The nexus between asylum seekers and defence spending in European NATO member states: a quantitative study of securitisation dynamics

ABSTRACT

This paper examines the relationship between the defence spending of European NATO members and their exposure to asylum seekers. While research shows that threat perceptions and domestic politics partly determine the defence spending of a state, the effect of migration has so far not been studied. This is remarkable because migration is increasingly framed as a security threat, also by NATO and European Union member states. Using a panel analysis, this paper explores the relationship between the defence spending of 23 European NATO members and the number of asylum seekers they register each year between 2000 and 2020. Results show a positive and significant relationship between the number of asylum seekers registered in a country and its overall military expenditure relative to GDP, equipment spending, and infrastructure spending. These findings shed new light on the origins of defence spending in collective security alliances, at a time when transatlantic burden sharing is at the centre of societal and academic debates.

KEYWORDS:

• NATO
• defence spending
• migration
• asylum seekers
• threat perception
• burden sharing

Introduction

In the last decades, research has convincingly shown that migration has become increasingly securitised in Europe (Adamson 2006; Colomé-Menéndez, Koops, and Weggemans 2021; Miholjcic 2017). Even the European Union (EU) frames migration alongside other security concerns such as terrorism and transnational criminal activity (Asderaki and Markozani 2021; Huysmans 2000; Léonard and Kaunert 2020). What is notable is that migration is also framed as a security threat by the North Atlantic Treaty Organization (NATO). The communiqué of the 2016 NATO Warsaw Summit establishes a direct connection between instability in the Middle East and North African regions, and the refugee and migrant crisis. More specifically, the communiqué characterises “massive flows of migrants” as “challenges and threats for international stability, security, and prosperity” (NATO, 2016). In particular, migration is not only perceived as a security threat by many political parties and governments, it is also an increasingly militarised issue (Nemeth 2018; Oliveira Martins and Strange 2019; Panebianco 2021). Since the 2015 migration crisis, some European countries (e.g. Hungary, Slovenia, Austria) assign the management of refugee flows to their armed forces (Nemeth 2018). While these securitisation and militarisation dynamics have been well-explored, we know little about their effect on defence policy. More particularly, how they affect defence spending commitments remains so far unaddressed. This is remarkable, because since Russia’s full-scale invasion of Ukraine in February 2022, both issues of defence spending and collective responses to migration crises have regained prominence in debates on NATO and EU burden sharing.

Scholars interested in transatlantic defence burden sharing have so far paid particular attention to the effects of domestic political and institutional factors, and threat perceptions on defence spending in collective security alliances (e.g. Becker 2019a, 2020; Becker and Malesky 2017; Eichenberg and Stoll 2003; Haesebrouck 2021). Studies mainly addressed the traditional and non-traditional security threats posed by Russia’s assertiveness and by transnational terrorism, respectively. Most recent analyses show that a higher threat perception and threat proximity tend to be positively associated with defence spending in NATO and Europe (Becker and Malesky 2017; Kim and Sandler 2020). Moreover, domestic political factors encompass right- and left-wing party politics (Potrafke 2011; Bove, Efthyvoulou, and Navas 2017), and authoritarian populism (Becker 2020), while institutional factors for example include common EU fiscal rules (Becker 2019a). The relationship between migration and defence spending has not been explored by burden sharing scholars, despite its increased securitisation including in NATO documents. This paper seeks to address this gap by answering the question: what is the relationship between European NATO members’ exposure to migration and their defence spending?

To answer this question, this paper follows the common line of reasoning that perceptions and proximity of security threats tend to increase defence spending. After showing that migrants are increasingly perceived as a security threat in Europe, the guiding hypothesis of this paper is therefore that the more a NATO member state is exposed to migrants flows, the higher its defence spending will be. Defence spending is operationalized with overall military expenditure relative to GDP, and four additional categories of disaggregated defence spending relative to GDP. Data on the number of asylum seekers registered in NATO member countries on a yearly basis is used to estimate their exposure to migrants (Migration Data Portal 2022). This study also assesses the relationship between right-wing populist politics and defence spending, for which data from the TIMBRO Authoritarian Populism Index is employed. To cope with a lack of data on bilateral migration, two ordered categorical variables are generated to estimate countries’ territorial exposure to refugees from Syria, and their territorial proximity to North Africa. The analysis covers the 2000–2020 period and the number of European NATO member states included as observations varies between 21 and 23 according to the availability of data for each variable. A panel analysis is applied because of the time-series and cross-sectional nature of the data.

The analysis shows that European NATO members in which more asylum seekers are registered are more likely to increase their military expenditure, equipment spending and infrastructure spending between 2000 and 2020. Moreover, countries in which right-wing populist parties receive a more important share of votes tend to spend a higher share of their GDP on military expenditure and defence equipment. Interestingly, being territorially more exposed to refugees from Syria tends to decrease the percentage of GDP that NATO members spend on the military overall, as well as on defence personnel, equipment, and infrastructure. Territorial proximity to North Africa is not significantly correlated with defence spending among NATO member states. These findings suggest that the increasing securitisation of migration might influence how much countries spend on defence. By proposing one of the first analyses on the link between migration and defence spending of European NATO members between 2000 and 2020, this paper contributes to ongoing academic and policy debates on defence spending behaviour and burden sharing in security alliances. Moreover, the results of this study provide a major contribution to the literature on burden sharing and the origins of defence spending because they show the importance of addressing the evolution and construction of security threats.

State of the art

Securitisation and militarisation of migration

Research indicates that most NATO members increasingly fear that migration flows will jeopardize their national security (Adamson 2006; Böhmelt and Bove 2019; Dreher, Gassebner, and Schaudt 2020). The numerous terrorist attacks that hit European countries between the 1970s and 1990s and the 9/11 terrorist attacks against the United States changed the way in which certain countries perceive migration, sometimes framing it as a major issue of national security (Adamson 2006; Miholjcic 2017; Colomé-Menéndez, Koops, and Weggemans 2021). According to research, there is no clear evidence that migration fuels more terrorism (Bove and Böhmelt, 2016). Nonetheless, migration tends to be securitised and instrumentalised by right-wing political parties, especially in Western countries (Helbling and Meierrieks 2020). Helbling and Meierrieks (2020) and Choi (2019) highlight how right-wing politicians scapegoat migrants to create an anti-immigrant sentiment among voters and avoid delving into the political and economic challenges that their country faces. Marine Le Pen’s political campaign for the French presidency in 2017 offers a clear illustration thereof, as her party claimed that stopping migrants from entering France could limit the threat posed by transnational terrorism (Benedicto and Brunet 2018; Choi 2018; Ilgit and Klotz 2018). Right and far-right wing parties in Italy, Poland, Hungary, Germany, Denmark and the Netherlands similarly frame migration as a vehicle for terrorism or as a threat to the cultural identity and national security of their country (Benedicto and Brunet 2018; Miholjcic 2017). Also, NATO and the EU express serious concern about the potential security threat posed by migration. For instance, the communiqué of the 2016 NATO Warsaw Summit establishes a direct connection between instability in the Middle East and North African regions, and the refugee and migrant crisis (NATO, 2016). Likewise, the EU frames migration alongside other security concerns like terrorism and transitional criminal activity, thus criminalizing migrants and more particularly asylum seekers (Asderaki and Markozani 2021; Huysmans 2000; Léonard and Kaunert 2020). Hence, migration flows have been increasingly presented as a security issue over the past years, both by governments and intergovernmental organisations.

In addition to a growing securitisation of migration issues, the 2015 refugee crisis in Europe has led to an increasing militarisation of migration (Musaro 2017; Nemeth 2018; Panebianco 2021). At that time, Hungary, Slovenia, and Austria deployed troops to supply equipment to police forces that were charged of driving back migrants to the border, and in some cases to take on border control tasks. Panebianco (2021) writes that the Hungarian and Italian governments became “obsessed with the defense of the EU borders” and “they constructed a rhetoric of state’s protection from the security threats (migration included) that clearly emerged at the (intergovernmental) institutional level” (p. 1403). At an international level, the EU also tackled migration issues, especially smuggling, by conducting military Operation Sophia (2015–2020) in the Mediterranean. Moreover, NATO’s standing naval forces are still deployed in military operations in the Mediterranean Sea and the Aegean Sea to fight smuggling and provide help to refugees. Some argue that such military involvement in the Mediterranean reinforce discourses on the security threat created by influxes of migrants (Oliveira Martins and Strange 2019). Research should therefore not omit to address the repercussions of growing flows of migrants on defence commitments.

Defence burden sharing

Scholarly interest in NATO burden sharing is almost as old as the Alliance (Hartley and Sandler 1999; Khanna and Sandler 1996; Kennedy 1979). The early research of Olson and Zeckhauser (1966) indicates that the NATO burden of collective security tends to be systematically unevenly shared by its members. The authors theorized the economics of alliances based on the observation that an ally’s Gross National Product (GNP) was positively correlated with its defence spending, and that smaller and less powerful allies gained more from their NATO membership than larger allies (Olson and Zeckhauser 1966). These states can be characterized as “free-riders”, and collective security in NATO should be understood as a public good (Olson and Zeckhauser 1966). Later, Sandler and Forbes (1980) found that from 1967 onwards, the collective security provided by NATO to its members corresponded to a “joint product model” that accounted for “private, impure public, and pure public outputs of defence expenditures” (Sandler and Forbes 1980, 425). This means that burden sharing scholars need to be attentive to shifts in the way states approach security and threats, and to the use of the military. These analyses of burden sharing identified GNP, Gross Domestic Product (GDP), population, and “exposed borders” as the main drivers of countries’ defence spending (Kim and Sandler 2020; Olson and Zeckhauser 1966; Sandler and Forbes 1980).

The origins of defence spending

Recent research on defence burden sharing seek to offer a more nuanced explanation of defence spending. A first strand of literature on the relationship between security threats and defence spending has so far offered varying results. Haesebrouck (2021) writes that Russia is perceived as the biggest security threat by NATO allies since the Crimean crisis. The full-scale invasion of Ukraine by Russia in February 2022 might reinforce the fear of the Russian threat among NATO members – not least among former Soviet republics. Research has found that Russia’s military expenditure is positively and significantly correlated with how much NATO members spent on NATO operations between 1999 and 2012 (Becker and Malesky 2017). George and Sandler (2018) expect that the fear of a Russian military offensive lead states to increase their defence spending, capabilities, and contributions, but their statistical analysis produces opposite results. Interestingly, the same study reveals that after the 9/11 terrorist attacks and until 2015, NATO member states that were most targeted by transnational terrorist attacks tended to spend more on defence (George and Sandler 2018). In another article, Becker and Malesky (2017) find that the number of victims caused by terrorist attacks is not significantly correlated with NATO members’ disaggregated defence expenditures. In sum, these findings suggest that it is not entirely clear what the impact of security threats is on NATO members’ defence spending.

A second strand of contemporary literature examines the effect of domestic politics on states’ defence spending. Defence spending is primarily shaped by domestic political and institutional factors (Becker 2017), even though international factors might also influence how much a country decides to spend on its defence (Gartz and Gleditsch 2004; Lumsdaine 1996; Rathbun 2007; Hofmann 2013). Many studies explored how left and right party ideologies can predict defence spending but led to contradictory findings (Eichenberg and Stoll 2003, Potrafke 2011; Whitten and Williams 2011; Bove, Efthyvoulou, and Navas 2017). Some even argue that party ideology only has a minimal effect on defence spending choices (Whitten and Williams 2011). Other recent studies indicate that populist politics are partly driven by high unemployment, which also reduces a state’s defence expenditures (Becker 2021), and tight EU fiscal rules, which tend to shift defence budget allocation (Becker 2019a). Yet, a recent study of Becker (2020) shows that populist parties do not necessarily have to be represented in a government to affect national defence spending, as NATO burden sharing is affected by the share of votes received by authoritarian right-wing populist parties within ally countries. Left populist policies, however, do not significantly alter defence resource allocation. The same study also highlights that right-wing populist politicians tend to be radically hostile to migrants. Yet, Becker does not investigate how migration might be associated with defence spending. It is even more remarkable that research burden sharing has not yet made the connection between the migration and defence, because recent securitisation studies found that current international issues (such as migration) are growingly framed as security threats.

In sum, we know that migration is increasingly regarded as a threat to national security by European countries (Adamson 2006; Colomé-Menéndez, Koops, and Weggemans 2021; Miholjcic 2017). Government and intergovernmental organisations such as the EU and NATO also take part in the securitisation and militarisation of migration (Nemeth 2018; Oliveira Martins and Strange 2019; Panebianco 2021). Most notably, NATO now also shows concerns about the security issue posed by migration, hence suggesting that migration has spilled over into debates on national defence (NATO 2016). We know that threat perceptions tend to affect defence spending of NATO members (Becker 2019b, 2020; Becker and Malesky 2017; Eichenberg and Stoll 2003; Haesebrouck 2021). However, we lack empirical insights about whether and how migration affects defence spending. This paper therefore examines the relationship between European NATO members’ exposure to migration and their defence spending.

Analytical framework

Following the common practice in the burden sharing literature, the analysis partly builds on the economics of alliances theory (Olson and Zeckhauser 1966; Sandler and Forbes 1980) by controlling for the theoretical assumptions that higher GDP and population are positively correlated with defence spending. Burden sharing studies typically compare NATO members’ defence spending with the security benefits they derive from alliance membership. However, it would be odd to add migration to the list of more traditional security threats commonly used to calculate the security benefits NATO members receive – such as their territorial proximity to Russia – given that defence (spending) is not explicitly presented by NATO as an appropriate response to migration flows (NATO 2006). Rather, as the literature shows that migration is increasingly perceived as a security threat, this paper focuses on exploring whether migration can be a potential domestic origin of defence spending.

Based on NATO documentation and recent burden sharing studies, defence spending is conceptualized as overall military expenditure as a share of GDP (“cash”), and defence spending on personnel, equipment, infrastructure (“capabilities”) and operations and maintenance (“contributions”) relative to GDP (Becker 2020, 2021; Stoltenberg 2019). The analysis therefore focuses on the relationship between migration flows and defence spending, i.e. overall military expenditure relative to GDP as well as the share of GDP spent on various categories of defence spending.

Building on studies on the securitisation and militarisation of migration issues, it appears that political scapegoating discourses which associate migration with transnational terrorism rarely distinguish between different types of migration (Benedicto and Brunet 2018; Choi 2019; Helbling and Meierrieks 2020). However, research shows that the scapegoating of migrants particularly increases during migration crises (Musaro 2017; Nemeth 2018; Panebianco 2021). To operationalise the exposure of countries to migration crises, one could therefore look at the number of asylum seekers or refugees registered in European NATO members. Based on definitions of migration (Amnesty International 2019; United Nations International Organization for Migration 2019), it seems that focusing on the number of refugees may be biased. It is highly dependent on the asylum policy of the host country and therefore does not capture the actual number of people present in the host country during migration crises. For instance, a country may have a high number of asylum seekers but a low number of refugees due to strict asylum policies. Focusing only on the number of refugees would exclude a potentially large number of asylum seekers present in the country who have not been or will not be recognised as refugees. Because of this, the number of asylum seekers, which measures how many people have applied for refugee status, is the most accurate indicator of a country’s exposure to migration flows, especially in times of migration crisis. Accordingly, the following hypotheses will be tested:

H1:

The higher the number of asylum seekers in a NATO member, the higher its defence spending will be.

Recent research on the domestic political origins of defence spending highlights that right-wing populist politics tend to be associated with higher hostility to migrants (Choi 2019; Helbling and Meierrieks 2020). This leads to the following two hypotheses:

H2:

The higher the number of votes received by right-wing populist parties in a country, the more the higher its defence spending will be.

H3:

As the number of asylum seekers increases in a NATO member, the effect of the share of votes received by right-wing populist parties in a country on defence spending will increase.

The securitisation literature and NATO documents emphasize the 2015 Syrian refugee crisis as the most recent migration crisis that affected NATO members, and more particularly the 21 EU members of NATO. The civil war in Syria raised security concerns for states that had to manage increasing migration flows, especially for those that are territorially closer to Syria. It also fuelled scapegoating political discourses that presented Syrian refugees as a channel for the diffusion of transnational terrorism (Benedicto and Brunet 2018; Chrisafis 2016; Choi 2019; Ilgit and Klotz 2018). This leads to the following hypothesis:

H4:

The more a NATO member is territorially exposed to inflows of refugees from Syria, the higher its defence spending will be.

In addition to the Syrian crisis, the most recent UN World Migration Report (2022) explains that in 2019, around 19 million African migrants were displaced in another African country than the one they were born in. Yet, the number of bilateral migrations between African countries and other regions of the world is collected every five years since 1990, which results in an insufficient amount of data for a regression analysis. Therefore, here the analysis focuses on assessing the effect of territorial proximity of NATO member countries to Africa. Based on the same rationale as that for H4, the next hypothesis is:

H5:

The closer a NATO member’s territory is to North Africa, the higher its defence spending will be.

As 22 out of 31 NATO member countries are also members of the EU, one must take into account the evolution of EU asylum regulations.¹ The Dublin Regulations set the rule that refugees would apply for asylum in the first EU country they enter (European Parliament Regulation 604/). This system has been criticized for imposing a greater burden on southern EU countries, for example during the 2015 migration crisis, when a large number of Syrian refugees flooded into Greece (Vie Publique 2021). The current Dublin III Regulation, which entered into force in 2014, does apply to all EU member states except Denmark. Iceland and Norway apply the Dublin III Regulation based on an agreement with the EU. Thus, one can argue that NATO members that are part of the Dublin Convention are more exposed to asylum seekers than those who are not, which leads to the following hypothesis:

H6:

NATO member countries that are part of the Dublin Regulations tend to make greater defence spending.

Data and methods

To test the hypotheses, a dataset is created combining data from open access databanks and datasets commonly used in research on NATO burden sharing. The dataset also includes several dummies and ordered categorical variables that were generated to cope with issues of data availability.

Dependent variables

In this paper, “defence spending” is measured with five dependent variables. A first dependent variable captures countries’ overall military expenditure relative to GDP. While military expenditure has often been criticized for being a “standard burden sharing indicator of the Cold War period” (Haesebrouck 2017, 2244; see also Oma 2012), the share of GDP that each member allocates to its military budget is an indication of its indirect funding to the alliance (NATO 2021). One should also note that NATO members have committed to a target of 2% of their GDP spent on defence by 2024; therefore, states can show their political support to the Alliance when they reach this 2% target.

To better assess how defence budgets are spent and how this explains states’ defence “capabilities” and “contributions”, the analysis relies on four additional dependent variables taken from NATO (2022) classification of defence expenditures (Figure 1). The share of GDP allocated to defence personnel, equipment and infrastructure capture the “capabilities” dimension of defence spending. Drawing on recent burden sharing research, defence “contributions” is operationalized with operations and maintenance spending relative to GDP (e.g. Becker and Malesky 2017). The source of data used to measure disaggregated defence expenditures is Becker’s 2021 dataset. The author used NATO documentations to obtain these data. For greater clarity, a description of the type of expenses covered by each of the categories of personnel, equipment, infrastructure, and operations and maintenance is presented in the Appendix.

Figure 1. Distribution of the dependent variables.

Independent variables

The number of pending asylum applications in a country at the end of the previous year is retrieved from the Migration Data Portal, which is provided by the International Organization for Migration’s Global Migration Data Analysis Centre. There are missing values for Latvia, Luxembourg, Norway, Portugal and Spain. The natural logarithmic of the number of asylum seekers is used in order to reduce the skewness of the data and facilitate the interpretation of the coefficients (Figure 2). This first key independent variable allows to test H1. To test H2, this paper utilizes the number of votes received by right-wing populist parties in a country during the latest election to the national parliament. These data are taken from the TIMBRO Authoritarian Populism Index (2019) and are only available for European countries from 1980 to 2010, with some disruptions (Figure 3).

Figure 2. Evolution of the number of asylum seekers (using log transformation).

Figure 3. Evolution of right-wing populist votes and unemployment.

The exposure to refugees from Syria is operationalized with an ordered categorical variable, inspired by Becker’s, 2019b own computations of the “Russian threat”. The variable territorial exposure Syria accounts for the evolution of the Syrian civil war and its consequences in terms of migration on NATO member countries. The computations are somewhat experimental but rely on strong evidence in both the literature and migration data (United Nations High Commissioner for Refugees 2013; 2016, 2022c; International Centre for Migration Policy Development 2020 International Organization for Migration 2022). The categories of the variable are organised in hierarchical order according to countries’ growing exposure to refugees from Syria (see Table 1). Within each category, the temporal dimension of the war in Syria is also taken into account, which explains why the peak of the conflict in terms of population displacement in 2015 and 2016 (UNHCR 2022b) corresponds to the highest value assigned to the group of countries in each category. The only exception to this is Turkey, which is put in the highest (most exposed) category because most Syrian populations that moved across Syria’s border since 2011 found refuge in bordering countries, one of which is Turkey (Connor 2018; UNHCR 2022c). Although the United States and Canada are not counted as observations in the analysis which zooms in on European NATO members, they could be considered the least exposed NATO member to refugees who have fled Syria. UNHCR statistics (2016) as well as contemporary studies (e.g. Melchionni 2018) allow for the identification of states considered to be part of the Western Balkan Route (category 3). These countries are considered to be the frontline transit area for refugees from Syria (International Centre for Migration Policy Development 2020). Below category 3, the classification reports that European countries with a Mediterranean coastline (category 2) are more exposed to migration from Syria across the Mediterranean Sea than other European countries that are not located on the Western Balkan route, and do not have a Mediterranean coastline (category 2) (International Centre for Migration Policy Development 2020). For countries that fall into different categories, the variable takes the value corresponding to the highest of all categories.

Table 1. Description of the variable territorial exposure Syria.

Category	Region	Variable takes value	Countries
	All	0 if year ≤ 2010	All NATO member states
1	Europe	1 if 2011 ≤ year ≤ 2014 2 if 2015 ≤ year ≤ 2016 1,5 if year ≥ 2017	Belgium, Czech Republic, Denmark, Estonia, Germany, Iceland, Latvia, Lithuania, Luxembourg, Netherlands, Norway, Poland, Portugal, Slovakia, United Kingdom
2	Mediterranean Sea	3 if 2011 ≤ year ≤ 2014 4 if 2015 ≤ year ≤ 2016 3,5 if year ≥ 2017	France, Italy, Spain
3	Western Balkans route	5 if 2011 ≤ year ≤ 2014 6 if 2015 ≤ year ≤ 2016 5,5 if year ≥ 2017	Albania, Croatia, Bulgaria, Hungary, Greece, North Macedonia, Montenegro, Slovenia
4	Border with Syrian Arab Republic	8 if 2011 ≤ year ≤ 2016 7,5 if year ≥ 2017	Turkey

Drawing on the same rationale as that of the variable territorial exposure Syria, an ordered categorical variable is generated to operationalize territorial proximity North Africa. The ordered categories are computed based on data presented in the 2022 World Migration Report of the International Organization for Migration, which reveals that Europe is currently the primary continent of destination for migrants from Africa. The most exposed countries to these migrations are those located on the Central, Western and Eastern Mediterranean routes, i.e. Italy, Spain, and Turkey and Greece, respectively. In contrast to the variable territorial exposure Syria, here the variable solely focuses on countries’ territorial proximity with North Africa. It is more difficult to assess the security context in Africa than in Syria (e.g. presence of war or violent conflict), as the later very much varies dependent on the regions and countries of the African continent. Alternatively, one might consider interacting territorial proximity with the number of migrants from or per African countries arriving in a host country, which is commonly referred to as “bilateral migration” (World Bank 2022a). However, data of the World Bank on bilateral migration are presented by blocs of five years and are available from 1990 to 2019, which results in an insufficient number of observations for this quantitative analysis. Thus, this paper only focuses on the territorial proximity of each country to North Africa. Northern European NATO members are put in the less exposed category (category 1), followed by Eastern and Western European countries (category 2), Central European countries (category 3) and countries situated on the so-called Mediterranean routes (category 4). A visualization of the five ordered categories is presented in Table 2.

Table 2. Description of the variable territorial proximity North Africa.

Category	Region	Variable takes value	Countries
1	Northern Europe	1	Denmark, Iceland, Norway, United Kingdom
2	Eastern and Western Europe	2	Albania, Belgium, Bulgaria, Croatia, Estonia, Latvia, Lithuania, Luxembourg, Montenegro, Netherlands, North Macedonia, Portugal, Romania
3	Central Europe	3	Czech Republic, France, Germany, Hungary, Poland, Slovakia, Slovenia
4	Mediterranean routes	4	Greece, Italy, Spain, Turkey

To test H6, a simple dummy variable is generated based on EU documentation on the Dublin Regulation (Regulation 604/2013; European Commission 2022). Observations are coded 1 for NATO member countries that are part of the Dublin Regulation, and 0 for those that are not, or that were not until they became EU member states. One exception is the case of Denmark, whose application of the Dublin Convention, Dublin II Regulation and Dublin III Regulation varied over time. Accordingly, Denmark complied to the Dublin Regulation between 1997 and 2003, and between 2006 and 2013.

Control variables

Based on the economic models of burden sharing (Olson and Zeckhauser 1966; Sandler and Forbes 1980; Sandler and Shimizu 2014), the analysis includes the natural logarithm of GDP and population as control variables (Figure 4). This accounts for the fact that some NATO members have greater economies and population, which are two indicators of state power. One should find a positive correlation between GDP and defence spending, as well as between population and defence spending (Olson and Zeckhauser 1966; Sandler and Forbes 1980). Moreover, defence spending is expected to decrease as NATO member states are territorially less exposed to Russia. The territorial exposure of a country to the “Russian threat” is measured with this country’s capital distance to Moscow. Data for this variable is taken from Becker’s 2020 Transatlantic Security Dataset. Although the years 2022 and 2023 are not covered by the analysis, the Russian invasion of Ukraine in February 2022 might raise NATO members’ security concern even more. In that sense, future studies could resort to other measures of the “Russian threat” than inverse capital distance. Finally, research shows that NATO members tend to reduce their overall military expenditure relative to GDP as unemployment increases (Becker 2021). Using data from the World Bank (2022b), the analysis thus includes a control variable on the unemployment rate in a country (Figure 3). A summary of the concepts, measurements and hypotheses is presented in Table 3.

Figure 4. Evolution of GDP and population (using log transformation).

Table 3. Summary of the concepts, measurements, and hypotheses.

Concept	Measure	Hypothesis	Expectation
Dependent variables measuring defence spending Military expenditure Personnel spending Equipment spending Infrastructure spending O&M spending	% GDP spent on the military % GDP spent on defence personnel % GDP spent on defence equipment % GDP spent on defence infrastructure % GDP spent on defence operations and maintenance
Threat independent variables Asylum seekers Territorial exposure Syria Territorial proximity Africa	Log number of pending asylum applications Ordered categorical variable (own computations, see Table 1) Ordered categorical variable (own computations, see Table 2)	H1 H4 H5	+ + +
Political and institutional independent variables Right-wing populist votes Dublin regulations	Share of votes received by right-wing populist parties Dummy variable (0 – not part of Dublin Regulations; 1 – part)	H2 H6	+ +
Interaction asylum seekers*right-wing populist votes		H3	+
Control variables GDP Population Territorial proximity Moscow Unemployment	Log GDP Log population Log country $\mathit{i}$’s capital distance to Moscow Unemployment in % labour force		+ + - -

Model of Defence spending

This study relies on quantitative time-series and cross-sectional data. The following model is tested for the five dimensions of defence spending:

$\mathrm{DEF}\mathrm{SPEN}{\mathrm{D}}_{\mathrm{i},\mathrm{t}}=\mathit{f}(\mathrm{ASYLU}{\mathrm{M}}_{\mathrm{i},\mathrm{t}-1}+\mathrm{POPULIS}{\mathrm{M}}_{\mathrm{i},\mathrm{t}}+\mathrm{ASYLU}{\mathrm{M}}_{\mathrm{i},\mathrm{t}-1}\times \mathrm{POPULIS}{\mathrm{M}}_{\mathrm{i},\mathrm{t}}+\mathrm{SYRI}{\mathrm{A}}_{\mathrm{i},\mathrm{t}}\text{break}+\mathrm{AFRIC}{\mathrm{A}}_{\mathrm{i}}+\mathrm{DUBLI}{\mathrm{N}}_{\mathrm{i},\mathrm{t}}+\mathrm{GD}{\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}+\mathrm{POPULATIO}{\mathrm{N}}_{\mathrm{i},\mathrm{t}-1}+\mathrm{MOSCO}{\mathrm{W}}_{\mathrm{i}}+\mathrm{UNEM}{\mathrm{P}}_{\mathrm{i},\mathrm{t}-1})$

Where $\mathrm{DEFSPEN}{\mathrm{D}}_{\mathrm{i},\mathrm{t}}$ is defence spending, which is then subdivided in five categories, corresponding to the five dependent variables employed in this study. The first, $\mathrm{MILE}{\mathrm{X}}_{\mathrm{i},\mathrm{t}}$ corresponds to military expenditure relative to GDP in a country $\mathit{i}$ during year $\mathit{t}$. $\mathit{DEF}\mathit{PER}{S}_{\mathrm{i},\mathrm{t}}$ is to the proportion of GDP spent on defence personnel, $\mathit{DEF}\mathit{EQUI}{P}_{\mathrm{i},\mathrm{t}}$ the share of GDP spent on defence equipment, $\mathit{DEF}\mathit{INFR}{A}_{\mathrm{i},\mathrm{t}}$ the share of GDP spent on defence infrastructure, and $\mathit{DEF}\mathit{O}{M}_{\mathrm{i},\mathrm{t}}$ defence expenses on operations and maintenance relative to GDP.

Regarding the key independent variables of the models, $\mathrm{ASYLU}{\mathrm{M}}_{\mathrm{i},\mathrm{t}-1}$ is the log number of pending asylum applications in a country i in year t-1, $\mathrm{POPULIS}{\mathrm{M}}_{\mathrm{i},\mathrm{t}}$is the share of votes received by right-wing populist parties in the latest parliamentary election in a country i during year t. As illustrated in Tables 1 and 2, $\mathrm{SYRI}{\mathrm{A}}_{\mathrm{i},\mathrm{t}}$ corresponds to the exposure to refugees coming from Syria of a country i during year t, and $\mathrm{AFRIC}{\mathrm{A}}_{\mathrm{i}}\mathit{\hspace{0.17em}}$ to the territorial proximity of a country i to North Africa. Moreover, $\mathrm{DUBLI}{\mathrm{N}}_{\mathrm{i},\mathrm{t}}$ is the dummy variable indicating whether a country i is part of the Dublin Regulations in year t. The model includes the four control variables described before. $\mathrm{GD}{\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}$ and $\mathrm{POPULATIO}{\mathrm{N}}_{\mathrm{i},\mathrm{t}-1}$ are the GDP and the population of a country i in year t-1, respectively. $\mathrm{MOSCO}{\mathrm{W}}_{\mathrm{i}}\mathit{\hspace{0.17em}}$ is the control for a country i‘s capital distance to Moscow, and $\mathrm{UNEM}{\mathrm{P}}_{\mathrm{i},\mathrm{t}-1}$ for its rate of unemployment during t-1 year.

One final regression analysis is presented for each dependent variable. For each model, the time unit of the analysis is one year. The analysis spans from 2000 to 2020 with some disruptions. This choice for this limited timeframe is driven by two considerations. First, compared to a timeframe that would cover the entire existence of NATO, this specific focus allows for better capturing the most important changes in the migration context in Europe, especially the migration crisis that followed the civil war in Syria from 2011 onwards. Second, it offers sufficient observations to use regression analysis. The unit of analysis is a NATO member state. A country is counted as an observation from the first year it became a member of NATO. Due to limited data on the dependent variables, some countries are not systematically included in the analysis, which can be seen in the tables and graphs of summary statistics presented in the appendix. The United States and Canada are removed from the observations as the analysis focuses on European NATO countries. Turkey cannot be counted as an observation because of missing data on the share of votes received by right-wing populist politicians (the TIMBRO index does not cover party politics in Turkey). Moreover, Iceland is not included as an observation at all because the country does not have a military.

The analysis starts by running a Pooled Ordinary Least Square (OLS) regression. Using the pooled OLS Estimator can lead to a violation of the Gauss-Markov theorem when strong unit heterogeneity exists, or if unit-specific factors are correlated with one or more independent variables (Hsiao 2003). Subsequently, five Random Effects (RE) regressions are computed, corresponding to the five dependent variables. The RE Estimator is a weighted average of variation between units and variation within units and assumes that unit heterogeneity is uncorrelated with the regressors. This is a difficult assumption to satisfy, as unobserved heterogeneity is arguably often correlated to the independent variables (Bartels 2005). In other words, the RE Estimator is appropriate if differences between NATO members influence the dependent variable. A Breusch-Pagan Lagrange-Multiplier-Test reveals a panel effect in the data sample. Therefore, the null hypothesis can be rejected, and a panel analysis can be used instead of the Pooled OLS Estimator.

The next step is to employ a Fixed Effects (FE) Estimator, which assesses changes within NATO members only. Due to their very constant nature, time-invariant attributes are not considered as potential cause for changes within countries (Kohler and Kreuter 2009), which is why the time-invariant independent variables SYRIA and MOSCOW are omitted in FE regression outputs. As presented below, the FE Estimator computes changes between country-years observations and the overall mean. It also controls for the omitted variable bias because constant unit-specific effects are allowed to be correlated with the independent variables (Hanck et al. 2020). Using a Hausman test allows to assess whether the unit-specific errors are correlated with the regressors (independent variables). Here, the Hausman test enables to choose between a RE and a FE after running each regression. Results of the Hausman tests are presented below the final regressions (see Table 4). Eventually, the standard errors are clustered for each model to deal with the issues of heteroskedasticity and serial correlation (Becker 2021).

Table 4. Models of disaggregated defence spending.

	(1)	(2)	(3)	(4)	(5)
Estimator VARIABLES	FE MILEX	FE DEF PERS	RE DEF EQUIP	RE DEF INFRA	FE DEF OM
ASYLUM	0,038* (0,022)	−0,020 (0,022)	0,017*** (0,007)	0,004* (0,002)	0,013 (0,011)
POPULISM	0,013** (0,006)	0,003 (0,004)	0,005*** (0,002)	−0,000 (0,001)	0,000 (0,002)
ASYLUM * POPULISM	−0,002** (0,000)	−0,001 (0,001)	−0,000** (0,000)	0,000 (0,000)	0,000 (0,000)
SYRIA	−0,041*** (0,011)	−0,039*** (0,010)	−0,009* (0,005)	−0,004*** (0,001)	−0,009 (0,006)
AFRICA	omitted	omitted	−0,027 (0,039)	0,002 (0,008)	omitted
DUBLIN	−0,155 (0,096)	−0,059 (0,062)	−0,062* (0,038)	0,008 (0,007)	−0,042 (0,035)
GDP	−0,225*** (0,076)	−0,211** (0,085)	−0,073*** (0,026)	−0,028*** (0,008)	−0,119** (0,044)
POPULATION	−1,136 (1,012)	−0,915 (0,632)	0,090** (0,038)	0,019** (0,009)	−0,308 (0,328)
MOSCOW	omitted	omitted	0,024 (0,048)	−0,041 (0,026)	omitted
UNEMP	−0,016** (0,006)	−0,007* (0,004)	−0,011*** (0,002)	−0,001* (0,001)	−0,006** (0,003)
Constant	25,917 (15,741)	21,676 10,232	0,014 (0,438)	0,753*** (0,267)	8,479 (5,257)
Obs. Number of groups Overall R-squared	354 23 0,090	321 21 0,056	353 23 0,139	353 23 0,210	353 23 0,017

Notes: Asterisks indicate *** p<001, ** p<0,05, * p<0,1. Robust standard errors are presented in parentheses. Results of the Hausman tests are: (1) Prob > chi2 = 0,007; (2) Prob > chi2 = 0,002; (3) Prob > chi2 = 0,240; (4) Prob > chi2 = 0,105; (5) Prob > chi2 = 0,068.

Table 4 displays the results for the five dependent variables of military expenditure and disaggregated defence spending. Due to missing data on right-wing populist votes for non-European countries, and missing data on defence spending, between 21 and 23 NATO member states were included as observations in the final regressions (see Appendix).

Results

Results show that a greater number of asylum applications is positively and significantly correlated with NATO members’ overall military expenditure relative to GDP. The variable ASYLUM also shows the expected effect on defence spending on equipment and infrastructure. For these two regression analyses, the null can be rejected and H1 accepted. However, there no significant effect of ASYLUM on DEF PERS and on DEF OM.

Interestingly, the analysis sheds light on a positive and significant correlation between votes received by right-wing populist parties and overall military expenditure, as well as with defence equipment spending. In contrast, there is no significant relationship between right-wing populist votes and defence spending on personnel, equipment, and operations and maintenance. Therefore, there is mixed support for H2, which can only be accepted for the models of military expenditure and equipment spending.

As for the interaction term used to test H3, results demonstrate that the positive effect of a growing number of asylum seekers on military expenditure and equipment spending tends to become less significant as the number of votes received by right-wing populist parties increases. This is the opposite result of what was hypothesized. Moreover, this interaction term has no significant effect in models of personnel spending, infrastructure spending, and operations and maintenance spending. The null must therefore be accepted for these three models.

Research on the securitisation of migration predict that closer territorial proximity to a country characterized by population displacement will increase the threat perception and scapegoating practices of migrants in Western countries (Chrisafis 2016; Benedicto and Brunet 2018; Choi 2019; Ilgit and Klotz 2018). Yet, between 2000 and 2020, NATO member countries that were territorially more exposed to refugees from Syria tended to significantly decrease the share of GDP they dedicated to their overall military expenditure, personnel spending, equipment spending and infrastructure spending. This relationship shows the opposite sense of what was expected in H4. Table 4 also shows that greater territorial exposure to refugees from Syria is not significantly correlated with spending on operations and maintenance. Moreover, the variable AFRICA is omitted in the three regression that use a FE Estimator. The variable has no significant influence on equipment spending, and also no significant effect on operations and maintenance spending. H5 must therefore be rejected, and the null hypothesis accepted.

The models of defence spending show poor support for H6. Indeed, the variable DUBLIN is only significantly yet negatively correlated with spending on defence equipment, while H5 expected a positive correlation between the two variables. Therefore, whether a NATO member state applies the Dublin regulations or not has no significant influence on its defence spending.

The first control variable GDP is negatively and significantly correlated with all five dependent variables. This finding is surprising, as burden sharing studies so far found that an increase in a country’s GDP tend to increase its defence spending. The second control variable POPULATION displays the expected positive and significant effect in models of equipment spending and infrastructure spending only. Such result is also somewhat unexpected, as studies of burden sharing generally find that the most populated NATO members are also those who put greater efforts on their defence. The control for the territorial proximity of a country to Moscow is omitted in three of the models as it is a time-invariant variable which the FE Estimator does not account for. Yet, the analysis suggests that there is no significant correlation between NATO members’ defence equipment and infrastructure spending and their territorial proximity to Moscow. Lastly, UNEMP is the only control variable for which all five the regression analyses exhibit the predicted effect. A higher unemployment rate in NATO members will tend to decrease these countries’ defence spending in terms of overall military expenditure, defence spending on personnel, equipment, infrastructure and operations and maintenance. A summary of the results is presented in Table 5.

Table 5. Significance of the results.

Note: This is a summary of the results presented in Table 4. Blue backgrounds are used for variables that show the effect that was hypothesized, while grey backgrounds apply to variables whose effect shown in the model is the opposite of the expected one. Non-significant results are identified by the absence of a coloured background. ▴ is used to show when time-invariant variables are omitted in models using a FE Estimator.

Discussion

This study provides mixed results on the relationship between asylum seekers and the defence spending of European NATO members. The key findings from the models are that there is a correlation between the number of asylum seekers registered in European NATO countries, as well as the popularity of right-wing populist parties, and defence spending over the period 2000–2020. It is important to see these findings as a generalisable pattern across 23 European NATO countries, and by no means as a causal relationship.

Two main implications can be drawn from these findings and suggest that more research is needed to clarify how NATO countries, at a national level, may or may not be considering migration issues in their defence spending plans. First, domestic politics seem to be the strongest driver of defence spending in European NATO members. As suggested by the negative coefficient obtained by interacting the variables ASYLUM and POPULISM, there is a point where the effect of a growing number of asylum seekers on defence spending becomes less strong when the representation of right-wing populist parties in parliament increases. That is, migration issues may become less securitised in the public debate, but a country may still increase its share of GDP spent on defence if, for example, right-wing populist parties are strongly represented in parliament. It may also be the case that even several years after a migration crisis, some countries – especially those where right-wing populist parties have strong political representation – consider that parts of the defence budget should be used to manage migration flows. Although it is unclear when and where these securitisation dynamics end, the results of this paper show that they can have long-term effects on defence spending. Ultimately, the impact of the securitisation and scapegoating of migrants on future defence priorities is arguably a societal concern at a time when right-wing populist parties are gaining seats in national parliaments across Europe.

Second, there may be some other domestic dynamics occurring within European NATO members. Some countries where anti-immigration discourses are prominent because of their territorial exposure to migrants may not actually spend more on defence. The negative relationship between territorial exposure to refugees from Syria and defence spending, and the absence of a statistically significant relationship between territorial proximity to North Africa and defence spending, indicate that this can be the case. More research is needed on this aspect, for example through case studies based on an analysis of parliamentary debates and national defence documents.

Finally, it is interesting to note that the models show no statistically significant relationship between the territorial proximity of European NATO countries to Russia and their expenditure on defence equipment and infrastructure as a share of GDP between 2000 and 2020. This adds to the mixed findings in the burden sharing research on the extent to which the fear of a Russian military offensive on NATO territory – which tends to increase the closer these countries are to the Russian border – can be a driver of defence spending. It can be expected that defence spending as a share of GDP will increase in the Eastern European NATO countries from 2022 onwards. The question remains as to how scholars can best capture these countries’ perceptions and fears of Russia.

Conclusion

Research on the origins of defence spending is of paramount importance at a time when security in Europe seems more fragile than ever. The renewed Russian invasion of Ukraine in February 2022 has reignited academic and political debates about the price that NATO members are willing to pay to ensure their ability to defend themselves collectively. Contemporary studies have shown that NATO member states’ exposure to security threats such as terrorism and Russia can be positively correlated with their defence spending relative to GDP (e.g. Becker and Malesky 2017; Kim and Sandler 2020). Moreover, recent studies on the securitisation of migration have demonstrated that migration is nowadays perceived as a security threat in Western countries, including NATO member states (Adamson 2006; Colomé-Menéndez, Koops, and Weggemans 2021; Miholjcic 2017). Therefore, this paper sought to better understand whether and how flows of migrants can affect European NATO members’ defence spending.

This paper finds that a higher number of asylum seekers is positively correlated with military expenditure, equipment spending and infrastructure spending among European NATO members. Moreover, the share of votes received by right-wing populist parties in NATO members is positively and significantly correlated with their military expenditure and spending on defence equipment. Surprisingly, the interaction between the number of asylum seekers and the share of votes received by right-wing populist parties tends to decrease overall military expenditure and equipment spending among NATO member countries. Closer territorial exposure to refugees from Syria tends to decrease military expenditure, as well as personnel, equipment, and infrastructure spending. Moreover, closer territorial proximity to North Africa displays no significant effect in the empirical analysis. Research on burden sharing and the origins of defence spending in NATO has so far not included any a measure of a country’s territorial exposure to (refugees from) Syria, nor to North Africa. As such, this provides scope for more extensive research on the influence that territorial proximity to conflict regions have on NATO members’ defence spending. Finally, the analysis reveals that NATO members that apply the Dublin regulations tend to decrease the share of GDP they spend on defence equipment.

By proposing one of the first studies on the link between exposure to migration flows and defence spending in NATO, this paper contributes to ongoing academic and policy debates on defence spending and burden sharing in security alliances. In light of the current war in Ukraine, it is important to note that the models presented in this paper overlook the displacement of more than six million Ukrainian refugees since the beginning of 2022 (UNHCR 2022a). This is because there is currently insufficient time series data on these refugees over several years. The burden sharing scholarship would strongly benefit from more research considering the effect that the Russian invasion of Ukraine will have on NATO members’ defence spending.

Acknowledgments

The helpful feedback from Dr. Yf Reykers and members of the panel Defence Cooperation and Military Assistance at the 2022 European Initiative for Security Studies conference were highly appreciated.

Disclosure statement

No potential conflict of interest was reported by the author.

Data availability statement

The author confirms that the data supporting the findings of this study are available within the article and its supplementary materials.

Additional information

Funding

The finalisation of this paper was funded by the Gerda Henkel Foundation [Grant for project AZ 12/KF/22].

Notes on contributors

Daphné Charotte

Daphné Charotte is a PhD candidate on civil society and miltiary transparency at the Faculty of Arts and Social Sciences at Maastricht University.

Notes

1 At the time of writing, Finland had not yet become a NATO member. Therefore, Finland is not included as an observation in the analysis.

Appendix

Table A. Categories of defense spending (NATO, 2022b).

	Nomenclature of NATO defense expenditure
	Personnel	Equipment	Infrastructure	Operations and maintenance
Type of expenses covered	Military personnel and civilian personnel: pays and allowances, employer’s contributions to retirement funds, pensions	Missile systems, missiles (conventional weapons), nuclear weapons, aircraft, artillery, combat vehicles, engineering equipment, weapons and small arms, transport vehicles, ships, harbour craft, electronic and communication equipment, research and development devoted to major equipment	National military construction, NATO common infrastructure: expenditures as host nation, payments to other nations, land and utilities	Ammunition and explosives (excluding nuclear), petroleum products, spare parts, other equipment and supplies, rends, other operations and maintenance

Table B1. Summary statistics of the dependent variables.

Variable	Obs	Mean	Std. dev.	Min.	Max.
Military expenditure	953	2,206	1,180	0,421	6,943
Personnel spending	728	1,146	0,620	0,171	4,342
Equipment spending	907	0,395	0,334	0,006	1,733
Infrastructure spending	760	0,071	0,067	0,000	0,581
O&M spending	758	0,497	0,373	0,068	2,615

Note: the distribution of the dependent variables is presented in graphs below.

Table B2. Summary statistics of the categorical and dummy independent variables.

Variable	Obs	Mean	Std. dev.	Min.	Max.
Territorial exposure Syria	1266	0,640	1,543	0	8
Territorial proximity North Africa	1277	2,189	1,255	0	4
Dublin	710	0,720	0,449	0	1
Territorial proximity Moscow	1277	7,741	0,562	6,672	8,964