Introduction
Russia’s February 2014 invasion of Ukraine and its subsequent annexation of Crimea shocked NATO members and renewed fears across the alliance of the Russian threat. At the Wales Summit later that year, NATO members pledged to increase defence spending, aiming at 2 per cent of their gross domestic product (GDP). After 2014, many European security researchers focused on increased threat perceptions along NATO’s eastern flank, as NATO allies closer to Russia perceive greater risk from their Russian neighbour compared to more geographically disparate allies in Western and Central Europe.Footnote 1 Other work focused on evaluating adjustments in NATO defence spending and adherence to Wales pledge commitments.Footnote 2 While these studies find defence spending across NATO increased after 2014, they also find heterogeneous defence investment patterns, with threat geography not always determining spending outcomes.
While this important research examines NATO’s evolving defence spending and threat perceptions, post-2014 adjustments in NATO military capabilities remain underexplored. Did the 2014 Russian invasion lead to military capability adjustments across European NATO member states? In this article, military capabilities refers to observable force structure outputs – such as personnel and major weapons platforms – rather than qualitative differences in training, readiness, or technological sophistication. While these qualitative dimensions are critical components of combat effectiveness, force structure measures remain central indicators of states’ ability to generate and sustain military power and therefore provide a useful measure of capability adjustment over time.Footnote 3 Moreover, they are also comparable across countries, a challenge with qualitative dimensions. Better understanding NATO military capability adjustments has increasing policy relevance as the US military’s end strength (total troops), platforms (number of tanks, jets, etc.), and force structure (number of units) shrink to their lowest levels since 1940, and the United States applies increasing pressure on its allies to contribute more to their own defence.Footnote 4 Better understanding these dynamics is particularly pressing in understanding contemporary US–European tensions in the NATO alliance, as European NATO member states have long been accused of ‘free-riding’ on US military preponderance.Footnote 5
Russia’s 2014 aggression represented an intensified threat for European NATO members’ security, and a compelling vector to empirically test how NATO members adjusted their military capability portfolios in response to that shock. Our outcome variables focus on important capabilities needed to deter a Russian military incursion: total numbers of active military troops, active land forces troops, armoured vehicles, and combat aircraft.Footnote 6 Reviewing trends in European NATO members’ military capabilities between 2008 and 2022, we demonstrate that despite increased NATO military spending, NATO troop and platform averages continue pre-existing declines after 2014. However, these European-wide alliance averages omit varying threat proximity across NATO. To evaluate this more directly, we next employ a difference-in-differences research design to examine changes in capabilities across twenty-nine European NATO allies between 2008 and 2022 (N = 435). Two treatment specifications are used to evaluate how proximity to Russia impacts post-2014 NATO capabilities: (a) NATO allies bordering RussiaFootnote 7 and (b) Eastern European NATO allies.Footnote 8 We find that allies bordering Russia significantly increased military personnel following Russia’s 2014 invasion, compared to more geographically distant NATO allies, with a similar but less robust pattern for armoured vehicles.
This paper offers contributions to both research and policy focused on NATO, burden-sharing, and alliance management. While some studies analyse changes in NATO’s defence spending before and after 2014, they focus on defence investment inputs (total defence spending, defence equipment spending, defence spending as a percentage of GDP, etc.), while not providing insights into investment outputs (number of troops and military capabilities). Our focus on European NATO members’ military capabilities provides a better theoretical understanding of how military capabilities change over time, both in response to external threat and in the context of alliance burden-sharing. While the connection between threat proximity and increased military spending is well established, our analysis shows European NATO members did not uniformly increase their end strength (personnel) and platforms (equipment) after 2014, despite increased defence spending. This suggests that future research evaluating allied contributions to NATO’s collective defence commitment also needs to consider capabilities, rather than only focusing on investments. This article seeks to contribute to ongoing European debates about NATO’s defence planning process (NDPP), capability targets, and the credibility of collective deterrence after 2014.
We proceed as follows. First, we review existing research examining alliances, threat perceptions, and military capabilities, before introducing a testable hypothesis. We next review some important historical background covering Russia’s 2014 invasion of Ukraine, before introducing data and operationalising variables capturing military capability and threat. We then present the empirical research design, introduce difference-in-differences models, and review the models’ results. After analysing findings, we conclude with research and policy implications.
Alliances, threat perception, and military capabilities
Traditional balance of power theory contends that states seek security from external threats by ‘external balancing’ through alliances and ‘internal balancing’ through self-arming to strengthen their own military capabilities.Footnote 9 Walt’s balance of threat theory suggests that geographic proximity to neighbouring states demonstrating offensive intentions is an important aspect of interstate threat perception and balancing.Footnote 10 During periods of intense security competition, arms races may emerge between rivals, leading to significant increases in military acquisitions.Footnote 11 However, because state resources are limited and states must provide a host of services beyond external security, they seek to gain the most security at the lowest cost, avoiding over-investment in military capability.Footnote 12
These dynamics in defence alliances, which aggregate military capabilities across states, might create ‘free-riding’, where moral hazard leads more secure allies to under-invest in guns to satisfy domestic constituents by spending more on butter.Footnote 13 In the context of military capabilities, this implies that more secure allies would not just under-invest in their militaries generally but might also direct what funds are available into areas that promise domestic political spillovers.Footnote 14 As summarised by Sloan, ‘The burden-sharing/free-riding issue is a normal one for alliances among sovereign, democratic states. Leaders of democracies are inclined to try to obtain the highest level of security for their country at the lowest possible price.’Footnote 15
Linking concepts of balancing, burden-sharing, and guns–butter trade-offs is research on defence spending and the political economy of security.Footnote 16 Research on defence spending has highlighted how budgetary and investment constraints impact force planning decisions.Footnote 17 Studies on domestic drivers of defence spending suggest that a country’s macroeconomic conditions significantly influence its defence investments in peacetime, especially in democracies.Footnote 18 Recent work in defence economics further highlights the constraints on NATO defence investment. Conyon demonstrates that defence burdens remain unevenly distributed across allies despite repeated alliance burden-sharing agreements, suggesting that collective action challenges identified by Olson and Zeckhauser continue to characterise alliance behaviour.Footnote 19 Conyon further demonstrates that fiscal conditions and debt constraints influence defence spending choices, showing that budget limitations remain considerations for countries responding to external security threats.Footnote 20 Recent survey work has shown that the micro-foundations of such constraints may be mitigated by public awareness of threats.Footnote 21 Less appreciated is the connection between defence spending and military capabilities. This is important because different types of defence spending have different effects – whether on economic growth,Footnote 22 or on capabilities,Footnote 23 and because even within capability portfolios different types of spending may manifest differently. This necessitates a direct examination of capabilities themselves.
To measure the self-arming behaviour that is at the core of internal balancing, ‘military capability’ captures how a state equips its military for combat.Footnote 24 Betts analyses how decisions regarding military capabilities can often take years to implement, as states design forces and develop capability portfolios by assessing current and expected future threats.Footnote 25 Most existing work on military capabilities examines it as an independent variable. Several studies examine how mechanised force structures and armoured vehicles impact counter-insurgent effectiveness and civil conflict outcomes.Footnote 26 Other work highlights how modern military capabilities, underpinning contemporary combined arms warfare, are important foundations for military effectiveness on conventional battlefields.Footnote 27 Empirical work suggests that armoured and air defence capabilities are more effective at deterrence, compared to light infantry, airpower, and naval capabilities.Footnote 28
While these studies explain how capabilities shape security outcomes, only a few studies examine military capabilities as a dependent variable, and little is known about how capability portfolios evolve over time in response to external threats. Sechser and Saunders conduct large-N regression analysis, examining the determinants of military mechanisation and the proliferation of armour capabilities.Footnote 29 Examining 150 countries from 1979 to 2001, they find that threatening neighbours play a significant role in states’ increasing their military mechanisation. Margulies focuses on determinants of military recruitment, finding that states select conscription or all volunteer systems based on influences from great power patrons.Footnote 30 Outside of arms-racing literature, little is known about how military capabilities systematically evolve over time. Thus, while it is clear that military capabilities impact force employment and conflict outcomes, it is less clear how states design their militaries in response to external security threats. With this gap identified, we proceed by analysing how European NATO member states adjusted their military capabilities after the 2014 Russian invasion of Ukraine.
Case background and conceptual framework
Having reviewed research examining alliances, threat perceptions, and military capabilities, we now introduce the case underpinning the subsequent quantitative analysis and detail a case-specific hypothesis motivated by the literature review. On 27 February 2014, Russia invaded Ukraine, secured Crimea, and illegally annexed the peninsula on 21 March.Footnote 31 Vladimir Putin referenced Ukraine’s increasing ties with the European Union and NATO to justify Russia’s intervention in his joint address to the Russian legislature shortly following the invasion. Putin had long viewed NATO with mistrust,Footnote 32 and Russian officials characterised the alliance’s post–Cold War enlargement into Central and Eastern Europe as a threat to Russian security.Footnote 33
Despite these Russia–NATO tensions, evidence suggests the 2014 Russian invasion and annexation came as a surprise to NATO.Footnote 34 Increasing Russian–European trade, European energy dependence on Russian oil, robust diplomatic communication, and intermittent NATO–Russia cooperation before 2014 all highlighted connections that obscured Russia’s forthcoming aggression.Footnote 35 Consistent NATO defence spending and military capability cuts before 2014 aligned with these trends and perceptions.Footnote 36 Similarly, the US Army deactivated its last permanent armoured units in Europe and shipped its tanks back to the United States in 2013, only to begin rotating armour units to Europe again after the 2014 Russian invasion.Footnote 37
Given the renewed threat from Russia, at NATO’s 2014 Wales Summit, all alliance heads of state and government agreed to ‘move toward’ spending towards 2 per cent of their gross domestic product (GDP) on defence, with 20 per cent of that spending allocated to equipment modernisation.Footnote 38 Following this 2014 Wales pledge, subsequent academic and policy research has carefully scrutinised defence spending trends throughout the alliance, generally finding defence spending increases across NATO but higher defence investments from Eastern European members, closest to the Russian threat.Footnote 39 Aligned with this work and tying back to Walt’s balance of threat theory, other research has demonstrated that varied threat perception across the NATO alliance can be linked to geographic proximity to Russia.Footnote 40 However, while these studies provide important insights into NATO defence investment (inputs), they do not examine if and how these investment changes lead to increased military capability (outputs).
NDPP is designed to convert this kind of investment into capabilities in a coordinated fashion among allies – ‘to provide a framework within which national and Alliance defence planning activities can be harmonised to enable Allies to provide the required forces and capabilities in the most effective way’.Footnote 41 The Wales pledge explicitly referred to NATO capabilities, as well as ‘NATO guidelines for deployability and sustainability and other agreed output metrics’ included in the NDPP.Footnote 42 Defence spending goals were linked to military–strategic reforms like the Readiness Action Plan and updates to the NATO Response Force. These capabilities-focused reforms were designed to ‘convert’ the cash allies pledged at Wales into capabilities through the NDPP.Footnote 43 While national capability benchmarks were not publicly released across the alliance, growing concerns in NATO over Russian aggression suggest that post-Wales defence spending increases should lead to increases in NATO military capabilities.
It is not, however, clear how efficiently increased defence spending yielded increased capabilities. While it may seem intuitive that higher NATO defence spending results in additional capabilities, this is not necessarily the case. First, military personnel and acquisition costs have increased over time, making it more expensive to maintain or acquire more capabilities.Footnote 44 Second, defence spending has multiple subcomponents, broadly ranging across investments in research and development, infrastructure, operations and maintenance, personnel, and procurement – which most directly impacts capabilities.Footnote 45 Further, member state NDPP capability outcomes are influenced by domestic political and economic factors, which might impede strict compliance with NDPP targets.
Despite these trends, strategic realities after 2014 suggest that NATO members had compelling reasons to increase their military capabilities. In the years following the 2014 invasion and Wales Summit, concerns mounted over an imbalance of forces in the Baltic Sea region, favouring preponderant Russian forces in its Western Military District and Kaliningrad.Footnote 46 Growing alarm over this NATO capability gap led to forecasts of Russian troops seizing the Baltic capitals in a few days.Footnote 47 At the same time, as multipolarity and strategic competition increased, the United States sought to ‘pivot to the Pacific’, and its limited military forces were being spread increasingly thin.Footnote 48 Because capability is understood as a critical aspect of conventional deterrence,Footnote 49 analysing NATO capability changes over time in response to the renewed Russian threat can inform our understanding of alliance burden-sharing and threat perceptions.
With these case details in mind, we proceed by analysing the 2014 Russian invasion and annexation as a shock to NATO security. Because concerns over Russian aggression initially started after Moscow’s 2008 invasion of Georgia, this was not a purely exogenous shock but rather an intensification of existing security concerns. Despite those pre-existing security trends, the 2014 Russian invasion clearly increased threat perception across the NATO alliance, offering an important window to analyse if and how NATO military capabilities adapted to that shock. This approach aligns with a difference-in-differences causal identification strategy, using the 2014 Russian invasion to define the pre/post-treatment period, and geographic proximity to Russia to define the treatment/control groups. Relating to this case, a testable hypothesis emerges:
H1: Following Russia’s 2014 invasion of Ukraine, NATO members that are closest to Russia, and most vulnerable to Russian territorial aggression, will respond by increasing their military capabilities more than geographically distant NATO allies in Central and Western Europe, even when controlling for additional defence spending on personnel and equipment.
Operationalising data
Having reviewed the case background and detailed a testable hypothesis, we next introduce data and operationalise variables with which we can empirically test this hypothesis. We organise these data in a country–year panel from 2008 to 2022, consisting of the twenty-nine European NATO member states. Non-European NATO members (the United States, Canada, and Iceland) are excluded, given their large geographic distance from Russia. The year 2015 is coded as the first year for the ‘post-treatment period’ following the February 2014 invasion.Footnote 50 The 2008–2022 panel with treatment occurring in 2015 allows for seven years before treatment and eight years after. Though North Macedonia, Montenegro, Finland, and Sweden were not NATO allies at the beginning of the panel, each country joined NATO during the covered period, or immediately following. Given their foreign policy alignment with NATO and perceived threat from Russia, they are fully included in the panel.Footnote 51 This produces 435 country–year observations total, across twenty-nine countries and fifteen years.
Dependent variables
Four dependent variables (DVs) measuring military capabilities are used: the military-wide total number of troops (across all services), the total number of troops in each military’s land forces, the total number of armoured vehicles, and the total number of combat aircraft. Total military and land forces troops are derived from the International Institute for Strategic Studies annual Military Balance report, from 2008 to 2022.Footnote 52 Armoured vehicles (consisting of tanks and infantry fighting vehicles) and combat aircraft are compiled from Souva’s Material Military Power (MMP) dataset.Footnote 53 Given research suggesting large, armoured land forces are important for attaining conventional deterrence, these capabilities provide a reasonable sample of military capabilities needed to deter Russian aggression.Footnote 54
Figures 1, 2, and 3 graphically depict these data across several categories between 2008 and 2022. First, Figure 1 captures NATO-wide averages across the alliance’s European member states. While average defence spending across NATO climbs following the Wales pledge, average capabilities (total troops in military, armoured vehicles, and combat aircraft) steadily decline.Footnote 55 However, Figure 1’s averages obscure varying threat perceptions in the alliance.
European NATO member averages, 2008–2022.
Figure 1 Long description
European NATO member averages from 2008 to 2022 across four categories: defense spending, active military troops, armoured vehicles and combat aircraft. 1. Defense Spending: The x-axis represents the year from 2008 to 2022 and the y-axis shows spending in real USD from 1,200,000 to 1,800,000. The data points indicate a general increase, particularly after 2014. 2. Active Military Troops: The x-axis shows the year from 2008 to 2022 and the y-axis represents the number of troops from 70,000 to 100,000. The graph shows a decline over the years. 3. Armoured Vehicles: The x-axis covers the year from 2008 to 2022 and the y-axis shows the number of vehicles from 1,400 to 2,000. The data points indicate a decrease, with a slight recovery starting around 2020. 4. Combat Aircraft: The x-axis represents the year from 2008 to 2022 and the y-axis shows the number of aircraft from 70 to 110. The graph shows a downward trend over the period.
Select NATO member troops (military-wide and land forces), 2008–2022.
Figure 2 Long description
Active troop numbers for selected European NATO members from 2008 to 2022 are displayed in graphs. Each graph represents a different country: United Kingdom, Germany, Czechia and Lithuania. The x-axis is labeled ′Year′ ranging from 2008 to 2022 and the y-axis is labeled ′Troops′ with varying scales for each country. Two series are plotted: Military (triangles) and Army (circles). United Kingdom: The Military starts at approximately 180,000 troops in 2008 and decreases to around 100,000 by 2022. The Army remains stable around 30,000 troops. Germany: The Military begins at about 250,000 troops in 2008, declining to approximately 180,000 by 2022. The Army begins at 150,000 troops and declines to approxomate;y 100,000 troops. Czechia: The Military starts at about 30,000 troops in 2008, slightly decreasing to around 25,000 by 2022. The Army remains stable at approximately 10,000 troops. Lithuania: The Military shows an increase from about 15,000 troops in 2008 to approximately 20,000 by 2022. . Each graph includes a legend indicating the symbols for Army and Military.Because Lithuania’s military is largely composed of its Army, the army reflects similar numbers.
Select NATO member armoured vehicles, 2008–2022.
Figure 3 Long description
Line graph titled, United Kingdom. The horizontal axis label is, Year, with values 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022. The vertical axis label is, Armoured Vehicles, with values 3200, 3300, 3400, 3500, 3600, 3700. Plotted points by year: 2008 about 3650; 2009 about 3680; 2010 about 3660; 2011 about 3640; 2012 about 3620; 2013 about 3600; 2014 about 3570; 2015 about 3540; 2016 about 3500; 2017 about 3460; 2018 about 3420; 2019 about 3380; 2020 about 3340; 2021 about 3300; 2022 about 3250. Line graph titled, Germany. The horizontal axis label is, Year, with values 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022. The vertical axis label is, Armoured Vehicles, with values 2000, 3000, 4000, 5000, 6000, 7000. Plotted points by year: 2008 about 6800; 2009 about 6500; 2010 about 6100; 2011 about 5600; 2012 about 5000; 2013 about 4300; 2014 about 3500; 2015 about 2800; 2016 about 2700; 2017 about 2700; 2018 about 2700; 2019 about 2700; 2020 about 2700; 2021 about 2700; 2022 about 2700. Line graph titled, Czechia. The horizontal axis label is, Year, with values 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022. The vertical axis label is, Armoured Vehicles, with values 200, 400, 600, 800, 1000. Plotted points by year: 2008 about 900; 2009 about 850; 2010 about 800; 2011 about 700; 2012 about 600; 2013 about 500; 2014 about 350; 2015 about 250; 2016 about 260; 2017 about 280; 2018 about 300; 2019 about 320; 2020 about 340; 2021 about 360; 2022 about 400. Line graph titled, Lithuania. The horizontal axis label is, Year, with values 2008, 2010, 2012, 2014, 2016, 2018, 2020, 2022. The vertical axis label is, Armoured Vehicles, with values 100, 150, 200, 250, 300, 350. Plotted points by year: 2008 about 160; 2009 about 170; 2010 about 180; 2011 about 175; 2012 about 165; 2013 about 150; 2014 about 130; 2015 about 120; 2016 about 140; 2017 about 170; 2018 about 200; 2019 about 230; 2020 about 260; 2021 about 290; 2022 about 320.
Figures 2 and 3 provide some initial insights into how geography and varying threat perceptions impact military capability. These figures show total military troops and armoured vehicles for four NATO members in the sample: United Kingdom (Western Europe), Germany (Central Europe), Czechia (Eastern Europe), and Lithuania (bordering Russia). These time series are consistent with H1’s predictions. The British and German graphs show persistent declines in all categories despite Russia’s 2014 invasion. In contrast, Czechia and Lithuania experience increases, although Lithuania’s increase is much sharper, perhaps reflecting their increased threat vulnerability.
Data limitations
Despite these data’s advantages, there are important limitations that require discussion. First, these military capabilities data do not capture military readiness. Possessing many troops and armoured vehicles does not imply that those platforms are properly maintained, their crews trained, or their units supplied.Footnote 56 NATO standardisation agreements, which enforce alliance-wide standards for capabilities and readiness, should partially overcome this shortcoming by NATO-wide minimum levels of readiness.Footnote 57 Second, these coarse count data obscure qualitative differences in the equipment. For example, an outdated Albanian T62 tank built in the 1960s receives the same weight in the data as a German Leopard tank built in 2024. Importantly, however, post-2014 military modernisation efforts were disproportionately concentrated among NATO’s frontline states. To the extent that states bordering Russia simultaneously improved both the quantity and quality of their military platforms, simple platform counts likely understate the substantive magnitude of capability improvements. Future research should therefore seek to incorporate quality-weighted capability measures as such data become more widely available. Third, these IISS data rely on national reporting and expert assessments, which may introduce measurement error stemming from different reporting practices, reserve force classification, or equipment storage status.Footnote 58
Despite these shortcomings, these measures remain theoretically appropriate for our research question because they capture observable force structure outputs – the personnel and major weapons platforms that states can mobilise for deterrence and warfighting. Our objective is not to measure overall combat effectiveness, which also depends on readiness, training, doctrine, sustainment, and technological sophistication,Footnote 59 but rather to examine whether NATO members altered the size and composition of their military capability portfolios following Russia’s 2014 invasion. Moreover, because any measurement error in platform quality, readiness, or reporting practices is likely to introduce noise into the dependent variables, these limitations should bias estimates towards finding no relationship rather than generating spurious positive treatment effects. Consequently, significant findings should be interpreted as conservative evidence of post-2014 capability adjustment among NATO’s frontline states.
Treatment and control variables
To test threat adjacency’s relationship to NATO military capabilities, we employ two treatment specifications. First, and following Conyon’s difference-in-differences (D-in-D) strategy to measure changes in NATO defence spending following the 2022 Russian invasion of Ukraine, we code a dummy for NATO states sharing a border with Russia, impacting six of the twenty-nine NATO states in these data.Footnote 60 Second, we include an alternate threat specification, capturing fourteen ‘Eastern European’ NATO members, as defined by the United Nations’ regional groups of member states.Footnote 61 We build an interaction term by multiplying these two treatment dummies with a term for post-invasion years (2015 and beyond). The resulting variables (Border Threat and Eastern Threat) are the primary independent variables measuring threat proximity, the effect of which on military capabilities is our primary outcome of interest.
Because more populous and wealthier countries might be expected to have larger militaries, we control for population and real GDP, both derived from the World Bank.Footnote 62 To control for defence spending, we include a country’s defence spending on personnel for regressions measuring troops in the military and army (land forces), and a country’s defence spending on equipment for regressions measuring armoured vehicles and combat aircraft.Footnote 63 Finally, we control for forward US military presence, to measure possible ‘free-riding’ dynamics within NATO. Using the US TroopData package, the number of US troops permanently based abroad in each country is included.Footnote 64 If ‘free-riding’ critics are correct, then the increasing presence of permanently assigned US forces should be correlated with lower military capabilities in the hosting European states.Footnote 65 Table 1 provides a summary of descriptive statistics for dependent, independent, and control variables.
Summary statistics of selected variables.
Table 1 Long description
Summary statistics are reported by variable using columns Count, Mean, Std. Dev., Min., and Max. Troops in Military has 435 observations with mean 71,398.5, standard deviation 100,532.2, minimum 410, and maximum 510,600. Troops in Army has 435 observations with mean 43,236.7, standard deviation 69,635.3, minimum 410, and maximum 402,000. Armoured Vehicles has 435 observations with mean 1,499.9, standard deviation 1,980.8, minimum 0, and maximum 10,368, and Combat Aircraft has 435 observations with mean 80.0, standard deviation 95.4, minimum 0, and maximum 465. Permanent US Troops has 435 observations with mean 2,266.9, standard deviation 7,247.8, minimum 0, and maximum 45,649. Def. Spending real USD has 435 observations with mean 1,130,655.2, standard deviation 1,840,861.9, minimum 6,321.9, and maximum 9,469,336. Def. Equipment Spending real USD has 426 observations with mean 243,130.0, standard deviation 439,566.6, minimum 94.9, and maximum 2,974,835.2, and Def. Personnel Spending real USD has 426 observations with mean 563,428.1, standard deviation 828,772.4, minimum 4,933.1, and maximum 3,726,312.2.
Table 2 presents average values for the dependent, independent, and control variables before and after Russia’s 2014 invasion of Ukraine across active NATO member states in the panel. Simple t-tests comparing pre- and post-2014 means yield null results for each variable when the alliance is analysed as a single, undifferentiated group. Notably, Table 2 shows that while average NATO defence spending and equipment investment increased following 2014, average military platform counts and personnel levels continued to decline. This apparent disconnect between inputs and outputs is consistent with the descriptive trends shown in Figures 1, 2, and 3. However, because Table 2 does not distinguish between NATO members facing differing levels of exposure to Russian military power, it reflects a naive comparison that obscures potentially meaningful variation in threat exposure across countries. To account for this heterogeneity and formally evaluate H1, we next introduce a D-in-D research design that is conditioned on geographic proximity to Russia and exploits the 2014 invasion as a shock that intensified fears over Russian aggression across the NATO alliance.
Examining data pre- and post-2014 invasion.
Table 2 Long description
Rows list measures with columns for Pre 2014, Post 2014, Difference, Std. Error, and N. Troops in Military are 78,202.98 pre 2014 and 65,444.53 post 2014, difference minus 12,758.45, standard error 9,653.50, N 435. Troops in Army are 47,429.44 pre and 39,567.97 post, difference minus 7,861.46, standard error 6,689.47, N 435. Armoured Vehicles are 1,570.97 pre and 1,437.74 post, difference minus 133.23, standard error 190.48, N 435; Combat Aircraft are 85.60 pre and 75.22 post, difference minus 10.37, standard error 9.17, N 435; Permanent US Troops are 2,306.57 pre and 2,232.11 post, difference minus 74.46, standard error 697.35, N 435. Population is 19,827,504.76 pre and 20,306,674.35 post, difference 479,169.59, standard error 2,435,572.11, N 435. Real GDP, World Bank is 561,000,000,000 pre and 626,000,000,000 post, difference 64,100,000,000, standard error 86,800,000,000, N 435. Def. Spending real USD is 1,033,742.49 pre and 1,215,453.91 post, difference 181,711.42, standard error 176,907.52, N 435; Def. Equipment Spending real USD is 207,067.82 pre and 273,285.51 post, difference 66,217.69, standard error 42,694.08, N 426; Def. Personnel Spending real USD is 564,247.66 pre and 562,742.70 post, difference minus 1,504.96, standard error 80,724.73, N 426.
Empirical framework
Having operationalised and defined key variables needed to test H1, we now introduce the paper’s quantitative research design. Prior research demonstrates that a wide range of political, economic, and strategic factors shape states’ military capability development.Footnote 66 However, these earlier global cross-national analyses are not able to isolate intra-alliance dynamics and geographically contingent threat responses. To address these challenges, we employ a D-in-D research design that exploits the 2014 Russian invasion of Ukraine as a major intensification of existing European security concerns and compares changes in military capabilities between NATO members facing higher and lower levels of threat exposure. This two-group, pre/post design helps account for unobserved time-invariant differences across countries and common temporal shocks, thereby mitigating concerns about omitted variable bias that arise in standard cross-sectional OLS regression models.Footnote 67 Following Baker et al., the D-in-D estimator identifies the causal effect of treatment as ‘the change in outcomes for the treated group minus the change in outcomes for the untreated group’.Footnote 68 In this application, the estimator captures differential post-2014 changes in military capabilities among NATO members most exposed to Russian military power relative to more geographically distant allies. This empirical specification is formalised in Equation (1).
\begin{equation*}{Y_{it}} = {\beta_0}Trea{t_i}+{\beta_1}Pos{t_t}+{\beta_2}\zeta +{\beta_3}Trea{t_i}\times Pos{t_t}+{\kappa_i}+{\lambda_t}+{\epsilon_{it}}\end{equation*} Here, 
${Y_{it}}$ represents several DV specifications for 
$Stat{e_i}$ in 
$Yea{r_t}$, capturing different aspects of a NATO member’s military capabilities, which include: total troops in the military, total troops in the army, total armoured vehicles, and total combat aircraft. 
$Trea{t_i}$ focuses on H1, how geographic proximity influences threat perception, and captures NATO members located closest to the Russian threat. 
$Pos{t_t}$ represents the shock of the February 2014 Russian invasion, with a dummy for all years after 2014. D-in-D models further allow for the inclusion of control variables, and the 
${\beta_2}$ coefficient captures the coefficients for the aforementioned controls. 
$Trea{t_i}$ x 
$Pos{t_t}$ is an interaction term of the previous two dummy variables, and the 
${\beta_3}$ coefficient captures the D-in-D estimator for the average treatment effect. 
${\kappa_i}$ represents country fixed effects for country i while 
${\lambda_t}$ represents year fixed effects for year t. These fixed effects account for unobserved, country-level trends within the twenty-nine NATO members in the data.
D-in-D models require the parallel trends assumption for internal validity, meaning that ‘trends would be the same in both states in the absence of treatment’.Footnote 69 This assumption does not require identical outcome levels across groups prior to treatment, but it does require that pre-treatment trends be comparable. This is a strong assumption: Sechser and Saunders show how domestic factors impact the acquisition of mechanised capabilities in their large-N analysis.Reference Sechser and Saunders2010 Country and year fixed effects partially mitigate concerns about omitted time-invariant domestic factors that influence military capability development, such as long-term force design choices. To more directly assess the plausibility of the parallel trends assumptions, and following Angrist and Pischke, we include the parallel trends test and Granger F-test in subsequent D-in-D models, which we analyse in the findings below.Footnote 70
Difference-in-differences results
The D-in-D results are depicted below, with Table 3 providing baseline models without controls and Table 4 providing fully specified models with all controls. Both tables use the Border Threat treatment specification, while Table 5 uses the alternative East Threat specification.
Baseline models: NATO military capabilities and Russian border treatment.
Table 3 Long description
Four regression models report coefficients for Border Threat and the Constant for Military Troops, Army Troops, Arm. Vehicles, and Combat Aircraft, with standard errors shown in parentheses. For Border Threat, the coefficients are 18,440.986 with standard error 6,310.77 for Military Troops, 13,201.642 with 5,352.09 for Army Troops, 264.474 with 143.00 for Arm. Vehicles, and 11.644 with 5.28 for Combat Aircraft. The Constant coefficients are 83,475.172 with standard error 4,000.80 for Military Troops, 50,823.724 with 2,407.26 for Army Troops, 1,707.793 with 111.25 for Arm. Vehicles, and 96.241 with 5.33 for Combat Aircraft. Parallel Trend F Test values are 2.03, 0.30, 4.46, and 2.93 across models 1 through 4, with Prob greater than F values 0.1654, 0.5867, 0.0438, and 0.0980. Granger Test values are 1.33, 1.12, 2.67, and 3.19 across models 1 through 4, with Prob greater than F values 0.2764, 0.3540, 0.0355, and 0.0160.
Treatment group is six NATO members bordering Russia; control is twenty-three other NATO states.
Values in parentheses are robust standard errors clustered on country.
All models include 435 observations and country and year fixed effects.
* p < 0.10,
** p < 0.05,
*** p < 0.010
Fully specified models: NATO military capabilities and Russian border treatment.
Table 4 Long description
Models 5 through 8 report coefficients with robust standard errors in parentheses for four outcomes: Military Troops, Army Troops, Arm. Vehicles, and Combat Aircraft. Border Threat is 8,997.087 with standard error 3,712.07 for Military Troops, 5,947.154 with standard error 2,074.76 for Army Troops, 277.813 with standard error 145.35 for Arm. Vehicles, and 6.590 with standard error 3.98 for Combat Aircraft. Permanent US Troops is 2.994 with standard error 2.44 for Military Troops, 4.825 with standard error 1.45 for Army Troops, 0.174 with standard error 0.04 for Arm. Vehicles, and 0.003 with standard error 0.00 for Combat Aircraft. Real GDP is 0.000 with standard error 0.00 in Military Troops and Army Troops, and minus 0.000 with standard error 0.00 in Arm. Vehicles and Combat Aircraft. Population is minus 0.014 with standard error 0.00 for Military Troops, minus 0.011 with standard error 0.00 for Army Troops, 0.000 with standard error 0.00 for Arm. Vehicles, and minus 0.000 with standard error 0.00 for Combat Aircraft. Def. Personnel Spending appears only in Military Troops at 0.013 with standard error 0.01 and in Army Troops at minus 0.007 with standard error 0.01, while Def. Equipment Spending appears only in Arm. Vehicles at 0.000 with standard error 0.00 and in Combat Aircraft at minus 0.000 with standard error 0.00. Constants are 347,401.699 with standard error 32,226.71 for Military Troops, 257,220.161 with standard error 35,464.65 for Army Troops, minus 692.736 with standard error 648.48 for Arm. Vehicles, and 222.562 with standard error 55.55 for Combat Aircraft. Parallel Trend F Test values are 0.33, 0.19, 5.90, and 3.36 with Prob greater than F values 0.5710, 0.6695, 0.0218, and 0.0774 for models 5 through 8, respectively; Granger Test values are 1.57, 0.55, 3.41, and 3.96 with Prob greater than F values 0.1937, 0.7802, 0.0118, and 0.0055.
Treatment group is six NATO members bordering Russia; control is twenty-three other NATO states.
Values in parentheses are robust standard errors clustered on country.
All models include 426 observations and country and year fixed effects.
* p < 0.10,
** p < 0.05,
*** p < 0.010
Fully specified models: NATO military capabilities and Eastern European treatment.
Table 5 Long description
Models 9 to 12 report coefficients with robust standard errors in parentheses for four outcomes: Military Troops, Army Troops, Arm. Vehicles, and Combat Aircraft. East Threat is 5,954.872 with standard error 5,711.29 for Military Troops, 1,171.413 with 3,167.46 for Army Troops, 97.820 with 160.90 for Arm. Vehicles, and minus 7.596 with 4.76 for Combat Aircraft. Permanent US Troops is 3.108 with 2.34 for Military Troops, 4.859 with 1.46 for Army Troops, 0.176 with 0.04 for Arm. Vehicles, and 0.002 with 0.00 for Combat Aircraft, with significance markers shown on the Army Troops and Arm. Vehicles coefficients. Real GDP is 0.000 with 0.00 in all four models, with negative signed zeros shown for Arm. Vehicles and Combat Aircraft. Population is minus 0.014 with 0.00 for Military Troops, minus 0.011 with 0.00 for Army Troops, 0.000 with 0.00 for Arm. Vehicles, and minus 0.000 with 0.00 for Combat Aircraft, each with significance markers shown. Def. Personnel Spending appears only for Military Troops at 0.011 with 0.01 and for Army Troops at minus 0.007 with 0.00, while Def. Equipment Spending appears only for Arm. Vehicles at 0.000 with 0.00 and for Combat Aircraft at minus 0.000 with 0.00. Constants are 341,346.475 with 32,363.21 for Military Troops, 259,555.102 with 36,247.69 for Army Troops, minus 667.262 with 565.21 for Arm. Vehicles, and 243.914 with 51.88 for Combat Aircraft, with significance markers shown on all except Arm. Vehicles. Diagnostic rows list Parallel Trend F Test values of 0.23, 0.61, 0.54, and 0.04 with corresponding Prob greater than F values 0.6346, 0.4415, 0.4698, and 0.8492, and Granger Test values 1.94, 0.50, 0.63, and 1.81 with corresponding Prob greater than F values 0.1089, 0.8051, 0.7035, and 0.1336.
Treatment group is fourteen Eastern European NATO members; control is fifteen other NATO states.
Values in parentheses are robust standard errors clustered on country.
All models include 426 observations and country and year fixed effects.
*p < 0.10,
** p < 0.05,
*** p < 0.010
Results from Table 3 show that the six NATO states bordering Russia – Lithuania, Latvia, Estonia, Poland, Norway, and Finland – are associated with increased post-2014 military capabilities across all four DVs, compared to NATO allies not bordering Russia. Because this treatment specification relies on six frontline NATO states, the estimates should be interpreted as evidence regarding NATO’s most exposed members rather than the alliance as a whole. Coefficients for the Border Threat interaction term, which captures the D-in-D estimator, are all positive and significant, consistent with H1’s expectations.
Table 4 includes fully specified models that include controls for population, economic size, defence spending composition, and permanent US military presence. The Border Threat coefficients’ magnitude here is reduced relative to the baseline models and only the troops and armoured vehicles estimates remain positive and statistically significant. The combat aircraft model’s treatment effect is no longer significant. This outcome rests on a particularly thin empirical base: Estonia, Latvia, and Lithuania report zero combat aircraft in every year of the panel, so half of the border treatment group contributes no within-country variation.Footnote 71 Notably, controls for defence spending on personnel and equipment are not statistically significant. While this null result does not establish that defence spending has no effect on military capabilities, it suggests the observed capability increases among frontline NATO members are not fully explained by higher defence expenditures alone, and that geographic exposure to Russian military power plays an independent role in driving these adjustments.
Results from Table 5 apply the East Threat treatment specification, expanding the treatment population to the fourteen Eastern European NATO members. Under this broader definition of threat exposure, none of the four coefficients reaches conventional significance, and the combat aircraft estimate is signed negative. This diluted treatment and loss of significance offer sharp contrast to the Border Threat models and is consistent with the expectation that more geographically proximate allies exhibit clearer capability increases than those facing more indirect exposure to Russian military power. In Tables 4 and 5, four of the eight controls for permanent US troops produce positive and significant coefficients, suggesting that US permanent forward presence is not systematically associated with reductions in host-nation military capabilities. Overall, East Threat’s reduced geographic proximity to Russia (compared to Border Threat) and its coefficients’ reduced statistical significance suggest support for Meijer and Brooks, who note heterogeneous threat perceptions across NATO.Reference Meijer and Brooks2021
Taken together, the main results and associated robustness checks provide support for H1. At the same time, assessing the D-in-D design’s internal validity requires careful consideration of the parallel trends assumption. The Granger causality test provides one tool to assess this assumption.Footnote 72 Applied to these data, the Granger test’s null hypothesis is that European NATO members are not altering their military capabilities in anticipation of treatment (Russia’s 2014 invasion). As an additional diagnostic to assess internal validity, we report parallel trends F-tests for all models. These entail a Wald test to evaluate whether pre-treatment outcome trends differ systematically across treatment and control groups.Footnote 73 Both tests are reported for every model in Tables 3, 4, and 5.
In all three tables, the troop outcomes satisfy the Granger and parallel trends F-tests. They consistently fail to attain standard statistical significance and thus fail to reject the null hypothesis of parallel pre-treatment trends. These results increase confidence that pre-treatment dynamics are consistent with D-in-D identification assumptions and lend support to a causal interpretation of the post-2014 increases in troop levels among NATO members most exposed to Russian military power. By contrast, the equipment outcomes do not satisfy these diagnostics in Tables 3 and 4. In these tables, both armoured vehicles and combat aircraft reject the null of common pre-treatment trends, and both reject the Granger null of no anticipatory adjustment. For armoured vehicles, the parallel trends statistic is significant at conventional levels (F = 4.46 and 5.90); for combat aircraft it is significant at the 10 per cent level (F = 2.93 and 3.36). These findings warrant caution in interpreting the corresponding D-in-D estimates as strictly causal for both equipment outcomes.
Figure 4 presents visual diagnostics for the parallel trends assumption in Table 4’s model. Each graph displays observed means for both treatment and control groups from 2008 to 2022. Because the parallel trends assumption requires similar pre-treatment trends rather than identical levels, it is important to assess the relative slopes of the two series prior to 2014 rather than their absolute values, which differ substantially given the smaller sizes of the six treatment states. For the troop-based outcomes (top graphs), pre-treatment trends are broadly parallel. Both groups exhibit declining troop levels before 2014, consistent with the null results of the parallel trends F-tests reported in Tables 3 and 4. The equipment panels (bottom graphs) reveal more visible pre-treatment divergence, particularly for armoured vehicles, where the control group declines steeply from 2008 while the treatment group is flat or gently rising. Substantively, the presence of pre-treatment divergence is consistent with heightened threat perceptions among frontline NATO members prior to 2014, potentially reflecting earlier developments such as Russia’s post-2008 military modernisation and force posture changes following its invasion of Georgia.Footnote 74 Across all four outcomes, the post-2014 period shows the treatment group stabilising or increasing relative to continued declines in the control group, offering support for the positive Border Threat estimates in Tables 3 and 4.
Observed means for all four DVs from Table 4; 2008–2022.
Figure 4 Long description
Graphical diagnostics for parallel trends. The image A showing a line graph titled, Observed means. The horizontal axis label is, Year. The horizontal axis shows years from 2008 to 2022. The vertical axis label is, Troops in Military. The vertical axis range is 20000 to 120000. A vertical dashed line is drawn at 2014. Legend labels: Control, Treatment. Control series coordinate pairs: (2008, 120000), (2009, 110000), (2010, 95000), (2011, 90000), (2012, 85000), (2013, 82000), (2014, 80000), (2015, 78000), (2016, 76000), (2017, 74000), (2018, 73000), (2019, 72000), (2020, 72000), (2021, 72000), (2022, 72000). Treatment series coordinate pairs: (2008, 33000), (2009, 32000), (2010, 30000), (2011, 29000), (2012, 28000), (2013, 28000), (2014, 28000), (2015, 29000), (2016, 29000), (2017, 30000), (2018, 31000), (2019, 32000), (2020, 33000), (2021, 33000), (2022, 32000). The image B showing a line graph titled, Observed means. The horizontal axis label is, Year. The horizontal axis shows years from 2008 to 2022. The vertical axis label is, Troops in Army. The vertical axis range is 20000 to 70000. A vertical dashed line is drawn at 2014. Legend labels: Control, Treatment. Control series coordinate pairs: (2008, 70000), (2009, 65000), (2010, 60000), (2011, 55000), (2012, 52000), (2013, 50000), (2014, 48000), (2015, 52000), (2016, 46000), (2017, 44000), (2018, 43000), (2019, 43000), (2020, 43000), (2021, 43000), (2022, 43000). Treatment series coordinate pairs: (2008, 25000), (2009, 22000), (2010, 18000), (2011, 16000), (2012, 16000), (2013, 16000), (2014, 17000), (2015, 17000), (2016, 17000), (2017, 18000), (2018, 20000), (2019, 22000), (2020, 25000), (2021, 23000), (2022, 22000). The image C showing a line graph titled, Observed means. The horizontal axis label is, Year. The horizontal axis shows years from 2008 to 2022. The vertical axis label is, Armoured Vehicles. The vertical axis range is 500 to 2500. A vertical dashed line is drawn at 2014. Legend labels: Control, Treatment. Control series coordinate pairs: (2008, 2400), (2009, 2100), (2010, 1900), (2011, 1800), (2012, 1700), (2013, 1600), (2014, 1550), (2015, 1550), (2016, 1550), (2017, 1550), (2018, 1560), (2019, 1570), (2020, 1600), (2021, 1650), (2022, 1700). Treatment series coordinate pairs: (2008, 750), (2009, 780), (2010, 800), (2011, 810), (2012, 820), (2013, 830), (2014, 830), (2015, 840), (2016, 850), (2017, 860), (2018, 870), (2019, 880), (2020, 900), (2021, 950), (2022, 1000). The image D showing a line graph titled, Observed means. The horizontal axis label is, Year. The horizontal axis shows years from 2008 to 2022. The vertical axis label is, Combat Aircraft. The vertical axis range is 40 to 140. A vertical dashed line is drawn at 2014. Legend labels: Control, Treatment. Control series coordinate pairs: (2008, 135), (2009, 120), (2010, 100), (2011, 95), (2012, 92), (2013, 90), (2014, 88), (2015, 86), (2016, 84), (2017, 82), (2018, 82), (2019, 82), (2020, 82), (2021, 82), (2022, 82). Treatment series coordinate pairs: (2008, 35), (2009, 38), (2010, 40), (2011, 39), (2012, 38), (2013, 38), (2014, 37), (2015, 38), (2016, 37), (2017, 37), (2018, 36), (2019, 37), (2020, 36), (2021, 36), (2022, 36).
While the challenges with parallel trends estimation warrant caution, these collective results are consistent with H1, particularly for troop-based outcomes. Because both equipment outcomes fail their parallel trends tests, the significant results for armoured vehicles offer only descriptive and not causal evidence. These findings are generally robust to several alternate model specifications, which are detailed in the appendix. These include an event study, log transformations of all four DVs, dropping late NATO joiners (North Macedonia, Montenegro, Finland, Sweden), using wild bootstrap error weighting, and altering the treatment year to 2014 (given the IISS reporting lag discussed above).
While our D-in-D models suggest that NATO allies bordering Russia significantly increased military capabilities relative to more distant allies after 2014, they cannot, on their own, explain why overall NATO capability levels declined even as defence spending rose. Following Spirling and Stewart, we acknowledge statistical inference is most valuable not as definitive causal proof but rather one piece of a broader inferential framework – establishing facts about the world that can then be evaluated against competing theoretical explanations.Footnote 75 Our D-in-D research design identifies that a geographically differentiated pattern of capability adjustment exists, but it is not able to identify the mechanisms producing the alliance-wide divergence between spending inputs and capability outputs. We briefly consider three complementary mechanisms that help account for this divergence.
First, defence spending entails far more than investment in just troops and equipment. As Becker et al. document using disaggregated expenditure data, defence budgets are divided across personnel, equipment procurement, operations and maintenance, infrastructure, and research and development. Only personnel and equipment spending directly map onto the capability outcomes analysed here.Footnote 76 Even substantial post-2014 spending increases may have been concentrated in operations, maintenance, and infrastructure without generating observable increases in force structure. This dynamic helps explain why, in our fully specified models, controls for both personnel and equipment spending are statistically insignificant: increased spending does not necessarily translate into more troops or platforms when much of it is absorbed elsewhere in the defence budget.
Second, the relationship between defence spending and force structure is further complicated by long-run cost escalation trends in NATO personnel and equipment. Defence political economists have documented how the unit costs of both military personnel and major weapons platforms have risen substantially faster than inflation over time.Footnote 77 Even real increases in defence equipment and personnel spending may therefore be insufficient to halt or even reverse declines in platform counts and force size, as each successive unit of capability becomes more expensive to acquire and sustain. This dynamic operates alliance-wide and partially accounts for the persistent downward trend in average NATO capability levels visible in Figure 1, even as average defence spending climbed following the 2014 Wales pledge.
Third, the divergence in capability outcomes across the alliance reflects the domestic political economies of democratic defence budgeting. Research on the guns–butter trade-off documents how political representatives in democracies systematically direct defence resources towards politically popular ends, including defence industry employment, dual-use infrastructure, and benefits packages.Footnote 78 NATO allies geographically distant from Russia face weak domestic political incentives to expand active force structure: the Russian threat is abstract, public support for visible capability investment is limited, and alternative uses of defence spending offer more tangible political returns. By contrast, frontline allies bordering Russia face a domestic political permission structure in which capability investment is both strategically necessary and politically justifiable to their publics.Footnote 79 The result is that geography shapes not only strategic threat perception but also the domestic political feasibility of converting spending into the specific capability outputs measured here.
Conclusion
This article considered if the 2014 Russian invasion of Ukraine and annexation of Crimea led to heterogeneous military capability adjustments across European NATO members. The difference-in-differences (D-in-D) research design considered twenty-nine European NATO members’ military capabilities from 2008 to 2022. Outcomes of interest were baseline military capabilities important for deterring Russian aggression: total numbers of active military troops, active army troops, armoured vehicles, and combat aircraft. Testing two treatment specifications, the D-in-D models suggested that the 2014 Russian invasion interacted with geographic proximity to Russia to create significant differences in military capability portfolio trends across European NATO member states.
These findings offer tentative support for research highlighting the importance of threat proximity in balancing behaviour,Footnote 80 and similar trends in a European-specific context.Footnote 81 Beyond the nuances of benchmarks for measuring allied contributions to NATO, this analysis carries important policy implications as the United States considers significant alterations to its forward presence in Europe in the coming months.Footnote 82 These findings suggest that some NATO members not bordering Russia continued shrinking their military capacity after 2014 despite the increasing Russian threat, while others had a much longer lag before beginning modest capability increases, consistent with selective burden shifting. Conversely, NATO allies who are closest to Russia, and thus most at risk from future Russian territorial aggression, have done the most to maintain and expand their military capabilities, despite broader capacity reductions within NATO. This suggests that carefully monitoring allies’ achievement of their capability targets through the NDPP remains essential – to the extent that ‘output metrics’ can be made available, doing so would enable more careful public scrutiny.
This divergence in troop and equipment levels underscores a critical insight: military planning remains highly sensitive to national and domestic politics. This research has important implications for the political economy of alliances, capability development, and long-term strategic planning. The persistent shrinkage of capabilities by NATO members distant from Russia suggests that, even after a clear systemic shock, burden-sharing challenges continue to manifest, but often in subtle and selective ways, consistent with recent evidence that defence burdens remain unevenly distributed across NATO allies despite post-Wales spending commitments.Footnote 83 This uneven response aligns with models emphasising joint productsFootnote 84 and burden shifting.Footnote 85 The maintenance of smaller, specialised forces – ‘bonsai armies’Footnote 86 – by wealthier, distant allies, without adequate long-term capital investment, could indicate a strategy of ‘paying for protection’:Footnote 87 buying allied presence with defence expenditures aimed at pleasing the larger ally more than any other purpose. The fact that national defence choices vary in terms of equipment, personnel, and operations necessitates the continued use of disaggregated defence spending data to accurately capture and analyse these behaviours.Footnote 88
The process of capability adjustment is deeply intertwined with domestic political economies, highlighting that governments manage complicated and heterogeneous ‘guns vs. butter’ trade-offs by shifting resources towards politically popular uses. Our finding that capabilities adjustments following Russia’s 2014 invasion of Ukraine are heterogeneous, varying according to geography, supports findings that, in democracies, domestic and (in the case of EU members) EU-level factors like unemployment, defence industry employment, and fiscal rules continued to outweigh strategic ones in all but the most vulnerable frontline states. Such phenomena risk alliance cohesion and collective deterrence, particularly when the alliance requires robust capability outputs in a period defined by great power competition. Diplomatic efforts are aimed at attaining agreements among allies to increase capabilities during the ‘fog of peace’,Footnote 89 like the Wales and Vilnius pledges, as well as the 2025 Hague Summit declaration, in which allied heads of state and government agreed to ‘invest 5% of GDP annually on core defence requirements as well as defence- and security-related spending by 2035’, explicitly tying this spending to having ‘the forces, capabilities, resources, infrastructure, warfighting readiness, and resilience needed to deter and defend’.Footnote 90 While the continued requirement to make such statements attests to the collective action challenges inherent in an alliance with the wide distribution of national attributes that NATO has had from its birth,Footnote 91 their effective use attests to the alliance’s ability to take such collective action through practices of alliance politics.
The observed heterogeneity in troop and equipment adjustments, particularly the sharp divergence related to geographic proximity, provides fertile ground for continued study. Future research could test how military platform and personnel cost escalation across NATO impact force size and force structure. Qualitative analyses can clarify the precise causal mechanisms linking security threats to specific capability outcomes at the country level. This could be achieved through process tracing using evidence from participant interviews or primary source defence planning documents and would improve our understanding of how individual countries address deterrence and collective action dilemmas. Future quantitative analyses could leverage the recent availability and accessibility of disaggregated defence spending data (personnel, equipment, O&M, infrastructure) to model the complex and likely bidirectional relationship between these spending categories and military investment outcomes.
Finally, scholars and policymakers alike should continue to explore the policy implication that regionalisation in defence planning – where allies focus resources on tasks for which they are ‘most geographically suited’Footnote 92 – may be the most effective strategy for reconciling national interests with collective defence needs. This is crucial for strengthening deterrence, particularly as the United States shifts strategic resources towards global competition with China, rendering Europe potentially a ‘secondary theater’.Footnote 93 Geography still matters – even as effective collective action among NATO allies mitigates its effects.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/eis.2026.10095.
Acknowledgements
Thanks to Katherine Rittenhouse, Mary Evans, Jesse Johnson, two anonymous reviewers, and seminar participants at the LBJ School of Public Affairs for invaluable feedback that improved this project. The views expressed are those of the authors and do not reflect the official policy or position of the United States Military Academy, the Department of the Army, the Department of Defense, or the US Government.
Funding statement
The authors received no financial support for the research, authorship, and/or publication of this article.
Competing interests
The authors report there are no competing interests to declare.
Ethical approval
There are no human participants in this article and informed consent is not required.
Data availability statement
Replication data, Stata do-files, and a codebook are available at the Harvard Dataverse: https://doi.org/10.7910/DVN/FI9I40. Troop counts were hand-coded by the authors from The Military Balance (IISS, 2008–2022); the underlying volumes are copyrighted and are not redistributed. Remaining source files are separately deposited by their authors and are cited in the codebook.
Open access
