a discussion of the main findings of the research. Further, an analysis will be conducted concerning the implications for theory as well as methodological choices which might influence the outcome
21 of the regressions. At the end of the section, shortcomings of the research will be presented and discussed as well as alternative explanations and areas for further research.
4.1 Descriptive Statistics
Before presenting the main findings of the analysis a section on descriptive statistics will be provided in order to increase the understanding of the data and the variables used for the research.
Table 2. Descriptive statistics
============================================================
Statistic N Mean St. Dev. Min Max
--- DEPENDENT VARIABLE:
Child Soldiers 781 0.9 0.4 0 1
INDEPENDENT VARIABLES:
Enrolment rate 742 71.3 23.8 15.0 99.9 Equality 742 0.8 0.2 0.4 1.0 Literacy rate 714 63.4 24.7 10.9 99.7 Expenditure 727 13.8 4.7 2.9 41.4 Teacher 773 -32.2 14.3 -90.4 -12.3 School Life 764 7.6 2.9 1.6 13.0
INDEX ON EDUCATION:
Total 779 -0.1 0.7 -1.5 1.3 Access 764 -0.05 0.9 -2.3 1.4 Quality & Relevance 779 -0.01 0.6 -1.9 1.7
CONTROL VARIABLES:
Intensity 741 1,566.2 3,807.0 25 49,698 GDP per capita 700 2,602.9 5,365.1 65.0 27,759.3 Democracy 781 0.8 7.5 -10 10 R have been used to generate the statistical results.
The dichotomous character of the dependent variable limits the variable to values of 0 or 1. In the descriptive statistics in Table 2 one can observe that the mean of the variable is 0.9. This indicates that a clear majority of the observations take value 1 (child soldier recruitment) compared to 0 (no recruitment) and consequently gives the analysis a limited variation in the dependent variable. More precisely 14,7% of the observations take value 0 in the dependent variable and 85,2% take value 1.
However, by consulting the descriptive statistic table an interesting variation can be observed among the independent variables. Among the six independent variables on education, a great difference between minimum and maximum can be seen. This indicates that the observations in the sample range from very low levels of education with 15% enrolled and an
22 average of 1.6 years of schooling, to very high, with 99.9% enrolment and up to 13 years of schooling. This gives the analysis an important variation in the independent variable which increases the possibilities to test the hypothesis. The mean of the different independent variables is fairly centered between the min and the max which indicates that the values are fairly normally distributed around the mean. Further, it might be so that not so many extreme outliers effect the results. Some exceptions to this statement concerns few observations in enrolment with low values which differs from the mean and some cases with very high levels of expenditure on education and teacher/pupil ratio8 which differs substantially from the mean of the variable. Concerning the control variables, the mean of Intensity is 1,566 and the maximum value is almost 50.000. This indicates that in the sample used for this analysis, few cases with very high levels of intensity are present when the average of intensity among the observations have lower intensity levels.
The variation in the three different index variables on education is small due to the standardized values for all independent variables which the index is built upon. The limited variation in the values makes the values in the descriptive statistics more difficult to analyze by only looking at the numbers.
Figure 2. Distribution of Index variables
In order to familiarize oneself with the index variables and to facilitate the analysis of the descriptive statistics, a boxplot showing the variation in the index variables can be observed in Figure 2. By looking at the figure, the index variables are more visually explained. One can observe that in the Quality variable several outliers are present with very low levels of quality of education.
8 In the statistics the variable of teacher/pupil ratio takes a negative value in order to follow the same scale as the other variables, a very low number indicates more children per teacher
23 4.2 Results
This research was conducted with the aim of providing a quantitative analysis of education’s effect on child soldier recruitment by providing a more inclusive measurement of the independent variable. To do so, a logit regression was conducted with several different independent variables, effecting both quality of the education itself and access to education.
The purpose of this paper is to provide an answer to how quality of education effect child soldier recruitment. With the background from the descriptive statistics this section will present the results of the logit regression made in this analysis.
In Table 3 the results from the logit regression between the dependent variable and the different independent variables as well as the index variables can be observed. The regression is made with a dichotomous dependent variable which make the interpretations of the coefficients slightly different from a linear regression model. A positive coefficient indicates that an increase in the independent variable relates to an increased possibility of 1 in the dependent variable. A negative coefficient indicates a relationship where an increase in the independent variable is associated with a decreased likelihood of 1 (Bjerling and Ohlsson, 2010). However, much more is difficult to say by only looking at the coefficient because the relationship between the variables is not linear due to the dichotomous character of the dependent variable (ibid.). In order to analyze the results in a more concrete way, a predicted probability analysis has been conducted. The results from the predicted probability analysis are presented in Figure 3. This shows the predicted probability of child recruitment when each independent variable increases from its minimum value, to its mean and finally up to the variables’ maximum value. By doing a predicted probability analysis it is easier to observe what the relationship between the variables mean and not just consulting the value of the coefficient (Bjerling and Ohlsson, 2010).
Although, some important things can be observed by looking directly at Table 3. In model 1-6 all independent variables are introduced alone together with the control variables.
One can observe that when the independent variables are introduced separately, all the different educational variables are statistically significant, according to the p-value. The p-value ranges from 0-1. The lower p-value, the greater confidence we have in that the relationship between the variables is systematic and not found by random change (Kellstedt and Whitten, 2009:147- 150). A larger sample size might, for example, increase our confidence in the relationship due to a more accurate representation of the population (ibid.). A p-value of <0.05 indicates a 95%
confident level and a p-value of <0,01 indicated a 99% confidence level (ibid.).
24 In model 7 and 8 all education variables are measured together, first without the control variables and then with the control variables. When analyzed together the variables have less significance. One reason for this might be multicollinearity between the variables, a risk that was observed in the VIF test. In model 9 one can observe the relationship between the standardized total value of education, through the index variable, and child soldier recruitment.
In model 10 the standardized value for all variables related to access are introduced and in model 11 the standardized value of all quality variables can be observed. All three index variables are also statistically significant. However, even if the results are statistically significant it does not prove the relationship to be strong or if the relationship is causal (Kellstedt and Whitten, 2009:147-150). Consult the analysis section for further discussion.
When examining the results of each model an unexpected relationship can be observed.
The correlation coefficients for the majority of the independent variables are positive. In all independent variables except expenditure on education, increased levels are therefore, according to the table, correlated with an increased likelihood of child soldier recruitment.
These results points to the opposite relationship compared to the theory of this paper and previous studies in the field. The same relationship can be observed in the index variables. An increase in the total score of quality of education is related to an increased risk of child soldier recruitment. Expenditure on education is the only variable which has a negative effect on child soldier recruitment. Consulting the predicted probability in Figure 3, one can see that when expenditure moves from its mean to its maximum value, the risk of child recruitment decreases from 86% to 57% while the other variables indicates an increased change when the independent variable moves from the mean value to the maximum value.
The only other variable which has a negative effect throughout all the models is GDP per capita. GDP per capita is statistically significant and an increase is associated with a decreased risk of child soldier recruitment, supporting the results from previous research (Achvarina and Reich, 2006; Brett and Specht, 2004; Vargas and Restrepo-Jaramillo, 2016). It is also the only control variable which is statistically significant.
Table 3. Results from the logit regression
==========================================================================================================================================
Dependent variable: Child Soldier Recruitment --- (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)
--- Enrolment 0.023*** 0.031*** 0.014 (0.007) (0.012) (0.016) School Life 0.183*** -0.120 0.143 (0.062) (0.107) (0.141) Equality 2.472*** 1.670 2.544 (0.860) (1.447) (1.592) Expenditure -0.056** -0.058*** -0.050**
(0.022) (0.021) (0.024) Literacy rate 0.012** -0.015 -0.011 (0.006) (0.010) (0.011) Teacher 0.029*** 0.011 0.025**
(0.009) (0.008) (0.010) Index Total 0.865***
(0.257) Index Access 0.582***
(0.182) Index Quality 0.574**
(0.258) Intensity 0.00002 0.00002 0.00001 0.00001 0.00003 0.00000 0.00001 0.00002 0.00002 0.00002 (0.00003) (0.00004) (0.00003) (0.00004) (0.00004) (0.00003) (0.00004) (0.00003) (0.00004) (0.00004) GDP per capita -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001*** -0.0001***
(0.00002) (0.00002) (0.00002) (0.00002) (0.00002) (0.00002) (0.00003) (0.00002) (0.00002) (0.00002) Democracy -0.007 -0.006 0.022 0.023 0.024 0.008 -0.019 -0.003 -0.005 0.011 (0.022) (0.020) (0.018) (0.018) (0.018) (0.019) (0.026) (0.020) (0.020) (0.019) Constant 0.377 0.720* -0.095 2.787*** 1.249*** 2.921*** 1.062 0.106 2.097*** 2.115*** 2.009***
(0.480) (0.426) (0.698) (0.370) (0.326) (0.339) (0.974) (1.039) (0.154) (0.156) (0.146) --- Observations 637 659 637 647 648 663 651 602 667 659 667 Log Likelihood -252.231 -256.456 -253.583 -254.286 -256.278 -261.318 -273.809 -229.673 -261.703 -255.770 -264.978 Akaike Inf. Crit. 514.461 522.912 517.165 518.572 522.555 532.635 561.619 479.346 533.405 521.541 539.957 Note: *p<0.1; **p<0.05; ***p<0.01, Dispersion parameter for binomial family taken to be 1
26 Figure 3. Predicted Probability (%) of Independent variables
4.3 Analysis
This section will provide an analysis of the results presented in the previous section as well as a discussion about shortcomings of the research and areas for future research. The research question of this paper is: How does quality of education effect the likelihood of child soldier recruitment? And the expected relationship, if the theory is true, is summarized in the hypothesis of the paper: Increased quality of education decreases the likelihood of child soldier recruitment in armed conflict.
However, the results of the analysis cannot confirm the hypothesis. According to the results the opposite relationship could be observed; increased quality of education increases the likelihood of child soldier recruitment. All at a statistically significant level. Some scholars have argued that education might actually have the opposite effect on child recruitment because schools might in some cases work as a recruiting ground and teachers might encourage students to join (Brett and Specht, 2004: 19; Vargas-Barón 2010:215). Nevertheless, this theory has little support empirically and even if the results might point to that relationship, I argue that there are several shortcomings with the data and the methodological choices that first need to be taken into consideration before giving too much value to the outcome of the regressions. I argue that
Min Mean Max
Access 63,48424 86,45861 93,52014
Index Total 64,37896 86,06383 95,10961
Quality 67,8013 86,34425 94,5399
School Life 68,35869 86,53935 94,56615
Enrollment 62,73749 86,19795 92,4077
GPI 64,16148 85,53982 100
Expenditure 92,11999 86,37179 57,08451
Literacy 77,34331 86,31775 90,60155
Teacher 95,6057 99,00377 91,7471
0 10 20 30 40 50 60 70 80 90 100
