Table C1 contains summary statistics for the key variables in the analysis, including these controls. The first five columns are devoted to the full sample used in the study. The final four columns are means by quartile of the instrument representing housing price shocks, the importance of which I describe below. Military Burden is the percentage of GDP devoted to military expenditures, the most commonly used measure of defense investment in the literature, and a favored measure of both NATO and the EDA. Personnel, equipment, operations & maintenance (O&M), and infrastructure are the proportions of national defense budgets allocated to each one of those categories of defense expenditure, as
reported by states, then verified and published by NATO and the EDA. These disaggregated defense expenditures are my key dependent variables, with my closest analysis reserved for personnel, which correlates negatively with each of the other three categories of expenditure.
Moreover, on the side of the dependent variable, simply disaggregating standard defense investment data into the four categories used by both NATO and the EU helps capture some of the more nuanced components of burden sharing behavior. From a policy standpoint, this technique of disaggregation helps clarify the relationship between “inputs”
focused approaches and “outputs” focused approaches.1 Such large-n analysis is
complementary to detailed case studies. Quantitative measures disaggregated in this way capture some of the same features of state behavior that recent qualitative studies do –
1 Distinctions between inputs, intermediate outputs, and final outputs are explored in detail in Melese, Richter, &
Solomon (2015). Sololsky & Adams (2017) recently highlighted the policy distinction between inputs and outputs.
For Peer Review
xii
taking advantage of those quantitative measures improves analysis and allows scholars to be assess whether their findings can be generalized beyond the cases they have studied in detail. It also offers insights into behavior that policy-makers are not likely to acknowledge in public – such as shifting resources away from areas they pledged to allies to invest in in order to stimulate domestic economies (Becker, 2017).
The share of defense budgets allocated to the four categories identified by NATO and the EU is the best available indicator to capture such burden-shifting. For example, doing so captures not simply numbers of personnel (which depend largely on population), or raw expenditures on personnel, but the extent to which governments prioritize
personnel spending relative to other areas. While very few European countries can expect any stimulus effect from equipment2 or operations, they can still hope for some effect by, for example, avoiding personnel cuts, increasing salaries, or maintaining retirement pay while cutting O&M.3
2 While countries without defense industries often receive offsets from the seller, there is no reason to believe that such offsets would have an effect approaching that of other forms of stimulus, including personnel expenditures.
Offsets are by their nature a fraction of the purchasing cost of the equipment: “In 2014, U.S. firms reported entering into nine new offset agreements with members of the EDA valued at $1.68 billion (U.S. Department of Commerce, Bureau of Industry and Security, 2016, p. 20),” approximately 3% of the approximately $50 billion in equipment spending among EDA members that year.
3 Only countries with significant defense industries can even hope to have any stimulus effect with equipment spending. There is no theoretical or practical reason to believe that O&M spending such as training or overseas deployment would have any effect on employment. While governments may believe that infrastructure spending may create jobs in, for example, the construction industry, the data indicate that the relationship between
unemployment and infrastructure spending is small, positive, and statistically insignificant. The dependent variable is the extent to which countries engage in this kind of behavior and not whether it has the desired macroeconomic effect. Supplementary File B contains detailed discussion of this phenomenon, along with a summary of data availability by country-year.
For Peer Review
Table C1: Summary Statistics
Variable
Obs Mean Std. Dev. Min Max Q1 Q2 Q3 Q4
Military Burden (Defense Spending/GDP) 1619 2.69 1.79 0.12 13.28 2.26 2.46 2.34 2.25
Personnel (% Defense Spending) 756 55.89 14.01 0.00 85.50 56.96 54.89 54.28 51.68
Equipment (% Defense Spending) 851 15.21 7.16 0.30 38.30 15.74 16.00 15.87 17.63
O&M (% Defense Spending) 702 24.14 8.85 6.83 48.49 24.39 24.68 25.53 26.27
Infrastructure (% Defense Spending) 704 3.48 2.70 0.00 19.00 3.04 3.02 3.36 3.51
Unemployment Rate 886 8.30 4.28 0.42 27.47 10.48 8.05 6.97 7.03
Housing Instrument (log change in OECD housing index) 700 0.97 6.26 -58.49 46.32 -5.82 -0.34 2.58 7.46
Spillins (log) 1604 13.52 0.33 10.58 13.93 13.64 13.53 13.55 13.59
GDP (log) 1335 25.26 1.95 20.37 30.45 26.24 26.74 26.50 26.43
Population (log) 1523 16.22 1.51 12.60 19.57 16.52 16.91 16.65 16.54
State Threat (capability*intent/proximity) 778 6501.42 2639.48 2145.31 13649.79 7522.27 6443.92 5919.22 5138.55 Vulnerability to Terrorism (citizens killed) 927 48.82 183.17 0.00 2521.00 55.15 47.90 36.26 59.89
NATO Strategic Excludability 1619 0.58 0.49 0.00 1.00 0.68 0.68 0.71 0.65
Right-Leaning Party 1592 0.27 0.44 0.00 1.00 0.36 0.43 0.40 0.36
Veto Points 700 4.28 1.43 1.00 16.00 4.56 4.42 4.38 4.53
Military Share of Labor Force 695 1.78 1.25 0.31 7.40 1.60 1.63 1.60 1.75
Atlanticism 577 6.02 7.63 -37.36 100.00 7.28 6.28 6.20 9.52
All Housing Quartiles
Mean by Quartiles of Housing Instrument
For Peer Review
xiv
Table C2 reports the correlations among my dependent and independent variables and the battery of control variables in my model, which are summarized in Table 1 in the main paper. Unemployment is positively correlated with the share of personnel in overall defense spending, and is negatively correlated with the share of defense spending in GDP, and the share of equipment, O&M, and infrastructure in defense spending. These
correlations are consistent with the theory that countries shift the burden of domestic unemployment to their allies by spending less on defense goods that benefit the group and more on defense goods that may generate private benefits. Also as theorized,
unemployment is negatively correlated with my housing instrument – negative price shocks in the housing market are associated with higher unemployment in subsequent years.
For Peer Review
xv Table C2: Bivariate Correlations
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
1 Military Burden (Defense Spending/GDP) 1 2 Personnel (% Defense Spending) -0.3857* 1 3 Equipment (% Defense Spending) 0.3455* -0.6494* 1 4 O&M (% Defense Spending) 0.1278* -0.8210* 0.2947* 1 5 Infrastructure (% Defense Spending) 0.0015 -0.3633* 0.0434 0.2169* 1
6 Unemployment Rate -0.1795* 0.3237* -0.1751* -0.2754* -0.1121* 1
7 Housing Instrument (log change in OECD housing index) -0.011 -0.1237* 0.1183* 0.0708 0.0866 -0.3170* 1
8 Spillins (log) -0.5962* 0.3115* -0.2144* -0.2172* 0.0973* 0.2513* -0.0579 1
9 GDP (log) 0.0864* -0.2514* 0.4958* 0.2546* -0.3186* -0.0245 0.0342 -0.0890* 1
10 Population (log) 0.4288* -0.3315* 0.4848* 0.2273* -0.2969* -0.0433 0.0136 -0.3474* 0.7714* 1 11 State Threat (capability*intent/proximity) -0.2253* 0.0283 -0.0955* 0.0142 -0.0433 0.1986* -0.2709* 0.5104* 0.0794* -0.0654 1 12 Vulnerability to Terrorism (citizens killed) 0.1895* -0.1003* 0.2321* -0.005 -0.0561 0.041 0.0103 -0.1551* 0.1593* 0.2702* 0.007 1 13 NATO Strategic Excludability -0.0383 -0.0225 -0.0127 -0.0561 0.0451 -0.0731 -0.0133 0.0682* -0.0111 0.0209 -0.1327* 0.1049* 1 14 Right-Leaning Party -0.0613* -0.0012 0.0884* 0.0436 -0.0455 0.0261 0.0117 0.1853* 0.1741* 0.0042 -0.1811* 0.0670* 0.1036* 1 15 Veto Points -0.0666 -0.0921* -0.0303 0.1480* 0.1139* -0.2328* -0.026 -0.0846* 0.1261* 0.005 -0.0236 -0.0674 0.1196* 0.0251 1 16 Military Share of Labor Force 0.8104* -0.0665 0.2407* -0.1857* -0.0888* 0.0127 0.0451 -0.2818* -0.0796* 0.2251* -0.3020* 0.2278* 0.4646* 0.0775* -0.1658* 1 17 Atlanticism 0.0069 -0.2908* 0.1592* 0.3152* -0.0117 -0.0912* 0.0541 -0.067 0.2034* 0.1577* -0.1039* -0.0038 -0.1206* 0.0373 0.0405 -0.027 1
* p<0.05
For Peer Review
xvi Figure C1: Bivariate Correlations, Full Sample
20406080100