Related Literature

Civilians’ migration calculus depends on many factors: violence, which drives refugee flightFootnote ¹³ and affects civilians’ destinations,Footnote ¹⁴ travel costs,Footnote ¹⁵ and economic opportunity at home and abroad.Footnote ¹⁶ Scholars have identified many policies that affect rates of refugee arrivals. Physical barriers deter some migrants,Footnote ¹⁷ though sometimes migration instead shifts to less fortified crossings.Footnote ¹⁸ Policies that make access to asylum more difficult are associated with fewer applications,Footnote ¹⁹ and can have spillover effects, leading to fewer arrivals in neighboring countries.Footnote ²⁰ Liberalizing policies have the opposite effect; making asylum easier to attain increases applications.Footnote ²¹ Many refugees choose destinations accounting for policy differences across receiving states.Footnote ²² Others, though less informed, seek refuge in places they suspect will be hospitable.Footnote ²³

States’ willingness to accept refugees depends on factors including domestic attitudes, international pressure, and economic conditions.Footnote ²⁴ Donors also provide aid strategically to conflict-affected states,Footnote ²⁵ hoping development will reduce migration.Footnote ²⁶ However, even targeted aid can create conflict between groups in recipient communities, potentially leading aid to be diverted from those most in need.Footnote ²⁷

I build on models that link economic activity and conflictFootnote ²⁸ and similarly find economic opportunity can motivate migration.Footnote ²⁹ My model is closely related to Sun which studies internal displacement and, in an extension, how violence affects civilians’ choice to remain, migrate internally, or flee abroad.Footnote ³⁰ I depart from that work in focusing on the effects of donor influence on civilian flight and government tactics. Using a richer model of civilians’ migration decisions, I show how different policies that affect civilians determine refugee flows. These features allow me to derive policy implications that could not be studied in previous work. Further, I find that migration can result in more repression because donor pressure changes the government’s incentives. This contrasts with Sun, which finds that increases in refugee flight reduce violence against civilians.Footnote ³¹ In this model, as in Sun, if the government is only concerned about a domestic conflict, it benefits from refugee flight, relying on displacement as a substitute for repression. I go beyond these results and show that when the government also responds to a foreign donor, outside pressure can instead complement its own incentive to repress as a means to deny resources to a rebel group.

The Model

I study civil conflict where a government combats an opponent by minimizing its economic capacity. Fighting may push civilians to flee as refugees. Government tactics and civilians’ migration choice depend on each other, and on constraints created by foreign donors.

Utilities

A civilian’s expected utility, accounting for their individual migration cost, ${\theta _i}$ , is

$${U_i}\left( {{m_i},x,v;{\theta _i}} \right) = {m_i}\left[ {\left( {1 - v} \right)\left( {b - \gamma {\theta _i}} \right)} \right] + \left( {1 - {m_i}} \right)\left[ {\left( {1 - v} \right)\left( {{Y_L}\left( {\left( {1 - v} \right)\left( {1 - \mu } \right)} \right) - cx} \right)} \right].$$

Civilians who remain in $H$ are indirectly affected by those other civilians who choose to migrate; economic prospects depend on the refugee migration rate, $\mu = \mathop \int \nolimits_0^1 {m_i}di$ , which endogenously determines wages. At the end of the game, civilians comprise three groups: (1) refugees; (2) civilians who remain in $H$ by choice; and (3) victims of repression, given by $v$ .

The government minimizes rents that accrue to the rebel group while maximizing its own returns from foreign aid. It chooses levels of fighting effort and repression that solve,

(1) $$\mathop {{\rm{min}}}\limits_{x \in \left[ {0,1\left] {,v \in } \right[0,1} \right]} \left( {1 - x} \right) \cdot {\tau _r}\left( {Y\left( {\left( {1 - v} \right)\left( {1 - \mu } \right)} \right)} \right) - \alpha f\left( {\mu ,v} \right) + a\left( x \right) + z\left( v \right).$$

I refer to the government’s objective in Equation (1) as $g\left( {x,v} \right)$ , with subscripts denoting partial derivatives (for example, ${g_{xv}}\left( {x,v} \right)$ is the cross-partial).

The timing of the game is as follows:

1. 1. The government chooses levels of fighting effort, $x \in \left[ {0,1} \right]$ , and repression, $v \in \left[ {0,1} \right]$ .

2. 2. Civilians decide whether to remain, ${m_i} = 0$ , or flee as refugees, ${m_i} = 1$ where the refugee migration rate is $\mu = \mathop \int \nolimits_0^1 {m_i}di$ .

The solution concept is subgame perfect equilibrium. Civilians’ strategy is captured by a cutoff in $\theta $ , which determines the proportion of civilians who flee abroad. The government’s strategy comprises levels of fighting effort and repression that account for refugee flight.

Comments on the Model

Before proceeding to the analysis, I offer brief comment on some features of the model. First, I deliberately black box some aspects of the conflict, notably by omitting the rebel group whose revenue the government minimizes. Governments’ motivations for controlling territory and denying rents to rebels are well studied,Footnote ³⁶ and other models consider competition between states and rebels over a civilian labor pool.Footnote ³⁷ In line with the experimental perspective in Paine and Tyson, I deliberately exclude certain factors—like rebels’ actions—from the model in favor of carefully explicating the mechanisms of interest—the effects of donor leverage and priorities on violence and refugee flight.Footnote ³⁸ In omitting the rebel group, I can focus on clearly defining how governments that value foreign aid respond to donor pressure, incentives that sit alongside concerns about fighting rebels.

I make no assumption about civilians’ ideological alignment with the rebels or the government and civilians see no direct benefits from the government retaking contested areas. The government may privilege elites in redistributing recaptured territory, and any increases in public goods following the reassertion of state control may come too late to affect civilians’ migration choice.Footnote ³⁹ In Appendix C, I allow the government to directly value controlling territory, modifying its objective so that instead of minimizing rebel revenue, it instead maximizes its own rents from controlling $H$ . I find the effects of this alternative specification on refugee flight depend on whether civilians can freely move from rebel to government-controlled land, in addition to fleeing abroad.

By studying the effects of violence in partial equilibrium, I can clearly trace the impact of donor policies on migration through their effect on violence. This reduced-form representation of civilians’ economic incentives captures an important feature of their migration calculusFootnote ⁴⁰ and reflects, intuitively, how civilians become more valuable to a rebel group as its support base shrinks. Given my findings are largely consistent with past empirical studies, which mostly estimate general equilibrium effects, the mechanisms I identify as driving violence likely complement other mechanisms identified in the literature.

Foreign Assistance and Refugee Flows

I first evaluate how donor policies affect the flow of refugees from $H$ , holding fixed levels of fighting and repression. These benchmark results show how donor incentives endogenously drive migration and contrast with outcomes when the government fights strategically. The comparison then highlights the challenges facing donors intervening to aid civilians in conflict.

Civilians flee when the expected benefits are greater than the costs, reflecting an assessment of both their well-being in $H$ and prospects abroad. The civilian indifferent between remaining and fleeing has migration cost that solves $b - \gamma {\theta _i} = {Y_L}\left( {\left( {1 - v} \right)\left( {1 - \mu } \right)} \right) - cx$ , where repression drops out of this calculus because victims of repression cannot choose to flee. Rearranging, the threshold below which civilians flee and above which they remain is represented by ${\theta ^{\rm{*}}} = {1 \over \gamma }\left( {b - {Y_L}\left( {\left( {1 - v} \right)\left( {1 - \mu } \right)} \right) + cx} \right)$ . This characterizes the unique equilibrium migration rate, as well as labor force participation in $H$ . Given the uniform distribution over migration costs, the equilibrium migration rate is,

(2) $${\mu ^{\rm{*}}} = {1 \over \gamma }\left( {b - {Y_L}\left( {\left( {1 - v} \right)\left( {1 - {\mu ^{\rm{*}}}} \right)} \right) + cx} \right).$$

The proportion of civilians who choose to remain in $H$ determines wages and thus affects this threshold and who decides to flee. Lemma 1 restates civilians’ equilibrium choices.

Proof of this, and all other results, are in Appendix A. Intuitively, civilians with relatively low migration costs are those who flee. How do donor policies influence civilians’ decisions?

Among the tools available to policymakers, policies affecting the gains of migration most directly impact refugee flows. By increasing barriers to asylum or restricting access to social services, a potential host state endeavors to become a less appealing destination. Providing humanitarian relief, on the other hand, mitigates the cost of conflict for civilians remaining in $H$ , which may discourage migration. I find these policies have their intended effects.

This mirrors how liberal policies allowing asylum seekers to work attracted large numbers of refugees to South Africa after the end of apartheid.Footnote ⁴¹ In contrast, recent changes to restrict work permits and limit asylum claims have seemingly worked to reduce immigration to Sweden.Footnote ⁴² Evidence from Europe suggests labor market conditions may have limited effect on asylum applications but, consistent with my model, easing access to asylum increases migration.Footnote ⁴³

Strategic Government Violence

I next establish the government’s optimal strategy to contest the conflict in $H$ and return to the effects of donor policies in the next section. Proceeding backwards, the equilibrium migration rate follows from Lemma 1. Now, however, the government strategically chooses levels of fighting and repression. How does the intensity of conflict affect refugee flight?

Lemma 3 highlights a core tension in the model. When the government directly fights the rebels, this increases costs borne by civilians who elect to remain in $H$ . More intense fighting, therefore, increases refugee migration. Repression, on the other hand, reduces migration for two reasons. First, as the government increases repression, since victims cannot flee, mechanically the number of civilians who leave $H$ decreases. Second, repression indirectly affects economic opportunity for civilians who choose to remain in $H$ . Victims of repression also cannot participate in the economy, making an individual’s labor more valuable—in the face of repression, each civilian is more instrumental to the rebel economy. Remaining civilians see marginal wage increases resulting from a tighter labor market, which may be sufficient for some with relatively high migration costs to elect to remain.

While some evidence suggests wages decrease during conflict,Footnote ⁴⁴ others find, closer to my result, that labor markets help absorb conflict shocks, such that violence shocks have no statistically significant effect on household consumption.Footnote ⁴⁵ Repression also has economic repercussions in terms of land or property. For example, land owned or occupied by conflict victims in Colombia was often redistributed by paramilitaries, creating opportunities for private gain.Footnote ⁴⁶ The indirect migration-reducing effect of repression suggests that some civilians, when they become more valuable to the rebels, are deterred from fleeing abroad, though they may be unwitting beneficiaries rather than deliberately seeking to create these economic effects.

Figure 1 visualizes how fighting increases refugee flight and repression decreases migration. Figure 1(c) also highlights the relationship between fighting and repression, where as the level of repression increases, more intense fighting moderates the deterrent effect of repression.

Figure 1.

Effects of violence on refugee migration

Note: I fix $\gamma=3.25$ , $b=1.5$ , $Y = 1.65{L^{{2 \over 3}}}$ and $c=.35$ .

Figure 1 Long description

The image contains three line graphs analyzing the effects of violence on refugee migration. The first graph on the left shows the effect of fighting effort on migration rate, with three lines representing different levels of repression (v = 0.25, v = 0.5, v = 0.75). The migration rate increases with higher fighting effort for all levels of repression. The second graph on the right illustrates the effect of repression on migration rate, with three lines representing different levels of fighting effort (x = 0.25, x = 0.5, x = 0.75). The migration rate decreases as repression increases, with higher fighting effort showing a more pronounced decline. The third graph at the bottom shows the effect of repression on the size of refugee outflow, with three lines representing different levels of fighting effort (x = 0.25, x = 0.5, x = 0.75). The refugee outflow decreases as repression increases. All values are derived from numerical solutions to parameterized versions of the model.

Turning to the government’s strategy, its aim is to defeat the rebel group by undermining its capacity, fighting and repressing civilians to compromise its ability to generate revenue. To understand how these incentives are amplified or constrained by donor pressure, I first show the government’s optimal use of force in the absence of foreign aid.

If the cost of fighting is low, $a'\left( 1 \right) \le \bar a$ , then the government chooses maximal fighting effort, ${x^*} = 1$ . It is never optimal for the government to choose maximal repression, ${v^*} = 1$ .

Lemma 4 shows that a government concerned only with domestic conflict intuitively balances the marginal gains, in terms of revenue denied to the rebels, with the marginal cost of employing each tactic. Fighting denies rents to the rebels directly while repression hurts them indirectly by eliminating their civilian support base. The rebels rely on civilian labor to generate revenue they can expropriate, and repression destroys part of this labor force.Footnote ⁴⁷

The final feature of this benchmark is the relationship between fighting and repression. These tactics are substitutes, meaning that governments trade off fighting for repression, or vice versa. In the absence of concerns about foreign aid, the government cares only to minimize rebel capacity, which can be achieved with either tactic. Because both accomplish the same goal, the government substitutes whichever tactic has better returns relative to its cost.

With this understanding of civilians’ and the government’s baseline incentives, Proposition 1 characterizes the game’s equilibrium. Equilibrium violence depends on the donor’s relative prioritization of migration versus repression, where $\overline {{f^\mu }} $ and $\overline {{f^v}} $ denote threshold values on the marginal responsiveness of aid to migration and repression and define the donor’s priorities.

1. (a) fighting, ${x^{\rm{*}}} \in \left( {0,1} \right)$ , and repression, ${v^{\rm{*}}} \in \left( {0,1} \right)$ , for low to intermediate $\alpha $ ;

2. (b) no fighting, ${x^{\rm{*}}} = 0$ and a positive level of repression, ${v^{\rm{*}}} \in \left( {0,1} \right)$ , if $\alpha $ is large and the donor prioritizes migration, $\left| {{f_\mu }\left( {{\mu ^{\rm{*}}},v} \right)} \right| \gt \overline {{f^\mu }} $ ;

3. (c) zero violence, $\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right) = \left( {0,0} \right)$ , if $\alpha $ is large and the donor prioritizes repression, $\left| {{f_v}\left( {{\mu ^{\rm{*}}},v} \right)} \right| \gt \overline {{f^v}} $ .

The unique equilibrium depends on the donor’s leverage and priorities.Footnote ⁴⁸ A donor who prioritizes migration more readily withdraws aid when refugee flows increase.Footnote ⁴⁹ In this case, fighting is both costly and, because it drives civilian flight, results in a loss of aid. An aid-dependent government then has no incentive to exert effort. However, because the donor prioritizes migration, a government under pressure to limit refugee flows may increase repression, crudely arresting migration.

When a donor prioritizes repression, the immediate threat of punishment prevents the government from repressing. The donor still penalizes migration as well (though not as intensely) and when the government is highly dependent on aid, this can lead it to also forego fighting, minimizing displacement. Finally, when the donor has low to moderate leverage, regardless of whether it prioritizes repression or migration, the equilibrium features both fighting and repression, which drives some civilians to flee as refugees. I focus on this equilibrium because it characterizes most conflict scenarios, where governments are not so concerned about foreign aid to avoid fighting but express concern when aid is suspended.

From a donor’s perspective, conflict in $H$ produces two types of negative consequences. First, fighting increases the flow of refugees, increasing the burden on host states. Second, repression directly harms civilians, and fighting imposes costs on those remaining in $H$ . Under the threat of conditional aid, the government is responsive to incentives created by donor policy. Some of these incentives, like the donor’s distaste for high levels of migration, compete with the government’s baseline interest in fighting the rebels while complementing its baseline incentives to repress. Then, because fighting and repression have countervailing effects on migration, the government employs both. Whether it substitutes fighting and repression or employs each to mitigate the externalities of the other, depends on the donor’s priorities.

Consequences of Donor Policies

Turning to the effects of donor interventions with strategic government violence, it is not immediately apparent that targeted humanitarian assistance or asylum policies in host countries would impact the government’s choices. After all, these kinds of policies are directed only at civilians. Yet because civilians’ migration choice affects rebel capacity, and therefore the government’s incentives to fight, policies designed to affect migration also impact violence.

I capture states’ imposition of barriers to seeking asylumFootnote ⁵⁰ by examining the effect of increasing $\gamma $ on state violence.Footnote ⁵¹

When barriers to asylum deter migration, the rebels maintain more of their support base. This incentivizes the government to fight harder and, because civilians face high costs of flight, it may do so without producing large refugee flows or triggering harsh punishment from the donor. However, when civilians are already inclined to flee, increasing $\gamma $ can lead the government to decreases fighting effort. In this case, barriers to asylum are insufficient to deter large refugee flows, pushing the government to reduce its effort to avoid driving more displacement. This outcome is less likely if benefits abroad, $b$ , decrease in the volume of arrivals, just like wages in $H$ . As shown in Appendix E, when lower benefits abroad make migration less appealing, the migration rate may be low enough that imposing barriers to asylum leads to an increase in conflict intensity.

To establish existence of the effects of $\gamma $ on violence and provide intuition for the impacts of changing asylum policies, I utilize two parameterized versions of the model that represent hypothetical countries which differ only in the rebel group’s economic capacity. In one country, the rebels’ economic output is more labor intensive and in the other, the rebels’ economic output is less labor intensive, that is, they control lootable resources like oil or minerals.Footnote ⁵² Figure 2 considers the effects of changing $\gamma $ across these countries.Footnote ⁵³ The left panel, Figure 2(a), shows changes in equilibrium fighting while the right panel, Figure 2(b) plots the equilibrium level of repression.

Figure 2.

Effects of asylum restrictions, $\gamma $ , on state violence

Note: I fix $\alpha=.95$ , $b=1.5$ , $c=.35$ , $\tau_r=.9$ , $f(\mu ,v) = 10 - {1 \over {{{(1 - \mu )}^7}}} - {{.5} \over {1 - v}}$ , $Y = 1.65{((1 - v)(1 - \mu ))^{{1 \over 3}}} $ when rebels have resource endowments, and $Y = 1.65{((1 - v)(1 - \mu ))^{{2 \over 3}}}$ without resource endowments.

Figure 2 Long description

The image contains two line graphs showing the effect of increasing asylum restrictions, gamma, on equilibrium fighting effort, x star, and repression, r star, for two different kinds of rebel groups. One rebel group has access to natural resources and the other has no resource endowment. The first graph on the left shows the effect of asylum restrictions on equilibrium fighting effort. The x-axis represents asylum restrictions, gamma. The y-axis represents fighting effort, x star. Two lines are plotted: a solid purple line representing fighting effort for a rebel group without resource endowments and a dashed green line representing fighting effort for a group with resource endowments. For both types of rebel group, increasing asylum restrictions increases equilibrium fighting. The second graph on the right shows the effect of asylum restrictions on repression, where the y-axis is now r star. A solid purple line plots equilibrium repression when rebels lack resource endowments, and a green dashed line shows repression for a group with resource endowments. The curve for rebels that lack resources is always below the curve for rebels with resource endowments.

Figure 2 shows the effects of increasing asylum restrictions depend on conditions in $H$ . When the rebels lack lootable resources, they are reliant on their civilian support base, whose wages are generally higher than when the rebels have resource endowments. Then, because civilians are sensitive to new barriers to asylum, the government can fight harder without triggering a significant increase in migration. When the rebels have meaningful resource endowments, they are less dependent on civilian support and civilians’ wages are lower regardless of government violence. They are thus more likely to flee even in the face of increasing barriers to asylum. Then, to curb the flow of refugees and sustain aid from the donor, the government chooses a substantially higher level of repression and a lower level of fighting effort. The effects of increasing $\gamma $ are in the same direction but of different magnitudes due to the differences in civilians’ underlying propensity to flee abroad.

Humanitarian relief can directly improve circumstances for civilians and has the perhaps desirable second-order effect of reducing refugee outflows, relieving pressure on refugee-hosting states. An increase in relief aid, however, may also embolden the government.Footnote ⁵⁴ Knowing relief efforts will mitigate the cost of fighting for civilians, Proposition 3 shows that humanitarian relief can sometimes increase conflict intensity.

Figure 3 illustrates this. Figure 3(a) shows optimal fighting increases as humanitarian relief reduces costs for civilians of remaining in $H$ . Consistent with findings that aid can prolong or intensify conflicts,Footnote ⁵⁵ the government fights with less restraint because aid can insulate it from some repercussions of conflict.

Figure 3.

Effects of humanitarian relief, $c$ , on state violence

Note: Parameters are fixed at the same values from Figure 2, with $\gamma=3.25$ .

Figure 3 Long description

The image contains two line graphs showing the effect of increasing humanitarian relief, or reducing c, on equilibrium fighting effort, x star, and repression, r star, for two different kinds of rebel groups. One rebel group (plotted with a solid green dashed line) has access to natural resources and the other has no resource endowment (plotted with a solid purple line). The first graph on the left shows the effect of increasing humanitarian relief on equilibrium fighting effort. The x-axis represents a percentage reduction in c and the y-axis represents fighting effort, x star. Both lines show an upward trend, indicating that fighting effort increases with more humanitarian relief, with the resource endowment scenario showing a steeper increase. The second graph on the right shows the effect of increasing humanitarian relief on repression, where the y-axis is now r star. Both curves are non-monotone. They are increasing for low levels of humanitarian relief and decreasing for high levels of humanitarian relief. The solid purple curve for rebels that lack resources is always below the green dashed curve for rebels with resource endowments.

The effect of humanitarian relief on repression is more ambiguous, which is also consistent with extant literature.Footnote ⁵⁶ Figure 3(b) shows the effect of aid on repression depends on both underlying conditions, in this case the rebel’s resource endowment, and can be non-monotone. When substantial humanitarian relief reduces civilians’ costs from conflict, the government substitutes fighting for repression, limiting violence against civilians to avoid sanction by the donor. On the other hand, the government increases repression at moderate levels of humanitarian relief. In this case, the government initially substitutes repression for fighting effort. When the rebels are resource endowed, civilians’ baseline propensity to flee is higher and without substantial humanitarian relief, intense fighting would lead to large refugee outflows. By increasing repression, and choosing a higher level of repression than when rebels lack resources, the government undermines the rebels’ capacity without increasing migration. As greater levels of humanitarian relief counteract civilians’ incentives to flee, the government shifts to fighting more and repressing less, effectively countering the rebels with less severe penalty from the donor.

Conditioning Aid

Donors have another tool to directly affect the government, rather than civilians, as they try to stop both migration and violence: conditioning assistance. Importantly, not all governments rely on foreign aid, which means some should be more responsive to conditions on aid than others. To capture how donor leverage affects violence, I evaluate the effect of the parameter, $\alpha $ . I contrast donors who are more sensitive to the level of repression and donors who prioritize reducing refugee flows.

1. (a) if the donor prioritizes migration, fighting, ${x^*}$ , decreases and repression, ${v^*}$ , increases if $x$ and $v$ are substitutes. If $x$ and $v$ are complements ${x^{\rm{*}}}$ decreases if and only if ${{{g_{vv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \over {\left| {{g_{xv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \right|}}\left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _x^{\rm{*}}} \right| \gt {f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _v^{\rm{*}} + {f_v}\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)$ while ${v^{\rm{*}}}$ increases if and only if ${{{g_{xx}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \over {\left| {{g_{xv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \right|}}\left( {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _v^{\rm{*}} + {f_v}\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)} \right) \gt \left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _x^{\rm{*}}} \right|$ .

2. (b) if the donor prioritizes repression, both ${x^{\rm{*}}}$ and ${v^{\rm{*}}}$ decrease when $x$ and $v$ are complements. When $x$ and $v$ are substitutes, ${x^{\rm{*}}}$ decreases if and only if ${{{g_{vv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \over {\left| {{g_{xv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \right|}}\left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _x^{\rm{*}}} \right| \gt \left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _v^{\rm{*}} + {f_v}\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)} \right|$ while ${v^{\rm{*}}}$ decreases if and only if ${{{g_{xx}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)} \over {{g_{xv}}\left( {{x^{\rm{*}}},{v^{\rm{*}}}} \right)}}\left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _v^{\rm{*}} + {f_v}\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)} \right| \gt \left|\ {{f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)\mu _x^{\rm{*}}} \right|$ .

Proposition 4 highlights how governments’ responses to aid conditionality depend on donor priorities and features of the conflict environment, which affect the government’s optimal strategy for countering the rebels. Figure 4 illustrates all four cases. Given a donor who prioritizes migration, as the government becomes more responsive to aid withdrawal, it will generally increase repression. This is shown in Figure 4(b) when rebels have resource endowments. Repression is a crude means of preventing migration while targeting the rebels’ base of support. As $\alpha $ increases, the consequences of driving migration by choosing a high level of fighting effort are more severe. When repression and fighting are substitutes, the government employs a higher level of repression to avoid increasing migration and incurring donor sanction; this holds in Figure 4 for low $\alpha $ . If instead repression and fighting are complements, the government, loath to cede control of territory to the rebels, does not want to totally forego $x$ . When the government expects substantial aid losses due to migration, that is, when the magnitude of ${f_\mu }\left( {{\mu ^{\rm{*}}},{v^{\rm{*}}}} \right)$ is large, then in order to keep fighting the government increases repression. This holds in Figure 4(b) for intermediate $\alpha $ . Only when donor leverage is great enough that the consequences of repression also bite does the government reduce ${v^{\rm{*}}}$ , illustrated for the largest values of $\alpha $ .

Figure 4.

Effects of donor leverage, $\alpha $ , on state violence

Notes: Parameters are fixed at the same values from Figures 2 and 3.

Figure 4 Long description

Two line graphs illustrate the effects of aid dependence on fighting effort and repression. The first graph on the left shows fighting effort, x star, on the y-axis and aid dependence, alpha on the x-axis. Two lines are present: a solid purple line representing a rebel group without resource endowments and a dashed green line representing a rebel group with resource endowments. Both lines show a downward trend, indicating that as aid dependence increases, fighting effort decreases. The line for rebels that lack resources is always below the line for rebels with resource endowments. The second graph on the right shows repression, r star, on the y-axis and aid dependence, alpha, on the x-axis. Again, two lines are present: a solid purple line representing rebels without resource endowments and a dashed green line representing rebels with resource endowments. The dashed green line shows an upward trend, indicating that as aid dependence increases sharply for low alpha, then increases slowly for moderate alpha, and decreases for higher alpha. The solid purple line shows a downward trend, indicating that as aid dependence increases, repression decreases. All values are derived from numerical solutions to parameterized versions of the model.

When a donor prioritizes repression, if $x$ and $v$ are substitutes, ex ante, the government prefers to employ whichever tactic is more effective against the rebels rather than accommodate the donor. As $\alpha $ increases, because the donor still prefers lower levels of migration, donor pressure may lead the government to reduce fighting effort, yielding lower refugee flows. But this can come at a cost. To maximize assistance while still undermining the rebels, the government may instead increase repression. On the other hand, Figure 4 illustrates how the government may reduce both fighting and repression. This is shown in Figure 4 when $\alpha $ is low and rebels lack resources, and occurs when the migration rate is low and violence has limited effect on civilians’ decision to flee. Then, expecting substantial loss of aid and with little ability to undermine rebel support, the government reduces violence.

Proposition 4 also highlights an optimistic case in which donors can achieve their aims. Governments that employ high levels of fighting and repression together are particularly troubling for donors. But when these governments are aid-dependent, if the donor commits to withdrawing aid in response to repression, it can induce reductions in both fighting and violence against civilians. Figure 4 illustrates this when rebels lack resources and $\alpha $ is high. This requires, however, that the donor strictly prioritize reducing repression over minimizing refugee flows and must condition its aid accordingly. This highlights the primary tension a strategic government exploits: the donor may care more about its own domestic concerns and prioritize minimizing refugee arrivals. A donor who prioritizes migration creates a tacit permission structure for the government to maximize its damage to the rebels by choosing high levels of both fighting and repression. Only when the donor is prepared to make significant investments in protecting civilians in $H$ can it reduce effort spent providing for those civilians abroad, after they have fled conflict.