What is the value of home?
A quantitative study on the effects of natural resource extraction on conflict-induced displacement
Department of Peace and Conflict Research Bachelor's thesis
Supervisor: Jonathan Hall 10th January 2021 Word count: 10,222
1. Introduction 4
2. Theoretical Framework 6
2.1 Literature Review 6
2.2 Theory 10
3. Research Design 14
3.1 Data 14
3.2 Dependent variable 15
3.3 Independent Variable 17
3.4 Control Variables 18
4. Results and Analysis 20
4.1 Descriptive statistics 20
4.2 Bivariate regression 23
4.3 Multiple Regression 24
4.4 Discussion 26
5. Conclusion 30
6. Bibliography 33
Conflict-induced displacement, a relatively novel term, is well researched but not well understood.
There is a significant amount of findings on this subject, but many have been disproven at a later stage, leaving behind a research field largely lacking in substantial findings. As conflict-induced displacement has steadily increased over the previous years, this is a significant problem. However, recent findings, hinting at a relationship between natural resource prevalence in armed conflicts and displacement, could help provide an explanation as for what causes these differences. A large-n study looking at 207 cases of armed conflict, varying over relative value of natural resource
extraction, is here conducted. The results find that while there is not a statistically significant relationship between the two variables, and this relationship varies depending on whether cross- or within border displacement is the focus, there is some level of covarying relationship.
What causes conflict-induced displacement, and when does it occur? It is per definition a
consequence of armed conflict, but not all conflicts generate the same quantities of displacement.
Research within the field of conflict-induced displacement has unfortunately not reached a consensus on exactly what causes this variation. While the field blossomed in the 90s and 00s, interest waned over the following decade. During this time however, new findings were made which promise to provide concrete answers where previously there were few. By embracing these ideas, the author aims to help reawaken interest within the field, in a time where forced
displacement is growing (IDMC, 2020a). This paper will attempt to utilize a new classification of forced displacement to analyze a promising new find, the causal relationship between natural resources in armed conflict and levels of conflict-induced displacement.
The theory that armed conflicts over natural resources may see increased displacement was first presented in Niazi and Heins (2017) paper. Conducting a qualitative study, it was found that the suggested relationship was not only present, but had a significant impact on all three examined cases. However, no causal mechanism was convincingly presented, and the results remained ungeneralizable. If the results produced by Niazi and Hein (2017) were found to be true in a qualitative study, and if they are statistically significant enough to be deemed generalizable, it could have significant real world implications. In a field where few causes are consistently found to be relevant, this could help guide policy makers and spark new questions for the field, and may help reignite interest in forced displacement as a research topic. The findings presented above are not the only new additions to this field, and the new definition of conflict-induced displacement, as presented by IDMC, Internal Displacement Monitoring Centre, may be even more important (2020b). Dividing forced displacement into disaster- and conflict-induced displacement, a much greater precision and accuracy can be used in research. Much of the previous findings were based on forced displacement at large, i.e. all types of involuntary displacements as a direct result of the actions of others, and may therefore not necessarily satisfy the need for validity. To show how significant this difference may be, approximately two-thirds of IDPs (Internally Displaced Person) in 2019 were not displaced by conflict, but would be classified as disaster-induced displacement
(IDMC, 2020a). Utilizing the new data collected by IDMC may thus finally solve some of the previous discrepancies, and could help highlight the complex dynamics of displacements.
This paper will therefore conduct a quantitative study, with the aim of applying the new available data to accurately find the causes of conflict-induced displacement, and more specifically find the causal relationship between natural resources in armed conflicts and said displacement.
The study will be looking at a large array of armed conflicts in which there are various levels of natural resource extraction, while controlling for variables which previous studies suggest affect conflict-induced displacement.
This paper will be divided into five sections. The first, being the introduction, can be found above.
The second part, theoretical framework, is itself divided into two subsections.
The first subsection, literature review, gives a summary of the state of the field of conflict-induced displacement research, explaining the key findings and trends.
The second subsection is the theory, where this paper's hypothesis and causal mechanism will be explained and motivated.
The third section, research design, presents the selected method, as well as data sources and operationalizations of the variables.
The fourth section, results and analysis, presentes the results of the paper and will show these in a concise manner. This will be followed by a discussion and analysis of the results, to conclude if the hypothesis is correct and discuss other potential findings of note.
The final section, Conclusion, will summarize the paper and present several avenues for future research.
2. Theoretical Framework
2.1 Literature Review
Conflict-induced displacement research is exhaustive, yet a few paradigms have come to dominate the field. While collecting them into easily identifiable groups may be simplistic and there is other important work outside these boundaries, this summary will focus on the most prominent, or to this case relevant, strands. Two particularly prominent strands are; the economic theory and the choice-based approach, both of which are necessary to understand the current state of the field.
First is economic theory. The literature within this branch has found that while low levels of income causes armed conflict (Chassang & Miguel, 2009:212), and while this conflict in turn may cause conflict-induced displacement, the economic factors were not a direct cause in- and of itself and did not directly affect levels of conflict-induced displacement (Mohammed, 2017:26). Prakash meanwhile found that higher levels of local economic opportunities reduced the likelihood of displacement due to conflict at communal and individual levels (Prakash, 2013:142).
Contradictions in the economic field can be found not only by looking at income and
opportunity, but also at macro-levels. For example, Schmeidl (1997:830) found that economic development did not have a mediating or causal effect, and variations in economic development and macro economic trends could not explain the differences in levels of displacement. This was not the findings of other researchers, however, who found direct links between economic development in terms of GNP and a reduction in forced displacement (Moore & Shellman, 2004:732). These contradictory findings permeate the field, and muddles the water regarding economic effects on conflict-induced displacement. The contradictions may be caused by some studies looking at non-typical cases (Prakash, 2013:142), or variations in premises and definitions (Moore & Shellman, 2004:725). Other findings within this framework fail to find substantial results due to low levels of validity in their chosen variables. An example of this last problem can be found in an article claiming a positive relationship between conflict-induced displacement and value of land, however it uses population density as a proxy of land value (Raleigh & Hegre,
2009:32-33). Ultimately, while the economic theory is useful and delivers several key insights, it does not produce a simple and satisfactory model.
The second framework is the choice-based approach. It argues that each displacement can be seen as a choice, and points at each displacement event showing groups of both displaced and
non-displaced, hinting that each individual here makes a choice between fleeing and remaining.
Most causal factors can thus, according to the choice-based approach, be viewed as factors
affecting the choice to leave one's home, and therefore tends to focus on the psychology, safety and opportunity of those who may be displaced (Moore & Shellman, 2004:725). This theory finds that state-based discrimination of human rights and states in the process of democratization both increase the risks of conflict-induced displacements (Davenport, Moore & Poe, 2003:42).
Melander and Öberg (2007:167) find the reversed relationship with democracies being far less likely to generate forced displacement along a linear trend however. As for human rights violations, there are conflicting findings. Schemidl finds no significant effect between human rights violations and conflict-induced displacement (2007:302), while Moore and Shellman find a substantive support for a positive causal relationship between the two (2004:739). As the choice-based approach attempts to find why the decision to flee is made, much of the research is based on individual and interpersonal factors. The micro-economic theories of income and opportunity can be fitted into the framework, but as stated above does so with mixed results (Mohammed, 2017:26;
Prakash, 2013:142). Highlighting interpersonal variables as one of the key factors however is done by Edwards, who finds that information networks passing person-to-person are the most
important factor, and that these are drivers of conflict-induced displacement rather than the violence itself (Edwards, 2009:40). These interpersonal relations are themselves subject to change during times of armed conflict, and the networks can function as both push and pull factors (Wood, 2008). Further, the threshold for being subjected to displacement varies from person to person based on other factors, such as level of education, age, land ownership or national commitment. Furthermore the idea of all potential victims of forced displacement as unitary is a false assumption (Melander & Öberg, 2006:146-147). The choice.based approach presents strongly supported theories, yet while some are broadly applicable, most are case-specific and highly contextual. This follows naturally from a framework that rejects structural explanations. It
is therefore, while interesting and useful in certain cases, lacking in generalizability and it’s lack of simple models makes it less useful for predicting and preventing displacement events.
Both frameworks produce useful insights, but are by their nature not sufficient to fully explain the macro-level factors of conflict-induced displacement. Thus, most useful and substantive findings in the field can be found outside of these, in individual relationships between causal factors and conflict-induced displacement. The first set of theories are focused on who the perpetrator is and why. One hypothesis that has been found to be false is the causal link between ethnic conflicts and increased quantities of displacements (Melander & Öberg, 2006:144). Schmeidl (1997:301) argues that this may be due to victims alternatively joining the fighting rather than leaving in ethnic conflicts, yet no explanation is offered as for why this is an alternative here but not in conflicts without an ethnic dimension. As for the nature of the perpetrator, there does not seem to be a significant difference between government or rebel forces in relation to displacements (Schmeidl, 1997:302). This leaves a few factors that do seem to affect conflict-induced displacement which needs to be discussed. First, one paper finds that conflict intensity does have a causal effect on the quantity of conflict-induced displacements (Melander & Öberg, 2006:144). However, the same authors later publish results to the contrary, finding no statistical significance in intensity of fighting with levels of conflict-induced displacement, yet they do not provide any explanation as for why these results differ (Magnus & Öberg, 2009:166). The same paper does find that
geographic scope and physical spread of the conflict has a large impact however, following similar logic as to the information theory presented by Edwards, yet utilizing push rather than pull factors (Melander & Öberg, 2007:168). The authors argue that a larger geographical spread of conflict will involve and affect more people, increasing the likelihood of feeling threatened as opposed to localised conflicts which only impact a smaller quantity of people. Another key factor is the presence of foreign military personnel in the conflict. Armed interventions result in significantly larger quantities of forcibly displaced (Schmeidl, 1997:301). Unfortunately, the author cannot confirm the findings as the data may be skewed as the results are somewhat built on outliers represented by actions during the cold war, and that the results may not therefore be generalizable to a post-cold war setting (Schmeidl, 1997:303). This reservation is countered by another study which finds that conflicts during and after the cold war show no significant difference in terms of
displacements generated (Melander, Öberg & Hall, 2009:24). Next, there is evidence that past events of forced displacement increases the risk of new displacement events (Davenport, Moore &
Poe, 2003:41). Finally, one last important factor is population size, which also correlates positively with displacement quantity (Magnus & Öberg, 2006).
Finally, a significant finding that has not gained traction in the field of conflict-induced
displacement, and which holds importance for this paper, can be found in the research of Niazi and Hein (2017). In order to put their work into context, it is worth taking a look at another important paper in the field. Azam and Hoeffler look at why the perpetrator uses forced displacement as a tactic. They find evidence that displacement is used to lessen insurgents effectiveness in key strategic areas, to great effect (Azam & Hoeffler, 2002:482). The idea follows that reducing population levels makes it harder for insurgents to blend in and hide, and therefore lessens their effectiveness. The paper which provides new key insights finds that conflicts over natural resources have a significant positive causal effect on forced displacement (Niazi & Hein, 2017:8). This is important for two reasons. First, it has not been previously researched, and is thus a new finding in the field. Second, if verified, will significantly help predict levels of
conflict-induced displacement events in the future, as natural resources are largely static and well-researched (Collier, 2009:127). This may also be beneficial in further research in the field of displacement as it ties in with another well-researched field, the connection between natural resources and armed conflict. There are a few reasons that this paper on its own is not satisfactory, however. First, as a small-n study it is useful for theory building, yet the results are not
generalizable. Second, it does not effectively argue from a theoretical standpoint for its results.
Both of these problems must be remedied for the results to be useful.
This field's frontier, as argued more important than large frameworks above, is building on individual and separate factors. As such, this paper will attempt to remedy the problems of Niazi and Hein’s (2017) article. Holding great promise, the aim is to make their findings useful and generalizable, and building a strong theoretical causal mechanism behind these findings. The first step will be to ensure generalizability, by conducting a large-n study. Second, building on the theory will be done relying on the findings of Azam and Hoeffler’s (2002) paper.
The dependent variable is conflict-induced displacement. This term will here be used as a person being forced to leave their homes due to the threat of or use of violence. This accounts both for refugees and IDPs. What constitutes a refugee was established by the 1951 refugee convention, and refers to a person crossing national borders as a result of fleeing from well-grounded fears of
violence or persecution (UNHCR, 1951). In order to look at both cross- and within border displacement, IDPs will also be measured. An IDP is a person who, like a refugee, has to flee their home but does not cross an international border, and is unlike refugee not a legal term (IOM, 2019:109). By using the term conflict-induced displacement rather than the often used alternatives of refugees or IDPs, two problems are solved. The first is by including IDPs, which are often overlooked in favour of refugees, as they are in comparison less visible and more difficult to
measure. By looking at both categories, we avoid overlooking differences which may affect the two categories differently (Lischer, 2007:150). The second is that we do not wish to research
disaster-induced displacement. While this is undoubtedly also important, this paper will focus on those displaced by armed conflict rather than all human activity (IDMC, 2020a).
The independent variable will be natural resources extraction. Definitions of these concepts are rare, and are often simply defined by a predetermined set of resources, such as specifically oil or minerals, (Ross, 2004:46; Niazi & Hein, 2017:5-7), which the author intends to research. This paper will attempt to capture a broader scope of resources, and will therefore offer up its own definition as; a resource which is geographically bound, not produced through artificial means, and with a substantial monetary value. This should encompass all relevant resources, without including non-monetary values, such as water or wetlands, or non-geographical resources, such as industries or taxable incomes. Research on natural resources are plenty, and these are very well mapped and understood. As such, natural resources are easy to apply to developing theories.
2.2.1 Causal Mechanism
This paper is based on the findings of Niazi and Hein (2017). They found evidence that conflicts over natural resources leads to higher rates of conflict-induced displacement. They find, by doing a qualitative analysis of three armed conflicts, connections between the natural resources and the levels of displaced. However, the causal mechanism is never developed in great depth, and the theory is never tested at a larger scale. My aim is to test this phenomenon on a larger set of data points, thus thus testing the generalizability, and to test the causal mechanism.
The causal mechanism proposed by this paper goes as follows; to extract high value resources, control over the surrounding land is required. This in turn will result in a purposeful displacement of local populations residing in the area, as a means of ensuring control over the resources revenue.
The process will be explained below.
2.2.2 Regional control
The need to control the land surrounding natural resources to guarantee revenue is
well-documented. The economic structure of land ownership has been shown to affect how the conflict develops, however, even more important was the economic value of land (Niazi & Hein, 2017:9). As natural resources require control of land to extract their value, the value of controlling said land can be considered equal to the resources it inhabits, and is directly correlated to strategic value (Collier & Hoeffler, 2004:567; Barnet, 2016:240). It is also noted that control of transit routes to and from the natural resource regions are important, as this is necessary for exports, and by extension revenue (Niazi & Hein, 2017:9). While the idea of transit routes is likely to be true to a certain extent, it will not be given attention in this paper, largely due to limitations in data. A further sign that control of natural resource-rich regions becomes an objective in and of itself, is that armed conflicts in these areas tend to last longer (Lujala, 2010:20). Further, the same study finds that active production and potential production, such as large reserves, have similarly large effects (Lujala, 2010:23). This necessity for control of territory is not just supported by empirics, but also follows previous theories (Azam & Hoeffler, 2002:482). A further cause of the strategic importance of natural resources, is that these have been theorized to function as guarantors in
terms of international support. Collier and Hoeffler (2004:565) proposes a theory that natural resource access can be promised to external actors, in return for startup resources for rebels, and support in terms of arms and/or training, and international recognition of post-conflict regimes.
This is however not found to be the case in another major study (Ross, 2004:51). Instead, no such relationship was found.
It should be mentioned here, that the value of exportable resources functions as a self-feeding cycle.
While the absolute value may not change, the relative value increases with conflict duration, as armed conflicts have a negative impact on economic development and general exports and thus weakens other sectors of the economy (Barnett, 2016:239). As the value of the natural resources thus become a larger section of the economy, and the relative economic gains compared to other sources increases, the value of the land itself will only grow. To make this effect even stronger, resources that could be aimed at economic and social development are instead pumped towards the military, furthering economic deprivation (Barnett, 2016:240). This results in higher relative demand for the natural resource revenue, exacerbating the situation.
The means by which natural resources alters armed conflicts and changes the importance of regional control is established above. Next, this paper will theorize as to why this affects levels of conflict-induced displacement. The second part of this reasoning is not as steadily researched in the past, and must follow a rational pathway of its own. The first, most concrete reason for forceful displacement is to avoid leakage. While leakage of revenue often occurs in favor of foreign actors (Gunton, 2003:69), this may be intended and is a method of ensuring international legitimacy and guaranteed support for the occupying force (Le Billion, 2007:95). A greater problem for the leadership can be internal, and low-tiered leakage of resources (Caselli & Coleman, 2013:162). This may only be relevant in cases where resources are easily accessible however, resources which in research are the same as lootables or contraband (Fearon, 2004:284). This mechanism will only be relevant in a limited amount of cases, as the majority of natural resources do not fit this specific criteria (Collier & Hoeffler, 2004:40). Unfortunately, the research on the topic is sparse as is the
available data on the matter, and the second pathway, which will now be presented, is thus more likely to be observed and remain relevant.
The second possible pathway begins with the risk of attacks on structures or workers extracting the natural resources, as a means of denying the occupying actor income. Displacement can, and has been, used to negate this risk of such attacks. One example is the large-scale displacement campaign in Darfur, South Sudan in 2002 (HRW, 2003). The theory behind one-sided violence fits well into settings of valuable land as found in the statement above. The use of violent targeting of civilians is theorized to be an effective way of maintaining loyalty among said population. One pathway in which this works, is that the use of selective violence promotes compliance, by showing loyalty or the desired behaviour the non-combatants are less likely to be targeted, and thus the violence reduces acts of resistance (Hultman, 2014:291). The use of violence against civilians has been shown to reduce insurgent attacks, and to be a successful preventative measure of violence from insurgents among the local population (Lyall, 2009:349). More importantly, the use of
indiscriminate violence has been a way of clearing out populations, with the express aim of
reducing insurgents' ability to hide and blend in (Lyall, 2009:336). This tactic has also been shown to be directly tied to forced displacement as a means of achieving the same goal. Further, displacing a certain fraction of the population directly lessens the effectiveness of insurgents, for the reasons stated above (Azam & Hoeffler, 2002:482). The use of strategic displacement has previously been used to ensure safety in tactically important or valuable areas (Lyall, 2009:336), a theory which is easily applied to cases of valuable natural resources which require certain zones of control.
The hypothesis is therefore that natural resource extraction in armed conflicts has a positive causal relationship with conflict-induced displacement. This stems from the need to control and ensure the value extracted from the land, which in turn results in displacement campaigns from the
3. Research Design
This paper attempts to find trends in terms of conflict-induced displacement, and as such, will conduct a large-n study. This will be in the form of a quantitative, comparative analysis. As most previous research on this issue is qualitative, a quantitative study will be more likely to contribute to the state of the academic field. This paper will first discuss the available data which will be used.
After which, the dependent and independent variables will be examined and operationalised, followed by a discussion on their validities and reliabilities. Finally, control variables will be discussed.
This paper will rely on four data sources, which have been combined here into one. The data has been collected from the Internal Displacement Monitoring Centre, IDMC, the World Bank and UN High Commissioner for Refugees, UNHCR as well as Uppsala Conflict Data Programme, UCDP (IDMC, 2020a; World Bank, 2020; UNHCR, 2020; UCDP, 2020). By combining these into one dataset, each individual set of data has to be slightly altered, however as no values are removed this should not affect the results if done correctly. The reason that a new dataset was created, instead of relying on older sets of data is twofold. First, several of the older quantitative studies data are inaccessible at the time of this being written. Secondly, previous
operationalizations of the dependent variable have been lacking in validity. The new dataset should solve this issue by more efficiently capturing definitions necessary to understand the phenomenon in question. This comes with a drawback however, as each added variable is a significant added workload. Therefore the new dataset has accurate operationalizations, but fewer variables to examine and to be used as controls. This should not be a problem however, as there are few potential control variables where data is accessible within the time span and at the correct level of analysis. As explained in the literature review, there are few concrete and undisputed findings in the field, and adding variables may only confound the results. As such, this study aims at
simplicity, and should a causal pathway be found, urges future studies to further test this by adding more control variables.
This paper will use the four causal criteria presented by Kellstedt and Whitten as determinants for causal relationship (2018:85-95). The authors present four criteria which must be fulfilled in order to claim that there is a causal link from the independent to the dependent variable. These are covariation, causal mechanism, time order and isolation. Covariation asks whether the two variables covary, if X changes, it requires to see a change in Y. This will be tested using bivariate regression. Causal mechanisms require a reasonable mechanism by which X causes change in Y.
The causal mechanism was explained and motivated in the theory section. Time order requires the change in X to occur before the change in Y, as to ensure there is no reversed causality. Time order will not be tested here, simply due to a lack of realistic mechanism for the dependent variable, conflict-induced displacement, to affect the independent variable, natural resource extraction.
Isolation requires that no other factor is the cause of change in both X and Y. This will be tested using a multivariate regression.
The dataset includes 207 observations across the timespan 2009-2018. The timespan was chosen because 2009 was the first complete year where the new operationalization of IDPs was in use, and the data was last updated in 2018. The 207 observations are taken from all countries during this time-span which experienced armed conflict during that year. The dataset was merged using R Studio, and is easily replicable. The method used for the regressions was OLS (Ordinary Least Squares). Regression will be used to satisfy isolation and covariation, as explained above. A simplified explanation of OLS is: a method which fits a straight line along the observed values of the dependent variable as such that the total sum of the squared differences will be as small as possible. Applying this in a bivariate regression shows the extent of covariation, while doing so in a multivariate regression using control variables will help ascertain isolation (Kellstedt & Whitten, 2018:198).
3.2 Dependent variable
This paper attempts to capture conflict-induced displacement as fully as possible. However, most data collections divide this concept into three subcategories: IDPs, refugees and asylum seekers.
While these concepts can be somewhat blunt and, occasionally, difficult to interpret (Hruschka &
Leboeuf, 2019), they are generally standardised over both time and space, and are thus easily replicable. While some sources do list these as one variable, they tend to use all forms of forced displacement, including non-violent persecution and disaster-induced displacement, rather than specifically conflict-induced. For this reason this paper will divide these categories upp individually and then add them together for a joint conflict-induced displacement value.
The first of these subconcepts is conflict-induced IDPs. This measure will use a rather thin operationalization, as used by IDMC. The reason that data is taken from IDMC, is because they have reliably measured IDPs based on conflict- and disaster-induced displacement separately, and thus have a wide array of comparable data points. By allowing for the disregard of disaster-induced displacement, it can be guaranteed that validity is high, and that any results are the effect of the conflict rather than other factors such as natural hazards. Further, IDMC collects data from a vast array of primary and secondary sources, reducing the risk of source bias (IDMC, 2020a). However, there is still bias in which conflicts are more heavily researched, and what groups remain invisible (Hruschka & Leboeuf, 2019;IDMC, 2020a). A final strength of the IDMC is that other similar studies use their data as well, and therefore any results will be more comparable.
Data on refugees will be gathered from UNHCR, and their operationalization will therefore be used. Refugees, as people who flee from violence or oppression, are functionally IDPs who travel across borders. While this difference may be interesting for certain research, and is vital for policymakers, it is outside the scope of this paper. There is one issue with using refugees as an indicator however. Unlike the data for IDPs, data on refugees separated into conflict-induced and non-conflict-induced has not been collected. As such, a certain section of refugees will be
generated by disasters or oppression. As UNHCR uses the legal definition as their
operationalization, conflict-induction cannot be isolated (UNHCR, 2018:28). However, since such isolated data is not available, this will be the closest non-primary research can get to a
functionally valid category of conflict-induced displacement. Similar to IDMC, this data source is often used by other researchers and will therefore be easily comparable, and as the data is compiled directly from primary sources it can be expected that the data is reliable (UNHCR, 2020).
Data on asylum seekers is also taken from primary sources within UNHCR. While asylum seekers are fewer than refugees and IDPs (UNHCR), they are nonetheless included in the category of conflict-induced displacements. The difference between refugees and asylum seekers is at which stage in the process of obtaining legal status (UNHCR, 2015:27). As such, asylum seekers as an indicator have the same issue as refugees, in that it does not isolate conflict as a cause. Despite this, ignoring them may have a greater negative impact than including them, and they will thus
represent the final part of conflict-induced displacements.
The three categories will be summarised to create a category called total displaced. However, as there may be other causes of the forced displacements, any cases where there are refugees or asylum seekers but no IDPs will be excluded from this list. The reasoning follows that the likelihood that people are forced to flee but that no one is unable to leave the country is highly unlikely, and thus the cause of the displacement is likely oppression or disaster rather than conflict. While this may omit certain data points from the analysis, it will increase the validity of results. In the end, both the total and IDP datasets will be used for the regressions. While the total forced displacements captures a greater set of cases, and may thus be more accurate in scope, the analysis of IDPs alone may help be valid as this more accurately captures the definitions proposed in the theory section.
3.3 Independent Variable
The independent variable, natural resource extraction, will be using a thin operationalization. In order to measure the value in a quantitative study, it is made quantifiable. While there are several ways to do this, each of which may yield interesting results, this paper will use a portion of GDP based on natural resources. This provides us with several advantages and disadvantages. First, it provides us with clear quantifiable, continuous values. Further, it directly measures the economic value of the resource. Since it is the value of natural resources that is hypothesised to result in the proposed effect, this is the most important aspect. This further shows the value of the resource in relation to alternate revenue. The function of economic gains in armed conflict is directly
dependent on alternate sources of income in a manner that people are more likely to fight over resources which are relatively valuable compared to alternate sources of income (Collier &
Hoeffler, 2004:569). One problem could be that it does not say that the displacement is occurring near the extraction, as both are measured at a country level. However, if a strong, statistically significant link is found this is unlikely to be a significant problem. There may however be a somewhat different causal mechanism, for example through the mismanagement theory proposed by Sachs and Warner (2001:836). These causal mechanisms tend to explain why armed conflicts start however, rather than its specific dynamics and how this affects forced displacement. A final problem would be what the data taken from the Worldbank considers natural resources; gas, oil, coal, minerals and forests (World Bank, 2020a). While the first four fit the causal mechanism proposed well in that they are densely populated and often very valuable, this may not be the case of forests. However, two results from this may arise. Either, forests are valuable, in which case they may fit the proposed causal mechanisms, or they are not valuable, in which case they are unlikely to significantly affect the data. In either case the results should remain relatively valid. Finally, this data and operationalization are more valid than the majority of previous research in the field.
Others are either qualitative or leave their exact data unspecified (Ross, 2004; Collier & Hoeffler, 2004:595).
3.4 Control Variables
This paper will use three control variables: population size, economic development and conflict intensity. The first, population size, is taken from the World Bank, and functions as a continuous variable (World Bank, 2020c). This follows on the findings of Melander and Öberg (2006), which claim that increased population sizes tend to generate larger quantities of displacements. While this may at first glance also seem to follow the argument that population density may function as a proxy for land value (Raleigh & Hegre, 2009:32-33), population is not measured in density (Worldbank). As previously mentioned, the validity of said claim is dubious, and the population here measured a total population in said country rather than density, again due to limitations in available data.
The GNI per capita values are also taken from the World Bank, and in a similar fashion functions as a continuous variable. It is operationalized along the world bank's definition as the gross value of
all citizens' production measured using standard price indexes (World Bank, 2020b). This will give a feel of the economic development of the country. While there are several criticisms of this as a measure of economic development (Prakash, 2013:10), it is functional in this case for two reasons.
First, situations of natural resource extraction are closely linked to inequalities in economic gain (Sachs and Warner, 2001:837). Thus, it can be expected that inequality will be present in the majority of cases here examined, and as such economic development may be more relevant than measures of equality. Second, there is little available data on household income or wages within the selected time span for relevant countries. As such this is the closest approximation of economic development to be found.
The third control variable, intensity levels, is taken from Uppsala Conflict Data Programme, or UCDP. The dataset called Armed Conflict Data will be used. This is an ordinal variable, with three separate values. If there are less than 25 battle-related deaths in the calendar year, it will be
registered as NA, as any intensity registered is below the armed conflict threshold. The numerical value one will be given to minor armed conflicts, where battle-related deaths fall between 25 and 999, and the numerical value two given to any major armed conflicts, which will see more than 999 battle-related deaths (UCDP, 2020). This choice was made as continuous versions of these
variables were given for dyads rather than at a country-level, and would therefore not be applicable to the rest of the collected data. Further, it is the same data and method used by previous studies, ensuring the results are comparable (Melander & Öberg, 2006:139). Intensity levels have, as previously been stated, found mixed results as to their relevance to forced-displacement. Thus, replicating data following the same methods as previous papers can be useful to understand causes behind conflict-induced displacement.
4. Results and Analysis
While causal mechanism and time order have been discussed, covariation and isolation remains untested. First, the descriptive statistics of the variables will be discussed and dissected. Second, a look at the bivariate regressions using both IDPs and total forced displacement will be reviewed and the correlation investigated. Third, the results will be analyzed including the control variables, and the causal mechanism will be discussed. Finally, the results will be analysed and examined to conclude how well the hypothesis fared as compared to the data found.
4.1 Descriptive statistics
Below, the descriptive statistics of all the relevant variables can be found summarized, with their 207 observations. As the table was taken directly from R Studio, some of the names may need clarifying. conflict_new_displacement represents IDPs, sum_displaced is the total displaced value, i.e. the sum of IDPs, refugees and asylum seekers. gdp_portion is the independent value,
GNI_value represents the GNI level, and intensity_level is the level of intensity.
Table 1 Descriptive statistics
Statistic N Mean St. Dev. Min Max
conflict_new_displacement 207 195,663.200 387,238.700 2 3,000,000
refugees 207 249,708.700 562,455.000 48 3,054,699
asylum seekers 207 35,536.070 58,656.490 110 369,061
sum_displaced 207 480,908.000 747,259.300 495 3,523,508
gdp_portion 207 9.040 10.137 0.001 51.152
GNI_value 206 6,386.019 5,206.067 770.000 30,080.000
population 207 127,809,601.000 270,497,960.000 4,248,334 1,352,617,328
intensity_level 130 1.262 0.441 1.000 2.000
The first value which needs to be examined is the independent variable, natural resources. By looking at the max and min values, and comparing these to the mean, it can easily be seen that the distribution is skewed heavily towards lower values of natural resource extraction, with a few large outliers with heavy dependence on natural resources. This can be seen in figure 1, showing a density plot of natural resource value. This can be compared to our measures of the dependent variable. Total displaced shows a similar distribution, with a heavy quantity of observations of lower levels of displacements, but a few major outliers, as can be seen in figure 2. IDPs too show a similar relationship as well, as seen in figure 3. This may hint at some level of covariation, as is expected at this stage.
4.2 Bivariate regression
Across 207 observations, no statistical significance could be tied to the impact of natural resource extraction to conflict-induced displacement when using the total displaced measure even at the 90% confidence level. This can be seen in table 2, showing the statistical significance of the covariation. As the p-value is greater than 0.1, it cannot be considered generalizable (Kellstedt &
Whitten, 2018:179). Further, while the correlation coefficient is positive, it is also small, and there is little correlation between the two variables. Whether this shows that the null hypothesis is correct, or if there is a problem with validity, can be tested using our second dependent variable, IDPs.
Table 2: Bivariate regression
--- Total displacements --- gdp_portion 3,097.010 (5,143.839)
--- Observations 207 R2 0.002 Adjusted R2 -0.003
Residual Std. Error 748,418.300 (df = 205) F Statistic 0.363 (df = 1; 205)
Note: *p<0.1; **p<0.05; ***p<0.01
The same amount of observations for IDPs provides a different view of the correlation. At the 95%
confidence level, thus having a p-value less than 0.05, the results are statistically significant and can be generalised to the larger population (Kellstedt & Whitten, 2018:179). Further the correlation coefficient is higher, showing a stronger correlation between the two variables.
Table 3: Bivariate regression
--- IDPs --- gdp_portion 5,631.967**
--- Observations 207 R2 0.022 Adjusted R2 0.017
Residual Std. Error 383,939.800 (df = 205) F Statistic 4.555** (df = 1; 205)
Note: *p<0.1; **p<0.05; ***p<0.01
4.3 Multiple Regression
The results of the bivariate regressions were detailed above. Next, the multiple regression using total displacements should be examined. First, looking at the independent value, it remains ungeneralizable, as expected. However, the coefficient has switched signs and now has a negative relationship with the dependent variable. The same negative relationship can be seen with all the control variables as well, aside from intensity level. As for statistical significance, the only variables that are generalizable are intensity level and GNI. These results alone do not support this paper's hypothesis that armed conflict over natural resources results in larger quantities of
Table 4: Bivariate regression
--- Total displacements --- gdp_portion -2,599.392 (5,278.480)
population -0.0002 (0.0002)
--- Observations 129 R2 0.398 Adjusted R2 0.379
Residual Std. Error 669,538.100 (df = 124) F Statistic 20.527*** (df = 4; 124)
Note: *p<0.1; **p<0.05; ***p<0.01
Using IDPs as the measure of the dependent variable provides a different picture. While the independent variable is no longer statistically significant at the 95% level, it remains significant at 90%. Despite being less generalizable, the coefficient shows a larger correlation between the independent and dependent variable for the cases shown. All the control variables had a positive coefficient, as opposed to the previous regression. However, none of the controls was statistically significant, except for intensity level, which was significant at the 99% level. Based on this
observation alone, it can therefore be said that while controlling for other causes, natural resource extraction is likely to have an impact on the level of conflict-induced displacements, but does not reach the 95% standard for generalizability.
Table5: multivariate regression
--- IDPs --- gdp_portion 5,253.133*
population 0.0001 (0.0001)
GNI_value 0.203 (6.349)
--- Observations 129 R2 0.164 Adjusted R2 0.138
Residual Std. Error 372,044.900 (df = 124) F Statistic 6.103*** (df = 4; 124)
Note: *p<0.1; **p<0.05; ***p<0.01
There are several phenomena of interest in the results of the regressions. First is the variation between the two selected measurements of the dependent variable. As previously discussed, this may be caused either by inherent differences between within- and cross border displacements, or the portion of the total forced displacement measure which was not caused by armed conflict is larger than anticipated. Both these options should be discussed, to determine how likely they are to cause this difference of results.
There are several factors determining whether a person becomes an IDP or refugee in the event of armed conflict. Two prevailing theories are the expected wages found in neighbouring states compared to local economic opportunity, and level of violence used against the non-combatants
(Moore & Shellman, 2006:619; Prakash, 2013:7). Neither of these theories would support lower levels of displacements across borders in armed conflicts with natural resource extraction. For one, as previously mentioned, income tends to be lower in states which rely on natural resources (Sachs and Warner, 2001:836), Therefore likelihood of relative economic opportunity in neighbouring states is higher. Further, as previously established, geographic areas associated with high monetary value are correlated with greater levels of targeting of civilians (Lyall, 2009:336). Following this, an increase in cross-border displacement should thus be observed, as opposed to the results of the regressions displayed in tables 2 and 4. This leaves a final theory suggesting that a base level of resources are required for cross border relocation, due to higher social and economic costs, leaving those without said economic capabilities behind as IDPs (Davenport, Moore & Poe, 2003:41;
Moore & Shellman, 2006:605). This would support an inherent difference between the
within-border and cross border displacements that correlate with the observed results, and could affect the hypothesis. Alternatively, the differences in results could be caused by the portion of total displaced which are not conflict-induced displacements. However, how large a portion of the results are represented by this number is difficult to know, as there is no data on levels of
conflict-induced displacement for cross border displacements.
Notably, adding armed conflict as a control flipped all other variables to negative in table 4. This would entail that the independent variable would have a negative impact on displacement. The theory mentioned above would allow armed conflict intensity to cause a negative relationship with natural resource extraction. This follows as the theory claims that cross border displacement only occurs at a base level of wealth (Davenport, Moore & Poe, 2003:41; Moore & Shellman,
2006:605). As income is often worse in regions dependent on natural resources (Sachs & Warner, 2001:836), this could explain the difference between the measurements. Following this logic, the same mechanisms as in table 5 are observed, with the exception that lower levels of income associated with natural resource extraction has a greater negative effect on cross border
displacement due to high costs than the positive effect on internal displacement. Thus, the sum total of conflict-induced displacements could be negative.
As has previously been motivated, intensity levels have had mixed results on displacement levels in previous studies (Melander & Öberg, 2006:144; Magnus & Öberg, 2009:166). Finding such strong correlation here is therefore somewhat surprising. However, it is not guaranteed that the causal mechanism is necessarily as strong as suggested. It is possible that the covariation between the dependent variable and conflict intensity levels themselves stem from the independent variable.
Natural resource extraction itself could cause the rising intensity in armed conflict. As explained in the theory section, the act of extracting natural resources results in grievances both in terms of land ownership, but also in terms of poor economic development and high levels of inequality, thus fueling armed conflicts (Collier & Hoeffler, 2004; Sachs and Warner, 2001:837). Further, fighting is not unlikely to be fiercer surrounding valuable resources, and the fighting over the revenue they provide could play a part in this data (Ross, 2004:64). If this is true, the independent variable could cause the change in displacement observed partly through increased conflict intensity, allowing intensity levels to function as a mediating variable. Since the data collected is at a national level, it cannot be determined whether fighting was fiercer surrounding the natural resources, and the concrete effects of natural resource extraction on armed conflict intensity is beyond the scope of this paper.
As for the other control variables, these relationships are less surprising, at least concerning table 5.
Population size was expected to see a positive correlation with conflict-induced displacement (Magnus & Öberg, 2006), although the results found are not statistically significant in the least. As for GNI, its effects were both small and statistically insignificant. This too was expected based on previous research (Prakash, 2013:129). It is however odd that the coefficient was positive. The results of GNI per capita in table 4 are also expected. An increase in economic activity should decrease potential economic gain from the crossing of borders, which some have found to be an contributing factor to cross border displacement (Moore & Shellman, 2004:619). While it is surprising that contrary to previous large-n studies (Magnus & Öberg, 2009:158), these results are somewhat generalizable, at the 90% level, they thus follow the results of qualitative research, who find correlation yet not to a statistically significant level (Prakash, 2013:129-130). The same effect as was observed with GNI per capita in table 5 can be observed regarding population in table 4.
The coefficient has the opposite from expected results, but they are neither large nor statistically significant.
With these factors considered, it can be determined whether the hypothesis was correct. There was a somewhat generalizable correlation between conflict-induced IDPs and natural resource
extraction in armed conflict, and if conflict intensity does have a mediating role, then this is a statistically significant relationship. This may be caused by both actions to reduce leakage, as well as means to protect natural resource revenue from sabotage. The same relationship was not found in regards to total displacement. Whether this difference was caused by inherent differences between within- and cross border displacement such as difference in costs and opportunities, or a lack of validity for the broader measurement, is as of now undetermined. However, it is possible that both these had an effect on the results.
Finally, the issues with this paper should be discussed. This paper has attempted to capture the relevant concepts as well as possible. However, questions regarding validity may be raised. The first is the independent variable. While it does capture both value and quantity of natural resources, it does include forestry. While this should not be a major problem, it may somewhat affect the data.
Second, are the measurements of the dependent variable, and has similarly to the independent variable been discussed previously. Third are the controls. Both conflict intensity and population are straightforward, yet GNI per capita stands out somewhat. This attempts to capture economic opportunities and macro-economic trends simultaneously. However, to more accurately capture economic opportunity, income would fit better. The reason this was not chosen is since accurate data is lacking within the relevant timeframe. These issues with validity are further discussed in the research design section. A general lack of appropriate data is a significant issue for this study. Data at local levels, for all measurements, would have been beneficial and may have provided more precise results. More importantly, the lack of data on conflict-induced cross border displacement is particularly problematic. Adding a variable measuring geographic spread would also have been useful, as it is one a strong predictor of conflict-induced displacement, yet again the data is nonexistent within the relevant timeframe. Further, there is the possibility of other potential control variables that could describe the relationship even better, and thus could affect the results.
While this would be useful to include, not all variables can be included, and the ones presented above were selected based on relevance and available data.
This paper attempted to find the effect that natural resource extraction during armed conflict has on conflict induced displacement. Following the previous literature within the field, the hypothesis natural resource extraction in armed conflicts has a positive causal relationship with conflict-induced displacement was formed. This follows the logic that in order to reduce leakage and sabotage, the occupying force will use displacement as a strategy. This theory was tested using a quantitative study, building a new dataset from IDMC, UNHCR, the World Bank and UCDP. This new dataset has a key advantage in terms of the research field, as it includes a measure specifically measuring conflict-induced displacement, although this is only available for IDPs, as opposed to the broader term forced displacement. Using conflict-induced IDPs, as well as total levels of forced displacement, two bivariate regressions were done using the portion of GDP represented by natural resources as the independent variable. After this, a multivariate regression was done for both measures of conflict induced displacement, controlling for population size, conflict intensity and economic development.
This paper found a series of unexpected results. First and foremost, only conflict-induced displacements of IDPs followed the expected results in a generalizable manner, even if it did not reach the 95%-threshold when controlling for armed conflict intensity. Whether the broader operationalization did not have the expected findings due to a difference between within- or cross-border displacement, or if the new use of specific conflict-induced data was the differing factor is undetermined. While the results are thus not certain enough to prove the hypothesis correct, it has shown that there is a likely causal pathway from natural resources to
conflict-induced displacement regarding at least within-border displacement.
Other interesting findings were that conflict intensity did have a generalizable and strong positive causal relationship with both measures of displacement. While this is not a new finding within the field (Melander & Öberg, 2006:144), it does contradict previous papers (Magnus & Öberg,
2009:166). Whether intensity levels function as a mediating variable between natural resources and displacement is as of now unclear, yet would fit the model well. The multivariate regressions further found support for the theory that there is a higher cost for relocating across borders rather than within (Davenport, Moore & Poe, 2003:41; Moore & Shellman, 2006:605). Thus, in contexts with heavy reliance on natural resources, and thus lower average income and wealth (Sachs &
Warner, 2001:836), fewer can afford the costs associated with crossing borders. As such, these cases should see a higher proportion of IDPs as compared to cases with lower reliance on natural
resources, where average income and wealth should be higher.
Finally, this paper has identified several areas for future research. Quantitative studies using data on conflict-induced displacement at sub-national level could closer examine the causal pathways, and look at intervening variables with greater accuracy. This would help establish the exact effects that armed conflicts over natural resources have on displacement patterns. Possibly the most interesting puzzle found in this study, is that of effects on within- and cross border displacement. Finding how the two differ in relation to the independent variable would shed light on many of the questions raised in this paper. This would be greatly helped by building datasets on cross border
conflict-induced displacement specifically. Further, while there are plenty of studies examining the effects of natural resources on intensity levels of armed conflict, looking at how this violence is directed could be useful. The implications and effects of conflict between two or more armed groups differ from those where the violence is one-sided. Finally, an interesting future puzzle is running this data with a reliable dataset using income levels and economic opportunity, to find differences between the effects of economic development at large and household income in particular.
There are several key implications of this study. The results might be useful in predicting
displacement events. Natural resources tend to be well-mapped (Collier, 2009:127), and as such it will be easier to predict the movement of displaced people within these regions. Further, greater
resilience can be built in these communities, and in the management of natural resources in regions where there is risk of armed conflict. This study has also contributed to the discussion on armed conflict intensity and its relationship with conflict-induced displacement. Finally, a new dataset has been built that, while there is room for improvement, allows for testing new more precise measures of conflict-induced displacement.
Azam, Jean-Paul., Hoeffler, Anke. 2002. “Violence against civilians in civil war: Looting or terror?”
Journal of Peace Research. 39 (4): 461-85.
Barnett, Jon. 2016. “Environmental Security”. In Collins, Alan. (ed). “Contemporary Security Studies”. 229-147. Oxford University Press. Oxford, England.
Billion, Philippe Le. 2007. “Securing Transparency: Armed conflicts and the management of natural resource revenues: Armed Conflicts and the Management of Natural Resource Revenues”.
March. International Journal. 61(1):93-107
Caselli, Francesco., Coleman II, Wilbur John. 2013. “On the theory of ethnic conflict”. January.
Journal of the European Economic Association. 11(1):160-192.
Chassang, Sylvain., Miquel, Gerard. 2009. “Economic Shocks and Civil War”. 20 october.
Quarterly journal of Political Science. 4(3):211-228
Collier, Paul., Hoeffler, Anke. 2004. “Greed and Grievance in Civil War”. 4 October. Oxford Economic Papers. 56(4):563-595
Collier, Paul. 2009. “Wars, guns & votes: Democracy in dangerous places”. Vintage Books, London.
Davenport, Christian,. Moore, Will,. Poe, Steven. 2003. “SOMETIMES YOU JUST HAVE TO LEAVE: DOMESTIC THREATS AND FORCED MIGRATION, 1964–1989”. International Interactions. 29(1):27-55
Eurostat. 2018. “Expert Group on Refugee and Internally Displaced Persons Statistics”. March.
[Online]. European Union and United Nations. [Accessed on 2nd January 2021]. Available at:
Fearon, James. 2004. “Why Do Some Civil Wars Last so Much Longer than Others?”. May.
Journal of Peace Research. 41(3):275-301
Gunton, Thomas. 2003. “Natural resources and regional development: An assessment of
dependency and comparative advantage paradigms”. January. Economic Geography. 79(1):67-94.
Hruschka ,C., Leboeuf. 2019. “Vulnerability: A Buzzword or a Standard for Migration”20 January. [Online]. Population Europe. 20. [Accessed on 9th January 2021]. Available at:
HRW. 2003. “LUNDIN: WILLFULLY BLIND TO DEVASTATION IN BLOCK 5A”.
[Online]. Human Rights Watch. [Accessed on 2nd January 2021]. Available at:
Hultman, Lisa (2014) Violence Against Civilians. In: Edward Newman & Karl Derouen Jr. (eds) Routledge Handbook of Civil Wars. New York: Routledge, 289–299.
IDMC. 2020b. “GLOBAL INTERNAL DISPLACEMENT DATABASE”. [Online]. Internal Displacement Monitoring Centre. [Accessed on 2nd January 2021]. Available at:
IDMC. 2020a. “Global Report on Internal Displacement 2020”. [Online]. Internal Displacement Monitoring Centre. [Accessed on 2 January 2021]. Available at:
IOM. 2019. “Glossary on Migration”. [Online]. International Organization for Migration.
[Accessed on 10th January 2021]. Available at:
Kellstedt, Paul., Whitten, Guy. 2018. “The fundamentals of political science research”. Third edition. Cambridge University Press, USA.
Lischer, Sarah Kenyon. 2009. “Causes and Consequences of Conflict-Induced Displacement”. 15 August. Civil Wars. 9(2):142-155
Lyall, Jason (2009) “Does Indiscriminate Violence Incite Insurgent Attacks? Evidence from Chechnya,” Journal of Conflict Resolution. 53(3): 331–362
Melander, Erik., Öberg, Magnus. 2006. “Time to Go? Duration Dependence in Forced Migration”. January 26. International interaction. 32(2):129-152