NATO at 70: Pledges, Free Riding, and Benefit-Burden Concordance

ABSTRACT

For 2011–2017, this paper revisits NATO burden sharing in light of recent developments and pledges to bolster members’ defense shares of GDP to 2%. Russian nationalism, enhanced transnational terrorism, and intrastate conflicts are apt to increase the publicness of NATO defense spending over the last eight years. When NATO allies’ defense shares of GDP are correlated with their GDP ranks, there is clear evidence of the exploitation of the large, rich allies by the small, poor allies, indicative of allies sharing purely public defense spending since 2011. Such exploitation and free riding has rarely been seen after 1967. In addition, there is an absence of concordance between NATO allies’ defense burdens and their derived benefit shares, consistent with greater defense publicness. Finally, we find further proof of exploitation and free riding for a broad-based measure of security spending.

KEYWORDS:

• NATO burden sharing
• joint product model of alliances
• exploitation hypothesis
• benefit-burden concordance
• free riding

JEL CLASSIFICATION:

• D74
• H42
• H56

Introduction

Since its founding in 1949, the North Atlantic Treaty Organization (NATO) has grown from 12 – Belgium, Canada, Denmark, France, Iceland, Italy, Luxembourg, the Netherlands, Norway, Portugal, the United Kingdom, and the United States – to 29 members. Four additional allies later joined the original members: Greece and Turkey (1952), West Germany (1955, becoming unified Germany in 1990), and Spain (1982). During 1999–2009, NATO expanded from 16 to 28 allies, an expansion of 75%. Montenegro joined in 2017 and North Macedonia will soon enter, thus bringing the alliance membership to 30. The average life span of a military alliance during the last 500 years is 15 years, with just ten surviving 40 years or more (The Economist 2019, 3). NATO is one of the most enduring and successful alliances in history. NATO’s success is tied, in part, with its victory in the Cold War in 1990–1991. Thereafter, NATO redefined its primary mission from deterring the ex-Soviet Union’s westward expansion to protecting NATO members’ interests in out-of-area places (Article 6) with peacekeeping missions in Bosnia, Kosovo, Iraq, Afghanistan, Libya, and elsewhere (North Atlantic Council 2014, 2018; Sandler and Murdoch 2000). With the collapse of European communist regimes, NATO allies took advantage of a peace dividend as they redirected some of their military expenditure (ME) to domestic programs, resulting in a fall in their share of gross domestic product (GDP) devoted to defense – i.e., ME/GDP fell.

These declining defense shares of GDP were greater for some NATO allies than for others. The great recession of December 2007 to June 2009 meant that allies needed to focus on monetary and fiscal stimuli to repair banking institutions and commercial enterprises, thereby turning allies’ attention further away from defense spending and procurement. In 2010, only five NATO allies allocated 2% of their GDP to ME (Stockholm International Peace Research Institute [SIPRI] 2019). The need to bolster NATO allies’ ME was highlighted in the departure speech of Robert Gates (2011) in which the then US Secretary of Defense characterized NATO as a two-tiered alliance. In particular, Gates viewed many NATO allies as incapable of “conducting ‘hard’ combat missions owing to aging weapons, insufficient research and development (R&D), reduced forces, inadequate training, and deficient ME. Gates’ criticism, emerging challenges, Russian defense buildup, and its provocative acts resulted in the Wales Summit Declaration of 2014 in which NATO allies pledged to contribute at least 2% of their GDP to defense spending and to increase their procurement of new military equipment to 20% of their annual defense budget within the next decade (North Atlantic Council 2014, 2018; Sandler and George 2016).

Noteworthy recent exigencies underscore the importance of an invigorated NATO. In February 2014, Russia annexed Crimea from Ukraine, which was followed by Russian support of insurgents in Eastern Ukraine. These Russian provocations heighten NATO’s concern about the territorial integrity of some eastern European members of NATO. In response, these eastern allies wanted to augment NATO’s presence on their soil, along with joint NATO exercises. Some, but not all, of these eastern allies have increased their defense shares of GDP to or near 2% in 2018 – i.e., Estonia, Latvia, Lithuania, Poland, and Romania (see Section 3). With Putin-fueled nationalism, Russia is again seen as a threat to NATO, thus calling for enhanced defense budgets by NATO allies. This worry is heightened by the US announced withdrawal from the Intermediate-Range Nuclear Forces Treaty (INF Treaty) in February 2019 in response to alleged Russian violations of this treaty. Thus, NATO may be confronted with a new arms race that includes enhanced ballistic missile capabilities and greater military utilization of outer space. Recent years also witnessed an increase in transnational terrorism by religious fundamentalist groups, such as Islamic State in Iraq and Syria (ISIS), al-Qaida, and their affiliates that target some NATO allies’ soil and interests (Hou, Gaibulloev, and Sandler 2019).

The worry about a two-tiered alliance and free riding dates back to the years following NATO’s origins. Virtually every US president and congress have raised these concerns – the Trump administration is not unique in this regard. During the 1950s and early 1960s, the nuclear allies – the United States, the United Kingdom, and France – devoted a much greater share of their GDP to ME when compared to smaller European allies in NATO. Their defense allocations were due, in part, to the procurement of their strategic arsenal. In a now classic contribution, Olson and Zeckhauser (1966) argue that the pure publicness of deterrence in NATO will result in an ‘exploitation hypothesis’ whereby the large, rich allies shoulder a disproportionately large defense burden in terms of ME/GDP for the small, poor allies (also see Olson 1965; Russett 1970; Sandler 1977). Olson and Zeckhauser (1966) find a significant positive correlation between defense burdens (ME/GDP) and GDP for NATO in 1964, based upon a Spearman rank correlation test. These authors only tested this correlation for 1964 to support their exploitation hypothesis. Since their exploitation finding, there have been numerous articles about NATO burden sharing for alternative time periods and burden-sharing measures (e.g., Hartley and Sandler 1999; Khanna and Sandler 1996, 1997; Kollias 2008; Sandler and Forbes 1980; Sandler and Hartley 1995; Solomon 2004; van Ypersele de Strihou 1967). The basic message of such studies is that changes in NATO’s strategic doctrine, its weapons technology, its membership, and its perceived threat affect the mix of public, impure public, and private (ally-specific) benefits derived from the allies’ ME, thereby influencing burden sharing (Murdoch and Sandler 1984; Sandler and Hartley 1999, 2001; Sandler and Murdoch 2000). Notably, the positive rank correlation between ME/GDP and GDP, indicative of the exploitation hypothesis and free riding, seldom held for NATO after the late 1960s (see Sections 2 and 3).

The purpose of the current study is to extend the study of Sandler and Shimizu (2014) to investigate NATO burden sharing during 2011–2017 when the Russian threat and that of transnational terrorism assumed a greater importance. Our extension underscores the impact of recent changes because our findings for exploitation, free riding, and, to a lesser extent, burden-benefit concordance differ markedly from 1999–2010, the period analyzed by Sandler and Shimizu (2014). Like their study, we examine burden sharing from three alternative perspectives: within-ally burdens, between-allies burdens, and broad-based security burdens. Within-ally burdens are based on each ally’s ME/GDP; between-allies burdens are founded on each ally’s shares of total NATO ME; and general security burdens are tied to each ally’s spending on ME, overseas development assistance (ODA), and UN peacekeeping spending. Between-ally burdens allow the researcher to gauge the concordance or agreement between allies’ defense spending burdens and their derived shares of the alliance’s benefits. A significant concordance supports a joint product model of defense for which allies’ ME provide a sizeable share of excludable benefits (Sandler and Forbes 1980).

Our study is appropriate for this 30^th anniversary issue for at least two reasons. First, the NATO burden-sharing debate is very much in the news, fueled by often disparaging remarks about NATO burden sharing by the Trump administration. Russian provocative actions in Ukraine and Russian efforts to influence elections in the United States and elsewhere also underscore the concern over NATO burden sharing. Second, Defence and Peace Economics (DPE) has been interested in NATO from the journal’s founding in 1990. Two of the six articles in the inaugural issue of DPE were on NATO – i.e., Hansen, Murdoch, and Sandler (1990) and McGuire (1990). Since then, DPE has published numerous articles on the demand for defense by allied countries (Bruce 1990; Christie 2017; Douch and Solomon 2014; Hilton and Anh 1991; Nikolaidou 2008; Skogstad 2016; Solomon 2005) and on NATO burden sharing (Khanna and Sandler 1996, 1997; Kollias 2008; Solomon 2004).¹

Our paper contains three noteworthy findings. During 2011–2017, there is evidence of the exploitation hypothesis for NATO, rarely seen after 1967. This evidence is consistent with an increased pure publicness of allied defense spending in light of current threats and continued intrastate instabilities. Additionally, in recent years, there is little or no concordance between defense burdens and defense benefits received, so that NATO allies do not appear to be assuming defense burdens according to impurely public and ally-specific benefits. This finding is also consistent with allies’ ME providing a predominant amount of alliancewide purely public defense benefits in recent years. Such benefits work against NATO spending pledges. Finally, there is indisputable evidence in recent years of an exploitation of the large, rich allies by the small, poor allies when security burdens are correlated with allies’ GDP.

The remainder of the paper contains seven sections. Section 2 provide a brief review of NATO past burden-sharing behavior, followed in Section 3 by the anticipated influence of recent changes on NATO burden-sharing predictions. Section 4 introduces the data and their sources. In Section 5, within-ally burden sharing and the exploitation hypothesis are examined for select years and 2011–2017. Section 6 investigates the concordance between defense burden and benefit; Section 7 studies the exploitation hypothesis for a broader security spending measure. Concluding remarks in Section 8 complete the paper.

NATO Burden Sharing – A Brief Review

In an influential article, Olson and Zeckhauser (1966) characterize NATO as sharing a purely public defense good in the form of deterrence of the Warsaw Pact, led by the ex-Soviet Union (Murdoch 1995; Murdoch and Sandler 1982). This deterrence is based on the strategic doctrine of mutual assured destruction (MAD) by which any Warsaw Pact action against the territorial integrity of NATO’s European allies would be met by a swift launch of nuclear ballistic missiles that inflict catastrophic damages on the aggressor. Ideally, MAD deterrence results in non-rival benefits for NATO allies, insofar as benefits from the threatened retaliation does not diminish with the number of allies on whose behalf the threat is made. Nonexcludability is more difficult to justify unless NATO possesses first-strike advantage so that the initial volley of missiles eliminates any second-strike capabilities by the Warsaw Pact. Prior to the Cuban missile crisis and the Pact’s subsequent nuclear stockpile buildup, the ex-Soviet Union had few strategic nuclear missiles for a second strike. Once the Soviets later acquired a second-strike capacity, deterrence’s nonexcludability became somewhat problematic since the United States could chose to ignore Warsaw Pact aggression given the devastating consequences that a launch would imply to US people and property. To curtail this potential reticence, the United States stationed large numbers of troops in Europe, especially in West Germany, which coupled with US foreign direct investment (FDI) and commercial interests throughout Europe, meant that Warsaw Pact aggression would greatly harm US interests, thereby bolstering the credibility of US commitment to Article 5 of NATO, where an attack on one ally would be viewed as an attack on all leading to an appropriate response.

The implications of MAD-induced pure publicness of NATO ME are at least threefold (see, e.g., Sandler and Hartley 2001). First, free riding is anticipated in which the large, rich allies shoulder the defense burdens of the small, poor allies.² This exploitation hypothesis is linked with the positive rank correlation between allies’ ME/GDP and their GDP. Second, alliance defense spending is viewed as suboptimal because allies do not account for the protective gains that their defense spending provides to other allies. Third, the pure publicness of allied defense expenditures implies that there is no crowding as more allies join; hence, there is no grounds to limit alliance size.

Following the publication of Olson and Zeckhauser (1966), evidence mounted against the purely public representation of NATO ME. In particular, Russett (1970) only uncover a significant positive rank correlation between ME/GDP and GDP for 1950–1967 and not thereafter. Russett’s result is confirmed by Sandler and Forbes for more years for which they only find a positive rank correlation for 1960–1966 and not after, except for 1973 (also see Khanna and Sandler 1996). Ironically, the publicness of NATO defense spending apparently changed shortly after the publication of Olson and Zeckhauser (1966), based on burden-sharing behavior.

To address this apparent change, a joint product model of alliances is put forward where allies’ defense spending yields multiple outputs that vary in their extent of publicness (Russett 1970; Sandler 1977; van Ypersele de Strihou, Jacques 1967). Strategic deterrence still provides purely public benefits, but (protective) conventional forces supply impurely public benefits, subject to thinning as given conventional forces are spread along a longer front or border. Thinning is a crowding concept that necessitates restricting the number of allies (Sandler 1977). Allied defense efforts may also yield ally-specific or private benefits that are excludable and rival among allies. For instance, an ally’s defense may supply national guard, home disaster relief, uprising protection, and flood control. Many other country-specific benefits may be associated with an ally’s ME. Thus, an ally’s ME can offer excludable (border protection and country-specific) and nonexcludable (deterrence) benefits. The share of excludable to total defense benefits determines burden sharing, exploitation, suboptimality, and alliance size restrictions (Sandler 1977). If, for example, allied defense is fully excludable, then burden sharing will be based on benefits received, there will be no exploitation or suboptimality, and alliance size must be founded on thinning considerations. There is then an anticipated concordance between defense burdens carried and benefits received. If, however, all allied defense benefits are nonrival and nonexcludable, then the Olson-Zeckhauser concerns apply including exploitation and free riding. Any changes to strategic doctrine, alliance size, or threats may influence the share of excludable to total defense benefits, thereby affecting burden sharing.

One highly relevant strategic change occurred around 1967 and beyond with directive MC 14/3 and the doctrine of flexible response, which permitted NATO to respond in a measured fashion to Warsaw Pact aggression. With this doctrine, a small conventional force incursion, initiated by the Warsaw Pact, would be met with commensurate conventional countermeasures. Challenges would be met with appropriate sequential responses. The nuclear option would be a last-resort response, deployed only after the challenge spiraled out of control. In light of this doctrine, NATO’s strategic, tactical (i.e., short-range nuclear forces), and conventional forces must be used in conjunction, thereby making allies’ defense efforts complementary to one another (McGuire and Groth 1985; Murdoch and Sandler 1984).³ As a consequence, free riding on conventional forces became more difficult. An ally that possesses inferior conventional forces to protect its own border and interior could invite Warsaw Pact challenges. To limit this eventuality, NATO European allies were motivated to build up their conventional arsenals and manpower during the 1970s and 1980s, and did so as documented in empirical studies (see, e.g., Murdoch and Sandler 1984; Sandler and Forbes 1980; Sandler and Hartley 2001). In fact, there was a noticeable shift in defense burdens from North America back to Europe during the initiation of this doctrine. This shift resulted in the absence of disproportionate burden sharing or exploitation. Moreover, the increase in excludable benefits, associated with flexible response, produced a concordance between defense burdens and defense benefits received, indicative of the joint product model (Khanna and Sandler 1996; Sandler and Forbes 1980). This concordance also held for the 1990s (Sandler and Murdoch 2000).

Recent NATO Changes and Burden-Sharing Predictions

There are at least four crucial NATO developments in recent years, each of which has burden-sharing implications. First, out-of-area missions in North Africa, the Middle East, and Asia (i.e., Afghanistan) gained prominence since 2000. As such, these missions are apt to create purely public benefits for the allies. Regional stability achieved by any contingent of NATO peacekeepers provide nonrival and nonexcludable benefits for NATO allies by fostering global stability, stemming refugee flows, encouraging trade, and bolstering FDI. Second, there is an augmented transnational terrorism threat posed by networks of Islamic fundamentalist terrorist groups that target some NATO allies’ soil and interests (e.g., Belgium, Canada, France, Germany, the United Kingdom, and the United States). Actions that weaken these terrorist groups provide nonrival and nonexcludable benefits to all NATO allies in the cross-hairs of these groups. Third, the rise of Russian nationalism creates inherent challenges to democratic principles, thereby threatening virtually all of NATO. Efforts to curb this threat offer purely public benefits to NATO countries. Russia also presents territorial threats to NATO allies contiguous to or in the proximity of the Russian Federation. Ameliorating these threats create alliancewide gains. Fourth, some allies’ addiction to the peace dividends of the 1990s places unfair defense burdens on NATO allies responding to new defense challenges.

The increased publicness of NATO defense spending has two clear implications. There is likely to be a return to the exploitation hypothesis last prevalent during 1950–1967. This return is not only in terms of narrowly defined defense spending, but also in terms of a more broadly defined security spending (see Section 7). This enhanced publicness is also apt to eliminate the concordance between defense burdens and benefits received, seen from 1990 through 2002 (Sandler and Shimizu 2014).

In Table 1, we display NATO allies’ shares of GDP devoted to ME for each year during 2010–2018 based on Stockholm International Peace Research Institute (SIPRI) (2019) data. All NATO members’ shares are displayed except for Iceland, which had almost no defense spending. In 2010, six allies were at or near spending 2% or more of their GDP on defense. In 2018, just five allies – Estonia, France, Greece, Turkey, and the United States – allocated 2% or more of their GDP to defense. Another four allies – Latvia, Lithuania, Poland, and Romania – allocated over 1.9%, but less than 2%, of their GDP to ME. Notably, five of NATO’s eastern allies were taking the 2% pledge seriously. Other frontline allies – Albania, Bulgaria, Croatia, the Czech Republic, Hungary, Slovak Republic, and Slovenia – were not near the 2% goal in 2018. Given the Wales Summit Declaration, only 6 of 28 allies displayed a decline in ME shares of GDP between 2015 and 2018, which included two eastern European allies, the United Kingdom, and the United States (also see George and Sandler 2018). However, the increases of ME shares of GDP for most allies were very small. These ME/GDP ratios are similar among a large number of countries, consistent with the notion of defense spending convergence (Arvanitidis, Kollias, and Anatasopoulos 2014; Sandler and George 2016).

Table 1. Military expenditure as a share of gross domestic product: 2010–2018 (in percent).

	2010	2011	2012	2013	2014	2015	2016	2017	2018
Albania	1.56	1.53	1.49	1.41	1.35	1.16	1.10	1.11	1.18
Belgium	1.08	1.04	1.04	1.01	0.98	0.92	0.92	0.91	0.93
Bulgaria	1.77	1.44	1.50	1.62	1.47	1.32	1.42	1.42	1.69
Canada	1.20	1.20	1.12	1.00	0.99	1.15	1.16	1.29	1.25
Croatia	1.70	1.77	1.69	1.65	1.57	1.52	1.36	1.43	1.47
Czech Rep.	1.20	1.09	1.07	1.03	0.97	0.95	1.00	0.97	1.12
Denmark	1.40	1.31	1.35	1.23	1.15	1.11	1.15	1.15	1.18
Estonia	1.70	1.68	1.90	1.91	1.92	2.02	2.08	2.03	2.06
France	2.34	2.26	2.24	2.22	2.23	2.27	2.33	2.35	2.29
Germany	1.35	1.28	1.31	1.22	1.18	1.18	1.19	1.23	1.23
Greece	2.73	2.48	2.41	2.36	2.33	2.45	2.54	2.52	2.39
Hungary	1.03	1.05	1.03	0.95	0.86	0.92	1.02	1.05	1.08
Italy	1.51	1.49	1.44	1.41	1.29	1.21	1.34	1.37	1.33
Latvia	1.09	1.04	0.91	0.94	0.94	1.05	1.47	1.68	1.98
Lithuania	0.88	0.79	0.77	0.76	0.88	1.14	1.48	1.72	1.96
Luxembourg	0.52	0.43	0.42	0.42	0.42	0.48	0.44	0.58	0.61
Montenegro								1.36	1.55
Netherlands	1.33	1.29	1.24	1.17	1.16	1.13	1.16	1.16	1.24
Norway	1.51	1.45	1.40	1.41	1.47	1.50	1.62	1.62	1.61
Poland	1.83	1.79	1.80	1.77	1.90	2.14	1.94	1.90	1.98
Portugal	1.98	2.00	1.91	2.09	1.79	1.79	1.73	1.67	1.78
Romania	1.26	1.30	1.23	1.28	1.35	1.45	1.40	1.72	1.93
Slovak Rep.	1.27	1.08	1.09	0.98	0.99	1.12	1.12	1.10	1.21
Slovenia	1.61	1.30	1.17	1.05	0.97	0.93	1.01	0.98	0.97
Spain	1.38	1.32	1.41	1.27	1.25	1.27	1.13	1.23	1.27
Turkey	2.32	2.08	2.05	1.96	1.90	1.85	2.07	2.07	2.50
UK	2.37	2.29	2.19	2.07	1.95	1.86	1.81	1.77	1.78
USA	4.66	4.58	4.23	3.81	3.48	3.27	3.21	3.11	3.16

These ME/GDP shares come from Stockholm International Peace Research Institute (SIPRI) (2019).

Data

Before we turn to our analysis of NATO burden sharing and burden-benefit concordance, we indicate key independent variables and their data sources. ME is in constant 2017 US dollars (Stockholm International Peace Research Institute (SIPRI) 2019) and GDP is in constant 2017 US dollars (World Bank 2019). As indicated in Table 1, allies’ ME/GDP for selected years are taken from Stockholm International Peace Research Institute (SIPRI) (2019). Population and price indices are from World Bank (2019). The latter indices are required for converting from current year US dollars to 2017 US dollars. NATO allies’ exposed borders equal their coastline plus their land boundaries with non-NATO countries including neutral countries (US Central Intelligence Agency 2019).

When computing derived benefits in Section 6, we include the transnational terrorism exposure for each NATO ally. Transnational terrorist attacks involve two or more countries due to the nationalities of the victims and/or perpetrators, or due to the venue country where the attack occurs (Gaibulloev and Sandler 2019). If a terrorist attack in Spain kills one or more foreign nationals, then the attack is transnational. When non-French ISIS operatives attack in Paris, the incident is transnational. This is also the case for a terrorist skyjacking of a commercial flight from Athens that is force to fly to Beirut or Algiers. By contrast, domestic terrorism is homegrown for which the victims and perpetrators are citizens from the venue country sustaining the attack. We chose transnational, rather than domestic, terrorism from which to gauge NATO allies’ terrorism-based benefits because most NATO members are more concerned about transnational incidents. Furthermore, transnational terrorism is a greater threat than domestic terrorism to an ally’s territorial integrity because, in part, the perpetrator may be based abroad or assisted by another country. Also, transnational terrorism tends to have greater economic implications than domestic terrorism (Gaibulloev and Sandler 2019). Our data on transnational terrorist attacks are drawn from International Terrorism: Attributes of Terrorist Events (ITERATE) dataset (Mickolus et al. 2018). We use two alternative measures for an ally’s exposure to terrorism: (i) share of venue-country-based attacks and (ii) share of victim-nationality-based attacks. According to ITERATE, victims are those who lose property, lives, or liberty because of a transnational terrorist attack. Up to three victims’ nationalities are identified for each incident, which is assigned to each victimized country.

When we construct a broader-based security burden measure in Section 7, allies’ spending on foreign assistance is included, since this assistance may foster world security by making for healthier indigenous populations that are less apt to be hosts for infectious diseases. Also, this ODA can bolster trade, stem refugee flows, quell domestic unrest, and secure resource supply lines, all of which make for a more secure world. NATO allies’ aid contributions are drawn from annual ODA figures in 2017 US dollars (Organization for Economic Co-Operation and Development (OECD) 2019). Security burdens are also tied to NATO allies’ contributions to UN peacekeeping operations (PKOs) based on assessed payments. Total UN PKO annual budgets are taken from UN (2011, 2012a, 2012b, 2014, 2015a, 2015b, 2017a, 2017b). Each ally’s share of these budgets is based on their assessment percentage, which changes at discrete points (UN 2009, 2012, 2015). Multiplying the annual UN PKO spending by an ally’s assessed share gives that ally’s nominal assessment for each sample year. By applying the appropriate price index, we convert each ally’s annual assessment into 2017 US dollars. The support of UN PKOs promotes security by limiting the spread of conflict to neighboring countries, stemming the flow of refugees, limiting the spread of conflict-related terrorism, and fostering the level of FDI and trade.

Exploitation Hypothesis: Within-Ally Burden Sharing

We return to the roots of NATO burden-sharing analysis by examining Spearman rank correlations between allies’ ME/GDP (defense burden) and their GDP for select years starting in 1961. We use a simple rank correlation measure, ${\rho }_{12}$, and two partial rank correlation measures, ${\rho }_{12,3}\mathrm{a}\mathrm{n}\mathrm{d}{\rho }_{12,34}$, that hold GDP per capita and both GDP per capita and exposed borders constant, respectively. There are two hypotheses associated with the corresponding Spearman rank correlation tests. The alternative hypothesis states that, within NATO, there is a positive rank correlation between allies’ GDP and their shares of GDP earmarked to ME. By contrast, the null hypothesis is consistent with there being no rank correlation association between ME/GDP and GDP. A rejection of the null hypothesis is then consistent with the alternative hypothesis of a positive relationship between ME burdens and allies’ GDP. This, in turn, supports the exploitation hypothesis in which the rich allies disproportionately carry the alliance burdens of the poor allies.

To offer some historical context on the exploitation hypothesis for NATO in Table 2, we display the three rank correlation measures for 1961, 1965, and then at five-year intervals until 2010. Thereafter, we list these rank correlations yearly for 2011–2017. For 1961 and 1965, there is evidence of a positive rank correlation that is 0.470 or greater. The partial rank correlation holding GDP per capita and exposed borders constant is significant at the 0.05 level, thus allowing for the rejection of the null hypothesis in support of Olson-Zeckhauser’s exploitation hypothesis and free riding by the poorer allies. Weaker evidence of exploitation is tied to the other two rank correlations that are only marginally significant for these two sample years. During the reign of the doctrine of flexible response and increased impurity of defense benefits, there is virtually no support for the exploitation hypothesis, except for 1985, until 2010. US defense burdens jumped greatly in 1985, which marked the height of the Reagan administration’s defense buildup (Khanna and Sandler 1996; Sandler and Hartley 2001). As a consequence, the United States and few other large allies shouldered a greater portion of NATO’s ME, making for a temporary increase in disproportionate burden sharing based on allies’ GDP.

Table 2. Spearman rank correlations between defense burden (ME/GDP) and GDP.

	${\rho }_{12}^a$	${\rho }_{12.3}^b$	${\rho }_{12.34}^c$
1961	0.516 (0.074)*	0.531 (0.076)*	0.485 (0.014)**
1965	0.484 (0.097)*	0.534 (0.074)*	0.470 (0.018)**
1970	0.204 (0.483)	0.249 (0.411)	0.181 (0.367)
1975	0.178 (0.542)	0.175 (0.568)	0.177 (0.377)
1980	0.358 (0.209)	0.364 (0.221)	0.360 (0.065)*
1985	0.475 (0.076)*	0.522 (0.055)**	0.461 (0.012)**
1990	0.250 (0.368)	0.295 (0.367)	0.244 (0.203)
1995	0.107 (0.705)	0.145 (0.622)	0.116 (0.548)
2000	0.185 (0.462)	0.333 (0.191)	0.217 (0.211)
2005	0.102 (0.628)	0.383 (0.065)*	0.121 (0.408)
2010	0.303 (0.125)	0.410 (0.037)**	0.267 (0.053)**
2011	0.375 (0.054)*	0.491 (0.012)**	0.340 (0.013)**
2012	0.397 (0.041)**	0.522 (0.006)***	0.363 (0.008)***
2013	0.367 (0.061)*	0.494 (0.011)*	0.339 (0.013)**
2014	0.382 (0.051)*	0.502 (0.009)***	0.348 (0.011)**
2015	0.365 (0.062)*	0.496 (0.009)***	0.314 (0.022)**
2016	0.264 (0.183)	0.391 (0.048)**	0.210 (0.132)
2017	0.223 (0.253)	0.404 (0.037)**	0.171 (0.211)

*indicates p < 0.1, ** indicates p < 0.05, and *** indicates p < 0.01. P-values are in parentheses.

Variables: 1 = ME/GDP; 2 = GDP; 3 = GDP/POP; 4 = exposed borders.

${ }^a$Simple rank correlation coefficient.

${ }^b$Partial rank correlation coefficient with GDP/POP held constant.

${ }^c$Partial rank correlation coefficient with GDP/POP and Exposed Borders held constant.

For 2011–2017, there is evidence of the exploitation hypothesis for at least one of two partial rank correlation measures, leading us to reject the null hypothesis in favor of disproportionate burden sharing. With and without exposed borders being held constant, there is no support for exploitation during 2016–2017 – the reason for this is not apparent. Exploitation is supported for these years when GDP per capita is held constant. The general findings for 2011–2017 are noteworthy because Sandler and Shimizu (2014) uncover almost no support of exploitation throughout 1999–2009. Clearly, something changed after 2009 to make for disproportionate burden sharing, which is consistent with NATO ME involving more alliancewide purely public benefits likely stemming from increased Russian nationalism, an enhanced transnational terrorism threat, and continued out-of-area missions.

Concordance between Defense Burdens and Benefit Shares

We now examine post-2010 changes in NATO’s burden sharing from an entirely different vantage based on the concordance between an average benefit share (to be defined) and a between-ally defense burden (Sandler and Forbes 1980). This alternative burden-sharing indicator corresponds to an ally’s ME as a proportion of NATO ME. We put forward three different average benefit share indicators; but, we fully acknowledge that there can be many other conceivable such measures. The first average benefit share is based on GDP, population, and exposed borders. For a given ally, we compute its GDP as a share of total NATO GDP. This proxy highlights an ally’s stake in NATO’s industrial base and is intended to capture potential protection to its livelihood, afforded by NATO defense spending. Next, we compute an ally’s proportion of NATO’s population to measure potential lives spared by NATO’s common defense and deterrence efforts. Third, we calculate an ally’s share of NATO’s exposed borders, where an interior ally (e.g. Luxembourg) possesses no exposed border, to denote a member’s share of the alliance’s potential front guarded.⁴ Because we do not know an ally’s relative preferences for protecting its GDP, lives, or territory, we take a simple average of the three shares to come up with an aggregate average benefit share for an ally (see Solomon 2004 for a dissenting view). Our other two average benefit share indices include GDP, population, exposed border, and transnational terrorist attacks. The latter is either based on the ally’s portion of attacks on its soil (venue) or its portion of attacks on its interests. For these two average benefit indices, we average the shares over the four measures to get a single indictor for each ally.

Next, we describe the concordance comparison between defense burdens and defense benefits received. If, say, French average benefit share is 6%, then France is viewed as receiving 6% of NATO’s potential defense benefits. A concordance between France’s defense burden and its derived benefits would have France assuming around 6% of NATO total ME. A nonparametric Wilcoxon signed rank test indicates whether or not these calculated average benefit shares sufficiently match allies’ assumed defense burdens during a given year. This Wilcoxon test assigns ranks based on the absolute value of the differences between the two measures for each ally in a given year. After computing the sum of these ranks with positive and negative differences, we calculate the z-test statistic using the sum of positive differences. The alternative hypothesis of the rank test for the current application is that the distribution of NATO’s defense burdens and the distribution of its average benefit shares are different; while the null hypothesis is that the two distributions are the same. A large Wilcoxon sum of ranks favors maintaining the null hypothesis of concordance. Not rejecting the null hypothesis implies concordance between allies’ average benefit shares and their defense burdens, which is consistent with the joint product model and a healthy share of excludable defense benefits.

In Table 3, we display defense burdens and average benefit shares for share proxy 3 for all NATO allies, except Iceland, during 2011–2017. There are a number of observations that follow from the table. In particular, there appears to be no concordance between average benefit shares and defense burdens. Moreover, this absence of concordance is extremely stable for each sample year. Based on displayed measure 3 in Table 3, the NATO European allies received about 50% of the derived defense benefits but only shouldered 25 to 30% of the defense burdens, which increased somewhat over time. By contrast, the United States underwrites about 70% of the burdens and receives 30% of the benefits. The most free riding appears to characterize Canada, Italy, Spain, Turkey, and Norway for share measure 3. Other benefit share measures (not shown) display similar patterns. Canada is disadvantaged by any benefit share measure based on exposed borders or territorial area, given its vast uninhabited area and associated exposed border.⁵ Perhaps, surprising, Germany is not much of a free rider based on this average benefit share measure, while most eastern allies are free riders.

Table 3. Defense burdens and average benefit share in NATO using GDP, population, exposed borders, and terrorism against targeted country’s interests (average share 3) as proxies for benefits: 2011–2017.

	2011		2012		2013		2014		2015		2016		2017
	Defense burden	Average benefit share	Defense burden	Average benefit share	Defense burden	Average benefit share	Defense burden	Average benefit share	Defense burden	Average benefit share	Defense burden	Average benefit share	Defense burden	Average benefit share
Albania	0.02	0.15	0.02	0.15	0.02	0.15	0.02	0.41	0.02	0.15	0.02	0.15	0.02	0.13
Belgium	0.46	1.15	0.47	0.93	0.49	0.67	0.51	1.18	0.49	1.42	0.50	1.43	0.49	1.51
Bulgaria	0.07	0.31	0.07	0.30	0.08	0.30	0.08	0.30	0.08	0.29	0.09	0.29	0.09	0.29
Canada	1.66	18.42	1.66	17.58	1.65	18.28	1.76	18.25	2.03	17.63	2.05	17.62	2.35	17.41
Croatia	0.09	0.68	0.08	0.68	0.09	0.68	0.09	0.68	0.09	0.93	0.08	0.67	0.09	0.67
Czech Rep.	0.19	0.48	0.19	0.47	0.20	0.46	0.20	1.23	0.21	0.71	0.23	0.45	0.23	0.75
Denmark	0.36	1.40	0.39	0.92	0.39	0.92	0.39	2.22	0.39	0.90	0.41	0.90	0.41	0.91
Estonia	0.03	0.35	0.04	0.62	0.04	0.35	0.05	0.35	0.05	0.35	0.06	0.35	0.06	0.35
France	5.12	9.41	5.30	6.53	5.68	7.92	6.04	6.69	6.33	7.32	6.53	6.29	6.66	8.54
Germany	3.91	5.13	4.19	4.99	4.25	5.03	4.43	5.04	4.60	4.82	4.76	4.87	5.00	4.89
Greece	0.53	2.00	0.49	1.50	0.49	1.80	0.51	1.99	0.55	1.71	0.57	2.04	0.56	1.44
Hungary	0.11	0.42	0.11	0.68	0.11	0.41	0.12	0.41	0.13	0.40	0.15	0.40	0.16	0.40
Italy	2.71	5.78	2.63	5.36	2.72	5.96	2.63	5.54	2.53	5.14	2.85	5.55	2.91	4.41
Latvia	0.02	0.15	0.02	0.15	0.03	0.15	0.03	0.15	0.03	0.14	0.05	0.14	0.06	0.14
Lithuania	0.03	0.19	0.03	0.19	0.03	0.19	0.04	0.18	0.06	0.18	0.07	0.18	0.09	0.18
Luxembourg	0.02	0.06	0.02	0.05	0.02	0.06	0.03	0.06	0.03	0.06	0.03	0.06	0.04	0.06
Montenegro													0.01	0.08
Netherlands	0.96	2.08	0.94	1.63	0.94	1.41	0.99	1.86	1.00	2.08	1.04	2.23	1.06	1.91
Norway	0.52	2.99	0.56	3.34	0.61	2.84	0.67	2.50	0.67	2.96	0.69	2.82	0.71	2.43
Poland	0.72	1.54	0.77	1.52	0.82	1.83	0.95	2.03	1.14	2.27	1.07	1.86	1.09	2.35
Portugal	1.57	0.60	1.66	0.57	1.56	0.57	1.63	0.57	1.75	0.55	1.61	0.55	1.77	0.84
Romania	0.18	0.84	0.18	0.82	0.21	0.83	0.24	0.82	0.28	1.34	0.30	1.21	0.40	1.39
Slovak Rep.	0.09	0.23	0.09	0.22	0.09	0.54	0.09	0.22	0.11	0.21	0.11	0.21	0.12	0.22
Slovenia	0.05	0.12	0.05	0.11	0.05	0.11	0.05	0.11	0.05	0.11	0.05	0.11	0.05	0.11
Spain	1.57	4.13	1.66	2.61	1.56	3.53	1.63	3.33	1.75	3.51	1.61	3.47	1.77	3.93
Turkey	1.24	4.25	1.33	4.71	1.46	6.58	1.54	5.23	1.60	6.30	1.85	6.82	1.96	6.02
UK	5.01	7.86	5.05	8.09	5.21	8.70	5.31	8.36	5.23	6.56	5.21	6.53	5.12	7.35
USA	72.77	29.29	71.99	35.27	71.21	29.73	69.99	30.27	68.82	31.97	68.03	32.79	66.74	31.30
NATO-Europe	25.57	52.29	26.35	47.15	27.14	51.99	28.26	51.47	29.15	50.40	29.92	49.59	30.91	51.29
NATO- North America	74.43	47.71	73.65	52.85	72.86	48.01	71.74	48.53	70.85	49.60	70.08	50.41	69.09	48.71

Table 4 presents the results of the Wilcoxon signed rank tests for the three average benefit share proxies. In Table 4, the first column indicates the year, the second column contains the sample size (number of allies), and the remaining columns display the rank sums with prob values in parentheses. For all three measures and years, we reject the null hypothesis of concordance at the p = 0.01 level. These results strongly suggest that excludable defense benefits are largely absent in recent years, thereby favoring a preponderance of purely public defense benefits. This preponderance may be attributable to the rise of Russian nationalism, the ongoing transnational terrorism threat, and continuing out-of-area conflicts. This non-concordance stands in stark contrast to many earlier studies on NATO burden sharing (e.g., Khanna and Sandler 1996; Sandler and Murdoch 2000).

Table 4. Wilcoxon signed rank correlations of ME/NATO ME and alternative average benefit shares.

	Sample size	Positive rank sum average share 1	Positive rank sum average share 2	Positive rank sum average share 3
2011	27	41 (0.00)***	54 (0.00)***	42 (0.00)***
2012	27	43 (0.00)***	57 (0.00)***	47 (0.00)***
2013	27	45 (0.00)***	45 (0.00)***	44 (0.00)***
2014	27	57 (0.00)***	76 (0.00)***	44 (0.00)***
2015	27	63 (0.00)***	46 (0.00)***	47 (0.00)***
2016	27	63 (0.00)***	61 (0.00)***	55 (0.00)***
2017	28	68 (0.00)***	56 (0.00)***	52 (0.00)***

*indicates p < 0.1, ** indicates p < 0.05, and *** indicates p < 0.01. P-values are in parentheses.

Average Share 1: GDP share, POP share, and exposed border share.

Average Share 2: GDP share, POP share, exposed border share, and transnational attacks share by venue.

Average Share 3: GDP share, POP share, exposed border share, and transnational attacks against targeted country’s interest share.

Security Burden Sharing

Following Sandler and Murdoch (2000) and Sandler and Shimizu (2014), we examine the exploitation hypothesis based on a more encompassing burden-sharing measure that includes NATO allies defense spending, their UN PKOs financial support, and their foreign aid (ODA) contributions. As mentioned in the Data section, allies’ UN PKO financial support and their foreign assistance payments augment global security in many ways by keeping conflicts contained, ending wars, limiting grievances, improving health and welfare, and promoting trade and investment. Mutual trade interests connect countries’ well-being, thereby increasing the opportunity cost of conflict.

For the broad-based measure of security spending, we merely sum each ally’s spending on ME, UN PKOs, and ODA in 2017 US dollars to generate security expenditure (SE), which is then normalized by the ally’s GDP, also in 2017 US dollars. To offer evidence of disproportionate burdens, we find the Spearman rank correlation between SE/GDP and GDP for a simple correlation and two partial correlation measures. Once again, the latter indicators hold GDP per capita and both GDP per capita and exposed borders constant.

In Table 5, we list the three Spearman rank correlation values by year for 2011–2017. For every sample year, these rank correlations are significant at the 0.05 level or better, thus supporting the exploitation hypothesis for the NATO alliance in very recent years. These findings are consistent with a large portion of the alliance’s defense benefits being purely public among the allies. Our results contrast markedly with those of Sandler and Shimizu (2014) in which there is little evidence of exploitation, except marginally for 2005 and 2008.

Table 5. Spearman rank correlation between security burden (SE/GDP) and GDP.

	${\rho }_{12}^a$	${\rho }_{12.3}^b$	${\rho }_{12.34}^c$
2011	0.514 (0.007)***	0.450 (0.018)**	0.461 (0.001)***
2012	0.498 (0.009)***	0.452 (0.020)**	0.447 (0.001)***
2013	0.480 (0.012)**	0.452 (0.020)**	0.427 (0.001)***
2014	0.480 (0.012)**	0.461 (0.018)**	0.430 (0.001)***
2015	0.510 (0.007)***	0.464 (0.017)**	0.438 (0.001)***
2016	0.455 (0.018)**	0.426 (0.030)**	0.388 (0.004)***
2017	0.424 (0.025)**	0.424 (0.027)**	0.356 (0.008)***

*indicates p < 0.1, ** indicates p < 0.05, and *** indicates p < 0.01. P-values are in parentheses.

Variables: 1 = SE/GDP; 2 = GDP; 3 = GDP/POP; 4 = exposed borders.

${ }^a$Simple rank correlation coefficient.

${ }^b$Partial rank correlation coefficient with GDP/POP held constant.

${ }^c$Partial rank correlation coefficient with GDP/POP and Exposed Borders held constant.

The findings here would be even stronger if we included non-UN PKO support, which from a NATO manpower perspective is primarily supported by the United States, the United Kingdom, and France. We do not use SE/GDP with the inclusion of NATO-supported non-UN PKOs because there are no expenditure figures to drawn upon.

Concluding Remarks

No measure of burden sharing is perfect and this certainly applies to the within-ally (ME/GDP and SE/GDP) and between-ally (ME/NATO ME) measures employed here. The best that one can do is to examine alternative measures to ascertain whether or not tests on these measures confess a consistent tale. In the current study, all three measures indicate that there is a disproportionate sharing of NATO defense burdens and there is little burden-benefit concordance during 2011–2017. Our findings are consistent with NATO defense yielding a large share of purely public defense benefits after 2010, which we link to Russian nationalism-fueled threat, the increased transnational terrorism concern, and continued out-of-area conflicts. Since 1999, NATO has grown greatly in size, thereby increasing the portion of relatively small allies that cannot match the prowess of the large allies. This imbalance may have contributed to the exploitation and non-concordance that the analysis uncovers. Our analysis is not based on political ideology; rather, it follows from science, however imperfect. We fully acknowledge that we use inputs (i.e., spending) rather than outputs (e.g., security level achieved) measures to base our analysis because data on defense outputs is difficult, if not impossible, to find. We also realize that we cannot attribute an ally’s defense spending to particular geographical areas – i.e., we cannot separate US defense expenditures into those that only protect Europe. In a world where threats can come from anywhere and power can be projected great distances, there is little or no reason to compute such separations. The attack on the United States on 9/11 was planned half a world away in Afghanistan. In fact, 9/11 is the only instance where NATO’s Article 5 was ever invoked!

The disproportionate burden sharing uncovered here is apt to continue unless the 2% rule is really implemented, along with its concomitant call for great investment and R&D in military equipment. NATO faces significant problems due to free riding and the disparity among allies’ size and capabilities, which appears to grow with NATO’s expansion. These problems may be addressed if challenges arise that threaten a larger number of NATO allies or a new strategic doctrine makes these allies’ defense spending more essential to alliancewide security. A changed doctrine that emphasizes up-to-date weapons and capabilities would further burden sharing the most.

Acknowledgments

We thank two anonymous reviewers for their very helpful comments on an earlier draft.

Disclosure statement

No potential conflict of interests was reported by the authors.

Additional information

Funding

Sandler’s research was supported by the Vibhooti Shukla Endowment at the University of Texas at Dallas.

Notes

1. Defense demand articles appearing elsewhere include Dudley and Montmarquette (1981), George and Sandler (2018), Hartley and Sandler (1990), McGuire and Groth (1985), Murdoch and Sandler (1982, 1985), Sandler and Murdoch (1990), and Smith (1980; 1995).

2. For demand estimates of an ally’s defense, this free riding is tied to the negative response of an ally to the aggregate defense spending (spillovers) of other allies.

3. Deterrence makes allies’ strategic forces substitutability, thus fostering free riding.

4. We could instead use share of an ally’s territorial area. Results of the concordance tests do not change using area instead of exposed border, since the two measures are highly correlated.

5. Other possible benefit share indices could incorporate allies’ share of FDI, their share of trade, or both.