This paper revisits old concepts and arguments that have surrounded NATO since its formation. This paper will revisit three of the five hypotheses of the Olson-Zeckhauser (OZ) Model of alliance burden sharing and use Poland and the Netherlands as case studies. We are interested in European dynamics and thus have not included the US or Canada in our statistical analysis. The data we used is from the Defence Expenditure of NATO Countries (2010–2017). The hypotheses we will re-test from the OZ model are as follows: hypothesis 1: correlation of size of GDP to percentage of GDP spend on defense; hypothesis 2: correlation between GNP and quota fulfilment; hypothesis 3: negative correlation between national income and infrastructure expenses. To conclude, we will revisit the issue of free riding and discuss some of the alternatives to the 2% GDP benchmark measurement.
By Jan Rempala and Tristan Roeven
The question of burden sharing has long been a contentious topic within NATO. These criticisms are chiefly based around accusations that member states are not pulling their “fair share” of the financial burden. This seems to be the most prominent complaint coming from the United States. In order to explain the perceived free riding of other NATO members in comparison to the United States, the Olson-Zeckhauser Model (OZ) was developed in the 1960s. The model was the result of research sponsored by the US Department of Defense.
Taking common defence as a public good, the OZ model is based on the assumption that a nation’s decisions in regard to devoting resources to any sort of military alliance is based upon national governments’ interests. The OZ model concludes that in any sort of organization or alliance where national interests are being represented, larger nations will end up bearing a disproportionate amount of the costs coupled with smaller members making either little or no contributions.[i] The results from this study, however, must be viewed in the context and time period in which this study was published. The OZ model is generally regarded as being a suitable method for describing the first 15 years of NATO’s behaviour.[ii] The problem that arose from using this model’s method of observing disproportionate burden sharing is that it only focuses on this disproportion. As such, Olson and Zeckhauser’s conclusion is hindered or incomplete. This is noted in numerous studies following the 1960s, first by Sandler in 1977 and again by Murdoch and Sandler in 1982, among others. These studies point to an “increasing relevance of private benefits jointly associated with the production of the public defense commodity”[iii] as being the main cause for why the OZ model fails to account for further empirical discrepancies.
The OZ model in reaching its aforementioned conclusion presents five hypotheses, of which this paper will apply three to the cases of Poland and the Netherlands.
Hypothesis 1: Correlation of size of GDP to percentage of GDP spent on defence
“H1: In an alliance there will be a significant positive correlation between the size of a member’s GNP (GDP for this study) and the percentage of its GNP (GDP) spent on defense.”[iv]
The first hypothesis applied in the OZ model stems from the implication that the bigger the size of a nation, the higher its valuation of the output of an alliance.[v] Essentially, this would mean that larger members of an alliance should devote a larger percentage of their GNP to common defence than smaller nations do.
In the original model, with empirical evidence from 1965, the data looked at the GNP of a country in order to test its hypothesis. This replication of the study will instead look at GDP, due to the significantly greater availability of this data. In the OZ model, several issues with the data were discussed. One of the main problems that the OZ model acknowledges in testing this hypothesis is that it is difficult to measure which part of a country’s total amount spent on defence is a contribution to common defence. To account for this the original study included a “noise” factor in its data collection.
Looking at military spending per NATO members’ GDP, what we can conclude from this hypothesis is that there is a tendency for nations with higher GDPs to spend more on defence. Using data from the NATO communiqué “Defence Expenditure of NATO Countries 2010–2017,” we can observe such a general trend amongst nations with higher GDPs.[vi] Therefore, from a basic comparison we can conclude that the original hypothesis still stands, albeit assuming we can substitute GDP for GNP. However, this does not necessarily mean that the member state is meeting its expected 2% goal even though it is outspending other member states. This point is further elaborated on in hypothesis 2.
In an attempt to replicate the mathematical side of the model to see if there is any sort of statistical significance, we decided to run a rudimentary spearman test for the first hypothesis for the year 2017. We used the spearman test because this kind of test was used in the original OZ model. Unfortunately, we were unable to replicate the “noise” factor used to account for any discrepancies in calculating common defence. We ran this test only for this hypothesis, simply to see whether or not the basic mathematical principles behind it still hold.
Our algorithm was designed to determine if there is still a relationship between GDP and military expenditure (exp2) as a percentage of GDP. The result was a very weak, positive monotonic correlation between GDP and exp2 (r=0.15, n=25, and p>0.001) as exemplified by Figure 1.
Figure 1: Countries’ GDP to Military Expenditure as % of GDP
However, we must caution against taking these results as conclusive or as anything other than a rough test. As mentioned before a truly comparative and thorough study of the OZ model would have to replicate all hypotheses.
Hypothesis 2: Correlation between GNP and quota fulfilment
“H2: In a voluntary organization with quota assessments that are not always satisfied, there will be a significant positive correlation between a member’s GNP and the percentage fulfillment or over fulfillment of its quota.”[vii]
As stated earlier in Hypothesis 1, a larger state can spend more money on defence than other member states yet not reach its 2% quota. Hypothesis 2 attempts to elaborate on such behaviour.
When applying this hypothesis to Poland and the Netherlands, it would appear that the OZ model begins to fall short. Of course, the model was never intended to be applied on such a miniscule (i.e., per-country) scale; however, the conclusions made from the original study should be applicable for any country What we see in this case is that while the Netherlands has a higher GDP than Poland, the Netherlands has a lower rate of contribution to NATO than Poland does in regard to meeting the 2% “quota” limit. (Figure 2)
Figure 2: Adapted from “NATO, Public Diplomacy Division (29 June 2017), Defence Expenditure of NATO Countries (2010-2017)[Press release]
The Netherlands, as shown, only contributes 1.17% of its GDP towards defence expenditures. Yet, the GDP of the Netherlands in 2017 was 908 billion USD (Table 1). Poland, on the other hand, had a GDP of 476 billion USD in 2017 and contributed 2.01% (Table 1).
Table 1: GDP per Capita (Thousands US Dollars) Adapted from NATO, Public Diplomacy Division (29 June 2017), Defence Expenditure of NATO Countries (2010-2017)[Press release]
To make a more equal comparison in line with the original study of the OZ model, in Europe, we notice that in 2017e only Greece, Estonia, the UK, Romania, and Poland reached the 2% NATO guideline. Yet, interestingly enough the UK is the only country out of these to have a GDP higher than the Netherlands. In fact, of the NATO countries in continental Europe with the largest GDP, Germany, France, Italy, Spain, Turkey, Netherlands, and Poland, again, only Poland meets the 2% criteria.
Therefore, from a precursory examination of the data, we can say that current NATO behaviour does not reflect the original hypothesis that members with higher GDP will meet their quota goals.
Hypothesis 5: Negative correlation between national income and infrastructure expenses
“H5: In the NATO alliance, there is a significant negative correlation between national income and the percentage of national income devoted to infrastructure expenses.”[viii]
The OZ model investigates the correlation between national income and the percentage spent by members on infrastructure, the definition of which is, according to NATO, “the static items of capital expenditure which are required to provide the material backing for operational plans necessary to enable the higher command to function and the various forces to operate with efficiency.”[ix] Some of the most relevant aspects included under this term are, amongst other things; the exercise of command such as permanent headquarters and permanent signal communications; the maintenance of forces and their equipment such as barracks, depots, warehouses, permanent hospitals, and permanent repair installers; and the permanent framework of a system of defence.
If the assumptions in the model were to be found correct, it would mean that in contrast to previously theorized positive correlations between national income and the percentage of this spent on NATO burdens, in the case of infrastructure, a negative correlation can be observed.[x] In this case, the higher the national income of a member state, the lower the percentage of this income is devoted specifically to infrastructure. The OZ model accounts for this by pointing out possible benefits that the smaller nations enjoy from taking up these costs.[xi] An example of these benefits would be the prestige a country gets from having a NATO base within their territory. As such, smaller nations end up devoting a higher percentage of their national income to infrastructure, especially when the benefits they gain are above average.[xii]
To test this hypothesis we used the available data from 2017. When assessing the distribution of defence expenditure, the category of infrastructure seems to confirm the model’s assumptions at first glance. Larger states devote a smaller percentage of their total defence expenditure towards infrastructure. Examples are the UK with 1.95%, France with 2.88%, and Germany with 3.48% (Table 2).
Table 2: Distribution of Defence Expenditure by Main Category: Infrastructure. NATO, Public Diplomacy Division, (29 June 2017), Defence Expenditure of NATO Countries (2010-2017)[Press release].
What is interesting to note, however, is that Poland spends a relatively high amount on infrastructure, 5.15% of their total defence expenditure. This is not so surprising given its geostrategic position. The smaller states that have a similarly important geostrategic position spend a significantly higher percentage on infrastructure. For example, Estonia devotes 11.25% and Latvia 13.56% of their total defence expenditure to infrastructure. The Netherlands is spending 3.34%, which is not that much compared to other smaller countries but still in line with the assumptions made by the model.
For hypothesis five we can therefore confirm the OZ model’s hypothesis that there is indeed a negative correlation between a member’s income and the percentage of national income devoted to infrastructure expenses. Although it is not possible to draw a definite conclusion as to what are the motivations of increased spending on infrastructure by a smaller member, the reason offered by Olsen and Zeckhauser seems plausible. Indeed, benefits such as the prestige obtained from having NATO bases are in the interests of smaller NATO countries.
Poland - The Netherlands: Nuance by numbers
Poland had a GDP of 475 billion USD in 2015 and 467 billion USD in 2016, which is significantly lower than that of the Netherlands, with 751 and 770 billion USD, respectively. If the OZ model would be applicable, it would suggest that the Netherlands would devote a larger part of their GDP to defence than Poland; but, the Netherlands has an expenditure of 1.17% of their GDP. In contrast to this, Poland does fulfil the 2% criteria, since it spends 2.01% of its GDP on defence (Figure 2). These trends go against the OZ model, which found a positive correlation between a country’s GDP and its spending behaviour on defence. Instead, they encourage us to look beyond GDP, go deeper into the contextual environment of the two countries, and break down each country’s expenditure to create more nuance.
Poland has a defence policy that is focused mainly on territorial defence, on the one hand, and its membership in NATO, on the other. Its proximity to Russia, which has displayed an increasingly assertive military behaviour, and Poland’s new government in 2015 have led to its New Technical Modernization Programme (TMP) for 2017–2026 that was launched in 2016.[xiii] This program is designed to increase capabilities, reform certain command structures, and strengthen Poland’s defence-industrial base.[xiv] Poland’s commitment to building a strong defensive force can also be found in the example of its announcement in 2016 in which it revealed plans to set up a “53.000 strong territorial-defence force organized in 17 light infantry brigades”.[xv]
In 2016, Poland had an active defence force with a total of 99,300 troops, of which the army constitutes 48,200, the navy 7,700, the air force 16,600, Special Forces 3,000, and joint forces 23,800.[xvi] The paramilitary constitutes another 73,400 troops, including Border Guards, the Maritime Border Guard and the Prevention Units (police), and the Anti-terrorist Operations Bureau.
As observed before, the Netherlands has a higher GDP but spends less proportionally on defence than Poland. It also has clearly more limited forces, with an active force of 35,410. The OZ model points out that smaller countries free ride on the larger ones by contributing less to nothing while reaping the benefits from the Alliance. At first glance one could perhaps observe this kind of pattern in Dutch behaviour towards NATO, with fewer resources committed to it and only 1.17% of GDP devoted to defence. However, this would be a conclusion too quickly made. The Netherlands does in fact make important contributions to both European Union military operations as well as to NATO itself. Examples of this are the air-policing agreement with Belgium and Luxembourg and the cooperation between the Dutch and German armies to establish a rapid reaction force.
Where in the past the Netherlands saw budget cuts for the military, these have halted, and it has been increasing spending since 2015.[xvii] Moreover, the Netherlands has confirmed it will acquire two tankers in cooperation with Luxembourg, which will be NATO property. Next to material contributions the Dutch army has also participated in multiple European Union-wide missions in Somalia and the Mediterranean in order to fight piracy and human trafficking. Its forces were also employed until the end of 2017 to provide training in the NATO mission in Iraq. The navy will soon start a mission to train Libyan coast guard personnel in light of the migrant crisis in the EU.[xviii]
The Netherlands has an active force of 35,410 troops, of which 18,860 belong to the army, 8,500 to the navy, 8,050 to the air force, and 5,900 to the military constabulary. The Netherlands also has a reserve force of 4,500.[xix]
Nuance and free riding revisited
As previously mentioned the OZ model attempts to empirically explain the fundamental tension that exists in NATO due to the cost of collective defence. Certainly, a key source of this tension arises from national legislatures, in charge of defence spending, not being accountable or beholden to the parameters set at the supranational level in NATO.[xx] We hope that our study has shown that the 2% benchmark is too imprecise to be able to calculate the true costs of burden sharing. In the recent NATO literature, this seems to also be a central critique. As a result, other models or metrics are being introduced or called upon to help delve into the nuance that, we believe, is required to adequately assess a member’s “free riding” status. These new methods extend from having an activity index of members, to an Alliance index, or just recalculating how defence expenditure is measured.[xxi] Scholars have recently begun to look at the relationship between the strategic culture of a nation and its reliability. This focus on reliability, in this case the willingness to participate in operations, opens up a new way of looking past this 2% benchmark. Member states are becoming increasingly quantifiably labelled as “Atlanticist” or “Europeanists” in their strategic culture, and this is leading to the reassessment of certain members’ values and priorities.[xxii] Atlanticist states tend to have high operations and maintenance budgets that are focused on mission readiness and deployment, whereas Europeanists are less focused on these tenets.[xxiii]
Presenting the numbers for Poland and the Netherlands through these alternative approaches, specifically the one advocated by Becker and Malesky and Frizzelle, the Netherlands has the second highest Atlanticism score following the UK. A look at participation in NATO operations under this system can allow for an alternative and more realistic approach to understanding member states’ behaviour in burden sharing.
In conclusion, the decades-long debate on burden sharing within NATO and which members are “free riding” is one that has resulted, in part, due to the lack of a metric that can account for nuances in member state behaviour. The OZ model was one of the first models to attempt to explain burden-sharing behaviour in the Alliance. The tests/comparisons we ran to try and replicate it had the following results:
For hypothesis 1, we concluded that there is a weak, however positive, correlation for the size of a nation and the percentage of their GDP being spent on defence. This does not mean, though, that the nation is meeting the 2% benchmark.
For hypothesis 2, we concluded that states with a higher GDP do not display a tendency to meet their quota level. Of the three hypotheses tested, this is the only one that we have come to fully reject.
For hypothesis 5, we concluded that we can indeed see a negative correlation between a member’s income and expenditures on infrastructure. We found that smaller states tend to spend more on infrastructure than larger members. One explanation given for this was that smaller states gain prestige from this action.
In terms of the nuances of members’ behaviour, when looking at Poland and the Netherlands in depth and especially in light of the alternative metrics to the OZ model, we hope that we have managed to at least shed doubt over the blanket statement that any member state not contributing 2% is a free rider. Although 2%, one could argue, is a useful and necessary benchmark, it does not provide enough information to make a strong argument in favour of the free-riding model that Olsen and Zeckhauser proposed. Discussions on burden sharing have been around since NATO’s existence, and different models have been developed to explain the phenomenon. By running one of the first models using recent data, we were able to find some trends still in line with the assumptions of the model; however, this also showed the need for more nuance. Newer models such as the Atlanticist versus Europeanist approach may offer a better explanation for the behaviour displayed by members in the Alliance. However, more research into these models would be needed, which does not fall under the scope of this paper.
Jan Rempala is a program trainee with the German Marshall Fund’s Europe and Mediterranean policy programs. He graduated with a master’s degree from Maastricht University’s European Public Affairs program and received his bachelor’s in Arabic and international relations at the Ohio State University. Jan is Polish-American and works on a variety of topics including cyber, Eastern Partnership and civil society, the MENA, and wider Atlantic issues. Jan speaks (rather rusty) Arabic and has lived and studied in both Jordan and Oman.
Tristan Roeven recently graduated with an master’s in European Public Affairs at Maastricht University. During his bachelor’s in European Studies he conducted an internship at NATO Joint Force Command, which strengthened his research interest in transatlantic relations. To complete his master’s degree, he completed an internship at the Dutch Foreign Ministry, where he was responsible for the coordination of Dutch diplomatic input into the working parties on European Neighborhood Policy. Tristan is aiming for a career in EU external policy and international relations.
[i] M. Olson and R. Zeckhauser, An Economic Theory of Alliances (Santa Monica, CA: The Rand Corporation, October 1966).
[ii] S. Weber and H. Wiesmeth, “Economic models of NATO,” Journal of Public Economics 46, no. 2 (1991): 190, doi:10.1016/0047-2727(91)90003-k190.
[iv] Olson and Zeckhauser, An Economic Theory of Alliances
[v] Ibid., 25.
[vi] NATO-Press Release, “Defence Expenditure of NATO Countries 2010–2017,” COMMUNIQUE PR/CP(2017)111, 29 June 2017, https://www.nato.int/nato_static_fl2014/assets/pdf/pdf_2017_06/20170629_....
[vii] Olson and Zeckhauser, An Economic Theory of Alliances.
[ix] NATO, Report By the Working Teams to the Standing Group on Definition of "Infrastructure" (Rep. No. S.G.68), 1950.
[x] Olson and Zeckhauser, An Economic Theory of Alliances, 32.
[xiii] International Institute for Strategic Studies (IISS), The Military Balance, 2017, 83.
[xvi] Ibid., 84.
[xvii] Ibid., 78.
[xx] B. Frizelle, “What Makes a Reliable Ally? A Fresh Perspective on NATO, Strategic Culture and Collective Defense,” War on the Rocks, 18 January 2018, accessed 30 January 2018, https://warontherocks.com/2018/01/makes-reliable-ally-fresh-perspective-....
[xxii] J. Becker and E. Malesky, “The Continent or the “Grand Large”? Strategic Culture and Operational Burden-Sharing in NATO,” [Abstract] International Studies Quarterly 61, no. 1 (2017): 163–180, doi:10.1093/isq/sqw039.
[xxiii] Frizelle, “What Makes a Reliable Ally?”