Refugees, Economic Capacity, and Host State Repression*

Abstract Does hosting refugees affect state repression? While there have been numerous studies that examine the link between refugees and the spread of civil and international conflict, an examination of the systematic links between refugees and repression is lacking. We contend that researchers are missing a crucial link, as the dissent-repression nexus is crucial to understanding the development of armed conflict. Drawing upon logics of the relationship between refugees and the spread of conflict as well as economic capacity, we argue that increased numbers of refugees lead to increased repression. We contend that willingness to increase repression when hosting refugees is in part conditional on a host state’s economic capacity. We argue that, on the whole, the greater the population of refugees in a host state, repression becomes more likely. That said, we argue that increased economic capacity will moderate this relationship. We find empirical support for both predictions.

The recent mass exodus of refugees fleeing Syria to seek asylum in Europe and North America has led to an impassioned public and policy debate over the management of this refugee crisis. Humanitarian arguments have clashed with concerns for the extent to which a host state's social fabric, economy, and security is affected by incoming refugees and how governments respond to them (Gambino, Kingsley, and Nardelli 2015;The Guardian 2016). Recent research focusing on the potential dangers of refugee camps for militant recruitment (Lischer 2005) and as transnational rebel bases (Salehyan 2009) have highlighted that influx of refugees into a country has consequences. Refugee militarization risks spreading civil wars across borders, or potentially laying the groundwork for interstate conflict between sending and host states (Salehyan 2008). That said, despite this potential, the hosting of refugees is much more commonplace than internationalized civil wars or interstate conflicts concerning transnational rebellion. This article seeks to explain and understand the recurring debate among states that host refugees by asking: do refugees affect a host state's propensity to repress? Do hosting refugees change the host state and its domestic political dynamics toward strengthening repressive behavior? We focus on the repression CONTACT  of physical integrity rights as a more pervasive and commonplace form of political violence. These rights of the person, which are violated via the use of imprisonment, torture, killings, and disappearances as a way to punish political opponents, are violated with great frequency throughout the world (Cingranelli and Richards 2010;Wood and Gibney 2010). Furthermore, political repression and the rise of civil conflict are related (Young 2013). Thus, understanding how the presence of refugees interacts with a host state's ongoing dynamics of dissent and repression is thus a crucial step in understanding how refugee movements may affect political violence more broadly.
In this project, we argue that refugee populations will lead to increased repression by a host state. Drawing upon logics of refugee mobilization (Lischer 2005;Salehyan 2009), linkages between refugees and violent dissent (Choi and Salehyan 2013) and the linkage between state capacity and repression (Cingranelli, Fajardo-Heyward and Filippov 2014;Englehart 2009), we argue that with increased refugees within a host state, grievances from both within those groups and the local population are likely to increase against the government. Exacerbated grievances against the government as well as real or perceived economic competition from refugees make repression an increasingly likely policy option to be pursued by states in order to maintain regime security. Refugee settlements themselves are likely to be places of little economic opportunity, which can generate grievances among the refugees themselves, who may turn to rebel groups and mobilization (Gleditsch and Salehyan 2006;Lischer 2005). Repression may be considered a viable policy option to stem the tide of refugee mobilization against the host state, or to prevent the risk of internationalized civil conflict. Consider the way in which the Paris attacks of November 2015 have been linked to the Syrian refugee crisis in Europe, and a prevailing worry about refugees causing a contagion of conflict (Smale and Lyman 2015). Refugee populations also present a form of economic stress for the host state, which can lead the local population to "otherize" refugees. Such processes can lead to increased grievances and dissent among the local population. There have been massive rallies organized in various European countries to protest against new refugee arrivals, particularly of Muslim faith (BBC Newsbeat 2015). Observing increased dissent by either refugees or the local population, the host state may choose to employ repression as a way of removing potential political threat. This may lead the host state to increase repression against the local population or refugees or both.
Despite the overall impact of refugees and repression, we argue that increased economic capacity may moderate the effect of refugees on repression. Higher capacity states will be better able to absorb the potential stresses brought by increased amounts of refugees, which may make grievances less likely to emerge and free up policy options other than repression when and if dissent does emerge.
We test our expectations on a cross-national sample and find some support for the general expectation that increased numbers of refugees from neighboring states lead to increases in repression. Furthermore, we find stronger support for the argument that increased economic capacity moderates the impact of refugees on repression. In all, we believe we present contributions to both research on how refugees contribute to political violence, as well as research that explores how economic capacity affects repression dynamics.

Repression dynamics
Repression of physical integrity-the use of imprisonment, torture, killings, and disappearances-occurs frequently throughout the world. While there is good reason to think that repression is becoming less common over time (Fariss 2014), it is still a far more common and pervasive form of political violence than outright civil or international conflict. States decide to repress political opponents when they feel they are being threatened at a high level relative to their political strength (Poe 2004). Over the past 20 years or so of research into the factors associated with repression, researchers have uncovered several key relationships. Factors such as increased democracy (Bueno de Mesquita, Cherif, Downs, and Smith 2005;Davenport and Armstrong 2004) and increased economic development are associated with less repression, while factors such as civil war and larger populations are associated with more repressive states (Davenport 1995;Davenport 2007a;Poe and Tate 1994). While some of these factors are stronger predictors than others (Hill and Jones 2014), what has emerged from prior quantitative research on repression is a set of structural factors that establish a baseline of repression. These factors, such as regime type, population, and level of economic development, are generally consistent, year-to-year. We can expect that democracies will be less repressive because their leaders must appeal to a broad constituency through the distribution of public goods and thus will have fewer grievances and dissent, and thus "need" to use repression less than their autocratic counterparts (Bueno de Mesquita, Morrow, Siverson, and Smith 2003). Richer countries will also generally have fewer grievances and need repression less, while more populous countries will likely have larger segments of the population to "win-over" and more opportunities for both dissent and repressive responses (Poe and Tate 1994).
Other factors may "shock" the political status quo. Factors such as civil conflict (Young 2013) and some forms of interstate conflict (Wright 2014) can rapidly alter a government's calculations of strength and threat. Civil conflict, which presents a direct challenge to the regime by rebel groups, ushers in what Davenport (2007a: 7) refers to the "law of coercive responsiveness", in which states will increase repression in response to direct, violent challenges. Other factors that may present a shock to baseline expectations for dissent and repression are economic sanctions (Wood 2008) or natural disasters (Wood and Wright 2016). Thus, repression dynamics can be thought of as a sort of punctuated equilibrium, where structural factors, such as regime type, population, and economic development, generate the baseline expectations for dissent and repression within a state, and shock factors, such as conflict and catastrophe, may rapidly alter these expectations and lead to short to medium term shifts in repression. We treat refugee populations as a shock to the ongoing political status quo of host states. While refugees often do not present a rapid onset of crises for host states, and many refugee populations stay in receiving countries for years, they present a new population that may require those states to adjust in a number of ways, including economically and socially. In order to understand how refugees affect host state repression, it is best to try and understand how the potential for dissent among both refugee groups and the local population affects the motives of the host state to maintain political stability and regime security.

Refugee mobilization, local dissent, and host state repression
The hosting of a population from outside the state may alter host regimes' assessment of political strength and threat, as well as place a strain on other factors that affect repression, such as the economy. Furthermore, refugees may mobilize against either the host state or the sending state, which may risk the spread of civil conflict or threaten the host state with interstate conflict. Given these domestic political and economic pressures, as well as the risk of international conflict, host states become more willing to clamp down on emerging dissent among refugees and the local population.
Recent path-breaking research on transnational conflict has found that civil conflicts often become internationalized and many interstate conflicts often have some domestic conflict at their root (Salehyan 2009). One of the mechanisms through which these conflicts become transnational is via refugee populations. The presence of refugees may lead to increased opportunities for mobilization within refugee camps, which while being focused against the sending state, also risks the spread of civil conflict into the host state. Lischer (2005) argues that refugees organized as states in exile are more likely to experience mobilization within the camps. These are groups that flee losses in civil wars in hopes of rearming in a new safe haven. Such groups, which are not typical for refugees, are essentially armies situated in camps. Both Salehyan (2009) andLischer (2005) describe the case of refugees fleeing from the aftermath of the genocide in Rwanda in 1994, which contributed to both ongoing conflict within Zaire (now the Democratic Republic of the Congo [DRC]) and eventually interstate conflict between the DRC and neighboring states, including Rwanda. Salehyan (2009: chapter 5) notes that fleeing genocidaires from Rwanda aimed to mobilize in the camps in order to attack the new Rwandan government. Support for rebels within the camps by the Zairian government led to Rwanda actively supporting Zaire's own rebels in the toppling of the Mobutu regime. Salehyan emphasizes that these refugee camps can serve as transnational bases from which groups can attack the sending state government.
Transnational rebellion can also bring civil conflicts across borders and risk interstate conflict between the refugee-sending and refugee-host states. Eventually, in the Congo, the new DRC government began to support rebelsin-exile against the Rwandan government and the Rwandan government began fighting against their former allies-this would become what has been referred to popularly as the African World War. Thus, refugee populations can contribute directly into the spread of civil and international conflict to the host state (Salehyan 2009).
Considering that rebel mobilization among refugee populations is of serious concern for host states, how do refugee populations affect ongoing dynamics of dissent and repression within host states? We believe that facing the possibility of mobilization within the camps, host states will become more likely to increase repression. Danneman and Ritter (2014) explain that states become more likely to engage in pre-emptive repression when there is a risk of civil conflict spreading within a region. They employ refugees from a neighboring state as a measure of proximity to civil conflict and find that as states see civil war becoming more prevalent within their neighborhood, they are more likely to increase repression in order to prevent the rise of rebellion. Choi and Salehyan (2013) find that in addition to providing opportunities for militarization, states hosting refugees may also receive large amounts of humanitarian aid, which provides looting opportunities and may contribute to an increase in the use of terror tactics by armed groups. Facing the threat of potential refugee mobilization as well as other motives for violence, host states will be more likely to increase repression when hosting more refugees.
While a fear of transnational violence may lead to some repression, a more common mechanism may be that host states may view refugees as sources of increased economic competition and strain. Nordas and Davenport (2013) explain that the presence of spikes in youth population may increase violent dissent and repression in response. A large amount of young men in a state with few economic opportunities become more likely to participate in organized violence, and repression becomes a likely response by the state. While refugee camps may or may not have large youth populations, the impact on the local economy and potential dissent may be similar regardless of the specific demographic characteristics of the refugee populations themselves. The presence of large amounts of refugees may lead to a net drain on the host state's economic output, by increasing the population of those with few economic opportunities and by providing economic competition for the local population. This could lead to dissent both from within and without the refugee population. Lischer (2005: 105) notes that during the 1994 elections in Tanzania, refugees were scapegoated by local politicians and blamed for increased crime and economic strain within the state. Thus, local populations, experiencing increased economic competition from refugees and facing economic strain, may also become more likely to engage in dissent against the host state government. Based on the above discussion, we argue that there are two mechanisms through which an increased presence of refugees may lead to increased repression by host states. The first is a transnational security mechanism. States fearing the spread of civil conflict into their own borders, or seeking to avoid interstate conflict over refugee mobilization, will seek to prevent it via the use of increased repression. Aside from this mechanism, states hosting refugees are seeking to maintain domestic stability. Refugees may present economic competition to the local population and lead to increased economic strain for the host state. This can lead to anti-refugee sentiment, increased grievances, and increased dissent by the local population against the state. The December 2015 rally in Dresden, Germany, is emblematic of this dynamic not just because it vocalized local grievances against refugees, but also because it was as aggrieved with the German government's refugee policy (Reuters 2016). States may thus increase repression against refugees, the local population, or both in order to stem the tide of increased dissent and promote order. Both of these mechanisms lead to the same core empirical expectation.
Hypothesis 1: As the population of refugees increases, host states become more likely to increase repression.
A state's choice to use repression tactics to quell dissent that may accompany refugee inflows is conditional on its capacity. A high capacity state is better able to (a) prevent refugee mobilization, and (b) respond to local grievances against refugees without resorting to increased repression. Lischer (2005) argues that the host state's capacity is influential for whether refugees become militarized and civil war spreads. She contrasts the experience of Rwandan refugees in two countries in the aftermath of the genocide and the subsequent Rwandan Patriotic Front victory. In Zaire, refugee camps were dominated by the statein-exile type of organization and found a willing, or at least tolerant Zairian government, but also a host state government with little capacity to stop the mobilization in the camps. In Tanzania, however, while they found a local population somewhat sympathetic to their politics, they also found a host state with no interest in having the conflict spread. Tanzania is notable for being a state that hosted hundreds of thousands of refugees from Rwanda from 1994 to 1996 but did not become directly involved in the regional war in the Congo (p. 30). Lischer points to both Tanzania's foreign policy desire to have no direct involvement in a conflict and its increased state and economic capacity over the Congo as being crucial for its lack of militarization in the refugee camps. A high capacity state, in this scenario, has little incentive to use repressive tactics to counter refugee mobilization.
How does capacity affect a host state's ability to respond to local grievances that may accompany refugee inflows? The presence of refugees in a host state may alter the dynamics between the citizen and the state because of largely economic concerns. Simply put, we believe that local populations may view refugees as potential economic competition, which may increase feelings of xenophobia against the refugee population, as noted by Lischer (2005) in her description of the 1994 Tanzanian elections described above. Increased negative perceptions against refugees may lead to both increased grievances against the state by local populations and increased repression and harassment by local police against refugees. The state may also repress the local population to manage tensions between them and the refugees. This dynamic may be particularly acute in countries with high unemployment or an already scarce economy. When the local population is already feeling particularly economically vulnerable, the influx of refugees, who may provide a comparatively cheaper labor source for employers, may lead to increased feelings of prejudice and fuel dissent or group violence, which may in turn lead to increased repression.
This dynamic was in play in 1990s South Africa. While there are many wellknown factors to point to for South Africa's human rights performance in the 1980s and 1990s, such as the end of apartheid and the transition to democracy, the country also hosted hundreds of thousands of refugees fleeing neighboring Mozambique and their long civil war that ended in 1992. After the war ended, over 200,000 refugees opted to stay in South Africa (Polzer 2007). During this early post-apartheid period, many parts of South Africa's economy were struggling. Mozambican refugees, and immigrants more broadly, were frequently made the scapegoat for the economic woes of the local population. Human Rights Watch (1998) reports that refugees and immigrants were viciously targeted by local police in some areas. In all, we believe that when states experience increased inflows of refugees while facing difficult economic conditions, repression becomes more likely.
Increased capacity allows for states to be better able to reach their citizens and better able to provide services. Furthermore, they are better able to respond to economic grievances of the population. A state with greater economic capacity will make violence a less attractive form of dissent, but also states with greater capacity can also better track and monitor dissidents. This provides the states with more policy options than repression when facing potentially violent dissent. State capacity as a concept can connote many meanings (Hendrix 2010). Prior research has emphasized military capacity, bureaucratic capacity, as well as economic capacity. In our conception, we focus on economic capacity because anti-refugee sentiment may often take the form of emphasis on economic competition. States with low economic capacity may find themselves particularly vulnerable to anti-refugee sentiment playing a salient role in how they conduct domestic politics, as indicated above in our example of South Africa. We also wish to distinguish the concept from military capacity, which might be thought of as the capability to repress, or power. There are many very capable states that may engage in high levels of repression, but do not. High capacity for repression and propensity for repression are two different ideas. Going back to the logic of strength and threat (Poe 2004), states engage in high levels of repression when they feel threatened relative to their level of political strength. In other words, as Arendt (1970) notes, power and violence (in our case, repression) are inversely related. States with power (capacity, political strength) are less likely to feel the impulse to increase repression because they feel more secure, either politically secure, or because they have such capacity that policy options other than repression are open to them.
Prior research on capacity and repression bears out this expected inverse relationship. Building from a bureaucratic capacity perspective, Cingranelli, Fajardo-Heyward, and Filippov (2014) argue that greater bureaucratic capacity makes states generally less repressive. Arguing from a principal-agent framework, Cingranelli et al. focus on administrative capacity's ability to increase bureaucratic accountability as a way of securing human rights. Cingranelli et al. employ reliance on taxes as a primary indicator of capacity. They argue that such reliance increases the accountability of the state to its citizens from whom it extracts taxes. Greater accountability leads to better governance across the board, which translates to better human rights performances. Furthermore, increased capacity may make certain forms of dissent less likely. Hendrix and Young (2014), trying to understand the relationship between state capacity and terrorism, separate state capacity into two components: military and bureaucratic. They find that increased bureaucratic capacity makes dissenters less likely to employ terrorist tactics while high military capacity makes terrorism more likely. Hendrix and Young (2014: 334) argue that increased bureaucratic capacity-which they define as the state's ability to collect and manage information about their population-allows for a state to provide different channels for dissent beyond violence.
Prior research debates the precise role that economics plays on the likelihood of repression (Davenport 2007a;Hafner-Burton 2014;Harrelson-Stephens and Callaway 2003). The record of findings of whether economic liberalization policies like increased trade (Harrelson-Stephens and Callaway 2003) or structural adjustment (Abouharb and Cingranelli 2006) is mixed. For our purposes, how states gain in their economic capacity is less important than the current state of economic capacity at the time they interact with a refugee population. Therefore, we are less interested in how the state interacts with the economy, rather than the overall level of economic capacity of a host country and how that impacts how flexible the economy is, in general, with respect to absorbing refugees into it. In general, we contend that higher levels of economic capacity should mean less repressive host states. For this reason, we focus on overall economic capacity rather than particular forms of economic organization.
We argue that states with greater economic capacity will be better able to absorb the "shock" of the refugee population. States with greater resources will likely have greater monitoring ability and may be able to prevent or better channel dissent from emerging in reaction to conditions of the camps, or emerging from the local population against refugees. Furthermore, states with greater capacity may simply be better able to manage the refugee population. States with greater reach can better serve citizens in border areas where refugees are likely to be focused and potentially better integrate refugees into the local economy without as much backlash from the local population. To the extent that there is greater economic competition in refugee areas, it may be mitigated by better governance and management of humanitarian aid in states with greater bureaucratic capacity. Thus, dissent, and repression in response to it, may be less likely in states with greater capacity, despite facing refugee crises. We believe that increased repression is less likely in more economically capable states, despite the potential for high refugee sentiment, because these wealthier economies should still be better able to withstand higher levels of dissent and have more policy options available to them. 1 Hypothesis 2: Increased economic capacity moderates the relationship between refugees and host state repression.

Research design
Our hypotheses predict that a state increases repression as it hosts increasing amounts of refugees. Additionally, we hypothesize that a host state with weak economic capacity increases repression in response to refugees. Considering our concern with levels of repression in a state, our unit of analysis is country-year. We conduct our analysis on primarily a sample of cross-national, annual data from 1952 to 2011. Our primary dependent variable, host state repression, is the Schnackenberg and Fariss (2014) variable for latent respect for human rights practices. This variable pulls together information from various sources of human rights performance, including the Political Terror Scale (Wood and Gibney 2010) and Cingranelli-Richards scales (Cingranelli and Richards 2010) and accounts for the increasing standard of accountability in human rights (Fariss 2014). 2 Its aim is to estimate the latent amount of human rights respect in a country. It is a normalized scale around a global mean of zero, with positive numbers indicating greater than average respect for human rights and negative numbers indicating lower than average respect for human rights. We have reversed the scores (multiplied them by −1) in order to have higher scores represent less respect for human rights, to make the models employing the latent variable comparable to our theoretical modeling approach that explains repressive behavior, rather than improved human rights 1 Thus, despite increased anti-refugee sentiment in countries like the United States, for instance, our theoretical expectation is that repression will not increase dramatically as a result. 2 The Schnackenberg and Fariss data also allow for tests using a much larger sample, as their data begin scoring countries in 1949. Taking into account missingness on our covariates (specifically, relative political reach), we were able to employ these data on samples ranging from 1952 to 2011.
practices. In our sample, the mean is −0.101 and ranges from −4.705 (least repressive) to 2.871. We also conducted robustness checks employing the Political Terror Scale (PTS) (Wood and Gibney 2010), which is based on yearly (beginning in 1976) human rights reports published by Amnesty International and the US State Department. 3 These results are largely similar on key models to those displayed below.
Our main independent variable is the natural log of the number of refugees a state is hosting in a given year. These data are primarily sourced from the United Nations High Commissioner for Refugees (UNHCR). However, since UNHCR does not count Palestinian refugees in Syria, Jordan, and Lebanon, we get those numbers from the United Nations Relief and Works Agency (UNRWA) and the US Committee for Refugees and Immigrants (USCRI). UNHCR, UNRWA, and USCRI provide refugee data in dyadic form for the years 1951 to 2013, that is number of refugees from an origin country in an "asylum" country. For generating a measure of total number of refugees in a host state in a given year, we employed Correlates of War country codes for host and origin states, collapsed the data to the host state, summing the total number of refugees from all origin states in a given year. As with prior research, we employ the natural log because the distribution of the number of refugees in countries is skewed. The majority of states do not receive any refugees, and even among those who do, some states receive far more than others. We primarily focus on total refugees because it provides the closest thing to a total picture of the refugee situation in a given country. Larger refugee populations are more likely to place strain or the perception of economic strain in the host state's citizenry and be more likely to lead to increased contention and repression.
The two mechanisms identified above, a transnational security mechanism and an economic competition mechanism, lead to the expectation that increased numbers of refugees will lead to increased amounts of repression. While hypothesis 2 allows for some testing of the second mechanism, we do not predict a clear way of distinguishing between the two mechanisms in the general relationship between refugees and repression. In order to make some inroads into this challenge, we employ some additional indicators of the number of refugees, based on from where refugees come. We measure the number of refugees from states engaged in a civil war in a given year. We rely on the Uppsala Conflict Data Program's Armed Conflict Data set (Themner and Wallensteen 2014) to identify states engaged in civil wars, which are coded as states that reach 1,000 battle deaths over the course of their internal armed conflicts. For generating both these additional variables, our procedure was similar to how we generated the total 3 PTS measures physical integrity violations within a country in a given year on a 5-point ordinal scale, in which 5 is the most repressive and a score of 1 is the least repressive. Countries perpetrating widespread, systematic abuses against their citizens get a higher score on the PTS scale than those that do so rarely. In keeping with common practice, we employ State Department scores with Amnesty scores imputed for country-years in which State Department scores are missing. number of refugees in a host state variable. Using the UNHCR/UNRWA/USCRI dyadic data that list refugee-hosting and refugee "producing" states, we summed refugees from civil war countries in a host state in a given year.
Our second hypothesis predicts that states with greater economic capacity will be less likely to increase repression as the number of refugees increase. Our theoretical approach argues that states with higher economic capacity are better able to absorb the "shock" of refugees. More capable states should thus be able to withstand increased grievances and dissent from both within the refugee and local populations and be better able to address such grievances with a lower likelihood of resorting to repression. These states should have more economic opportunity and resources. For this concept, we employ two different indicators. We employ the natural log of Gross Domestic Product (GDP) per capita as our primary indicator of economic capacity of the host state. We chose this variable because it is the most straightforward and widely available of economic capacity measures, which perhaps gives the measure broader applicability, and has been used in notable studies testing the linkage between capacity and violence (for example, Fearon and Laitin 2003). Prior work on state capacity notes that although this measure may have advantages over other measures of capacity, it also may represent multiple causal pathways between capacity and violence (Hendrix 2010: 277-278). That said, since our theoretical logic builds on a more holistic notion of capacity (that is, how well the state can absorb shocks and whether it has more options than violence), a broad indicator of capacity like GDP per capita is useful. Furthermore, it is one of the major variables that has empirical ties to repressive behavior (Hill and Jones 2014), with high levels of GDP strongly associated with less repressive human rights practices. Our measure of GDP per capita is taken from the Gleditsch (2002) measure of real GDP per capita at 2005 levels. We take the natural log of this measure because the distribution of GDP across countries is skewed.
We also employ Mousseau's contract intensive economy (CIE) measure (Mousseau and Mousseau 2008;Mousseau 2015Mousseau , 2016. This variable captures the amount of formal participation in the national economy. Specifically, it is meant to measure the concept of an economy that is rich in contractualist norms. Mousseau and Mousseau (2008) argue that these contractualist norms of interpersonal reciprocity and the upholding of contracts promote the rule of law and demote the use of violent repression as a means of combatting dissent or political threats by the state. Mousseau and Mousseau (2008) find contractintensive economies to be negatively related to repression. 4 We employ this measure, for which the proxy is the logged average of life insurance premiums in 4 This concept and measure is also employed in the ongoing debate between advocates of a "contractualist peace" argument over the "democratic peace" argument (see Mousseau 2016). We have no stake in that debate and it is tangential to this research question, and we contend that it is one of many potential measures that capture the overall economic capacity of a state and as a check on the robustness of our findings employing GDP per capita. a country, as another indicator of economic capacity. 5 We contend that states that have greater intensity of the upholding of contracts are more likely to have diverse economies and be better able to absorb the shock of a refugee population.
Existing literature on repression suggest a number of factors that affect statesponsored repression that need to be controlled for in the analysis. Presence of civil conflict in a country is expected to increase state-sponsored repression. Data on civil conflicts are taken from the UCDP/PRIO Armed Conflict Data set (Themner and Wallenesteen 2014). This dummy variable codes whether an ongoing intrastate or internationalized intrastate conflict has reached 25 battle deaths in a given year. Previous studies have shown that population size is positively related to repression (Poe and Tate 1994;Nordas and Davenport 2013). We include the log of population from Gleditsch (2002). 6 Additionally, democracies are less likely to engage in systematic abuses against their population than autocratic states (Davenport and Armstrong 2004). We use the 21point Polity2 measure from the Polity IV data set (Marshall, Jaggers and Gurr 2011) that ranges from −10 to +10. The control variables, aside from civil conflict, are lagged one year. Civil conflict is not lagged because it can suddenly alter a state's repressive output. 7 We employ fixed effects regression to evaluate our hypotheses, because our dependent variable is continuous. The fixed effects approach allows us to model the impact of the independent variables while controlling for country-specific variation by employing different intercepts in the statistical model for every country (or year if year fixed effects are employed) (Beck and Katz 1995: 645). This allows us to model the impact of the individual variables and interactions in states over time. We also cluster standard errors on the country to account for within unit correlation. We do not include a lagged dependent variable in these models to avoid potential Nickell bias (Gaibulloev, Sandler, and Sul 2014). We also include standard errors clustered on country and in some models; we employ year fixed effects as well. Models that have alternate modeling corrections, as well as models employing an AR1 specification, are included in our online Appendix

Results
Before assessing our main results from the fixed effects regression models, in Table 1 we include some comparable descriptive statistics of the variables used in our models. In order to try and better understand the distribution of the most 5 This variable is less widely available than GDP per capita, and so our sample with this variable is restricted to only through 2010 and only 163 countries. 6 We multiplied Gleditsch's (2002) population variable by 1,000 before logging it to be comparable with our measure of refugees. This change does not affect the results. 7 We did conduct tests of a model where both the current year and lagged versions of our independent variables were included. Only for democracy and population did the lagged version of the variable reach significance when the current version was included.
important variables (refugees, economic capacity, and repression), in Table 2 we compare the average number of logged refugees and the level of repression across three different ranges of economic capacity. The observations are restricted to those that have at least some refugees. As Table 2 shows, observations that are one standard deviation above or higher than the average for GDP and one standard deviation below or lower than the average for GDP host similar average amounts of refugees, but their average amounts of repression are very different. This indicates to us that refugees do not appear to "cluster" in a particular type of country, based on wealth, as the majority of country-years that have any refugees at all are contained within a standard deviation of the sample average. It also seems to indicate that countries with higher levels of GDP will be less repressive, despite the presence of refugees, than countries with lower levels of economic performance. This provides some suggestive evidence for our theoretical approach.
In Table 3, we report the findings from our main analyses. In Models 1 and 2, we present the test for the general relationship between total refugees hosted and repression, both with and without year fixed effects in addition to the country fixed effects. These models present tests for our first hypothesis that an increased refugee population leads to increases in repression by host states. While Model 1, without year fixed effects, does not show this result, Model 2 does, which indicates mixed support for our first prediction of the general expectation that as the total number of refugees increases, increases in repression are likely to occur as well. While both models have fixed effects for countries, Model 2 takes into account the specific year effects as well, which we believe controls for effects within panels of specific years, that may be outliers with respect to refugees.
The other models in Table 3 provide tests for our second hypothesis, that increased economic capacity among host states will moderate the relationship between refugees and host state repression. We argued that as states have greater   Table 3. Fixed effects regressions of refugees and economic capacity on repression. (1) (3) Note. Fixed effects regressions on the Schnackenberg and Fariss (2014) Latent Respect variable (reversed). Standard errors clustered on country in parentheses. Two-tailed significance: ***p < 0.01, **p < 0.05, *p < 0.1. Due to space considerations, numbers are rounded to the second decimal place, which resulted in some coefficients being multiplied by 10, as per journal style guidelines.
These include the coefficients of the theoretically relevant variables, Total Refugees on Models 1 and 2, Total Refugees X CIE in Model 6, Ref/Civ War X GDP for Model 8.
economic capacity-the ability to integrate refugees into the local economy-they would be better able to absorb the "shock" of increased refugee populations. Wealthier host states should therefore face lower risks of refugee mobilization against them, transnational rebel recruitment in the camps, and experience less negative political feedback against refugees from the local population. In Models 3 and 4 of Table 3, we test this relationship by interacting the log of total refugees with the logged measure of GDP per capita of the host state. In Models 3 and 4, which both have country fixed effects, and Model 4, which also has year fixed effects, the interaction term for "Refugees X GDP" is statistically significant and negative. The lower-order term for refugees is significant and positive in both models, which shows that the general relationship holds when states have the lowest level of economic capacity, 8 while the lower-order term for GDP per capita is significant and negative in observations with no refugees, as expected. Models 3 and 4 provide combined support for our second hypothesis that states with higher economic capacity can better withstand refugee crises and are less likely to engage in repression as a result of them.
In Models 5-8, we provide some alternative specifications of our empirical model to better assess the robustness of the key findings. In Models 5 and 6, we test our interactive prediction employing an alternate indicator of economic capacity, Mousseau's (2015) measure for a contract-intensive economy. In Model 5, without year fixed effects, the interaction between total refugees and CIE is significant and negative, as expected. In the stricter Model 6 that includes year fixed effects, the interaction is not significant. In Models 7 and 8, we attempt to understand the relationship between refugees from countries experiencing conflict and economic capacity. In Model 7 of Table 3, the interaction between logged refugees from a civil war country and logged, lagged GDP per capita is significant and negative, as expected. Model 8 displays the results of the model with year fixed effects, but that model is not as supportive. In all, these results indicate that both a domestic stability mechanism (exemplified by the effects of economic capacity) and a transnational security mechanism (exemplified by the effects of refugees fleeing civil wars) may be at work when leaders make decisions regarding repression when facing refugee crises. We believe future work should work to further unpack this relationship.
While the results of the models provide support of our theoretical framework, according to research on interpreting the significance of interaction terms, one cannot rely on statistical significance in hypothesis tests alone to determine whether or not multiplicative terms have substantive significance (Brambor, Clark, and Golder 2006) and must instead calculate marginal effects of the coefficients.

8
Technically, the lower-order terms provide the effect of the term when the other term in the interaction is zero, which is impossible for GDP. That said, as displayed in Figure 1, the effect of refugees when GDP is at the lowest observed logged value (around 5) the effect of refugees is positive and significant.
In order to understand the substantive significance of our key independent variables, we have calculated the conditional marginal effect for refugees on repression, across the range of GDP per capita from Model 3 in Table 3. In Figure 1, we display the conditional effect of refugees across the range of GDP per capita. Continuous controls were set at their mean level and civil conflict was set at zero. What this figure displays is the effect of a one logged unit increase in refugees' impact across increasing levels of economic capacity. Figure 1 clearly shows that states with lower GDP are more likely to increase repression as a result of increased refugees, while that increase in repression become decreasingly likely as economic capacity increases. Our sample mean for GDP is around 8.2, which is right where the line crosses zero. States below about 1 standard deviation (7.05) are more likely to have repression scores above 0, while states about 1 standard deviation above (9.47) our sample mean for GDP (8.26) are much less likely to have repression scores increase. Figure 2 presents a different way of examining the conditional effect, by displaying the effect of refugees on repression at two different levels of GDP per capita, one standard deviation above and below our sample mean for economic capacity. Figure 2 shows that states with high and low economic capacity experience refugees and repression in markedly different ways. The figure shows that at relatively low levels of economic capacity, states hosting large numbers of refugees are more likely to increase repression than comparatively wealthier states. States at one standard deviation below the sample mean for GDP will increase from just above a score of 0 (the mean) of  Table 3. All continuous control variables held at their mean and civil conflict are set to zero. Confidence intervals are set at the 95% level.
repression to a score of 0.5 as refugees approach their maximum amount. For reference, the natural log of 100,000 is about 11 and the natural log of 1 million is almost 14. Figures 1 and 2 display what our theoretical framework expects, that states with high economic capacity will be better able to absorb the shock of refugees and integrate them into their economy, which should lead to fewer grievances among refugees and the local population, and run fewer risks of refugee mobilization, leading to a lower probability of repression. Taken together, the substantive effects displayed in Figures 1 and 2 show that there is a conditional effect of refugees on repression and that not all host states should be expected to react with increased repression. The effect of refugees on repression is moderated by a host state's economic capacity. States with low economic capacity are more likely to increase repression when facing refugee crises than the average state or states with high state capacity.
In all, we received support for our theoretical framework. Hypotheses 1 and 2, which predicted that increased refugees would lead to increased repression by host states (H1) but also that the relationship would be tempered by economic capacity (H2) received either moderate or strong empirical support across many models, as well as employing alternate indicators of economic capacity and refugees. That said, models that incorporate year fixed effects in addition to country fixed effects were not as supportive of our theoretical framework.  Table 3. All continuous control variables held at their mean and civil conflict are set to zero. Confidence intervals are set at the 95% level.

Robustness checks and empirical extensions
While our findings from the models discussed previously are supportive of our theoretical framework, we conducted several robustness checks in order to gauge the empirical limits of our predictions. In addition to the models testing on the Schnackenberg and Fariss (2014) latent repression variable, we conducted additional analysis employing the PTSs (Wood and Gibney 2010). The PTS is based on yearly (beginning in 1976) human rights reports published by Amnesty International and the US State Department. PTS measures physical integrity violations within a country in a given year on a 5-point ordinal scale, in which 5 is the most repressive and a score of 1 is the least repressive. Countries perpetrating widespread, systematic abuses against their citizens get a higher score on the PTS than those that do so rarely. In keeping with common practice, we employ State Department scores with Amnesty scores imputed for countryyears in which State Department scores are missing. These results are largely similar on key models to those displayed above.
We conducted tests on whether or not "inflows" of refugees were more important than stocks of refugees in determining repression. In order to account for this, we subtracted the current year's refugees from the prior year's refugees and took the natural log. When the change in refugees was positive, it represents an inflow. We treated decreases in refugees as zeros. When incorporating this variable into a model similar to Model 3 of Table 3, both the results for logged inflow and logged total refugees were positive and significant, and the interactions for both variables with logged GDP per capita were negative and significant. To us, this indicates that both stocks and inflows of refugees present a "shock" to the political status quo and can alter the repressive calculations of the state. Future research should explore in more detail the key conceptual and empirical differences between inflows and stocks and their effects on political violence. We also conducted tests employing refugee population as a percentage of the total population of the state, but that variable was not significant in a model similar to Model 3 of Table 3, nor did its inclusion affect the significance of our key finding.
Additionally, we conducted tests on models incorporating whether or not a state's neighbor was experiencing a civil war [taken from Salehyan and Gleditsch's (2006) replication data]. 9 We know from prior work that as civil wars become more proximate, states may become more likely to increase repression as a way of preventing dissent within their country from escalating into a rebellion (Danneman and Ritter 2014). Furthermore, the theoretical linkage from hosting large numbers of refugees, and increasing repression because of an upset social balance, and from viewing refugees as a potential threat for conflict escalation is not a difficult linkage to make (Lischer 2005;9 These models' time sample was restricted to end at 2001, because of employing a variable from Salehyan and Gleditsch's data. Salehyan and Gleditsch 2006). In these models, we find that our key variables still attain statistical significance in the expected directions. To us, this suggests that extensions of this work should more fully explore the linkages between neighboring civil wars, increased refugee flows, and repression. While work by Danneman and Ritter (2014) employ refugees as an indicator of proximate civil war, as well as other characteristics of proximate civil wars, another way to test this linkage is to examine the level of repression experienced in those neighboring civil wars and whether particular groups of refugees represent particular threats with respect to civil war diffusion, in order to capture the degree to which civil wars and refugee flows combined form a hostile externally based threat for states, or whether a more domestic mechanism of maintaining social order is key to understanding repressive behavior. This may also help to clear up another theoretical mechanism, which we are unable to do at current levels of data aggregation. Danneman and Ritter (2014) describe a pre-emptive increase in repression (but in reaction to refugees) in order to prevent the spread of civil war. This argument fits with the transnational security mechanism described above. What we are unable to clear up with our approach is whether states increase repression in anticipation of (a pre-emptive motivation) increased dissent or violence because of refugees or in reaction to increased dissent. Because our data are aggregated at the year level, the timing of increased refugees and repressive reactions are difficult to disentangle. Aside from more precise temporal disaggregation, spatial disaggregation would be helpful for this as well. If we could model where increased repression occurs (in locations with high levels of refugees vs. locations where anti-refugee dissent is high), then we might also be better able to disentangle the state's motivations. We also conducted analyses on the effects of refugees from neighboring states. In order to collect these data, we measure refugees from neighboring states by employing the Correlates of War contiguity data (Tir et al. 2008) in order to determine the number of refugees coming in from nearby countries, using their variables for contiguity at 24 miles. Because the Correlates of War (COW) contiguity data are only updated through 2006, we were not able to capture past that for those particular refugee variables. On these samples, which range from 1952 to 2006, we observe similar effects as those displayed above.
The results of our additional analyses incorporating neighboring civil conflict and refugees from neighboring states indicate that the role of local external threat on the relationship between refugees and state repression is complex. Our theoretical framework focuses primarily on how increased refugee flows upset social order, along with the potential for civil war diffusion, but does not aim to specify whether external or internal threat is the root cause. Our findings on state economic capacity suggest that perhaps both domestic and international factors may play a role. That said, our findings from our extended analyses suggest that there are many useful future directions for research aiming to understand the linkages between conflict, refugees, and repression. 10

Conclusion
Refugees are caught between a rock and a hard place-they are people who have left their country of origin under physical threat (Davenport et al. 2003;Moorthy and Brathwaite 2016;Rubin and Moore 2007) and their presence in their host state leads to increased repression as well. Refugees are viewed as disruptions to their destination state's polity and security-either as potential carriers of violence (Choi and Salehyan 2013;Salehyan 2008;Salehyan and Gleditsch 2006) or as economic burdens and competition. This dynamic can lead to grievances against the state, from both refugee and the local population, leading to increased levels of repression by the host state in response. Additionally, states that are unable to pre-empt or address these grievances-states that have weak capacity-are likely to resort to systematic physical integrity abuses when faced with an influx of refugees.
Our findings generally support these two expectations, and our study presents a good step toward a better understanding of how refugees affect political violence, but there are many directions this research could go to gain a more comprehensive understanding. First and foremost, while determining a link between refugees, state capacity, and overall levels of repression in a host state is helpful, there are questions about repression left unanswered. Specifically, who is the state targeting with repression? Nationally aggregated indicators such as the PTS or the latent human rights performance indicator (Schnackenberg and Fariss 2014) do not distinguish between whom the state is targeting with repression. For a better understanding of the micro-processes that underlie the general pattern we are observing in our analysis above, it would be helpful to know if the state is targeting incoming refugees that may mobilize for a diffuse civil conflict, or if they are targeting their local population because of increased grievances due to economic competition. Knowing which repressive path is more likely would lead to a better understanding of which refugee crises may lead to civil conflict diffusion and thus a better understanding of the linkages between repression and civil conflict. Furthermore, distinguishing between groups of refugees might also be helpful for future work. Second, Lischer (2005) and some migration studies (Neumayer 2005) make us think that some countries consider some "outsiders" less threatening and dissentious. If future work can systematically differentiate between refugee groups to understand whether some types of refugee groups (states in exile) are more likely to lead to a repressive state response, then we would have a much better understanding of these processes.
Despite the fact that we cannot uncover these micro-processes with the empirical strategy employed above, our study has uncovered a general relationship between hosting increased numbers of refugees and increased repression. Furthermore, the conditional moderating relationship of increased economic capacity on refugees and repression indicates that refugee crises need not lead to extra human suffering if a host state is better equipped. Future research might consider examining the degree to which foreign humanitarian aid might also play a role in mitigating the potential for increased repression in the wake of refugee crises, although foreign aid itself is a complicated phenomenon (see Wright and Winters 2010). While Lischer (2005) does focus on the role of aid, as does Choi and Salehyan (2013) with respect to refugees and terrorism, more could be done to understand the link between aid, refugees, and repression. Our results indicate that increasing economic capacity could help ease suffering, but that is a longer term process than increased aid, so more systematically understanding how aid might impact the ongoing dynamics between refugees, the local population, and the host state would be of great use to both policymakers and scholars.