Abstract
The implications for funding a military, though important, are still not fully understood. Existing work often surmises that military spending is higher in majoritarian electoral systems that are predicated on personalistic ties. However, further examination casts doubt upon these findings. Accordingly, we present a pooled time-series cross-sectional analysis of military spending and electoral institutions and we find that party-based electoral systems, rather than majoritarian ones, foment higher military spending levels—which we attribute to these systems’ predilection for public goods spending. These results are robust even when a host of control measures and four different military spending metrics are employed.
Keywords
Introduction
Recent research on the factors affecting levels of military spending has identified institutional arrangements as important variables that influence defense funding. 1 Among the numerous explanations offered to account for variations in military expenditures, multiple works have reported that regimes with personalistic or candidate-centered political institutions (e.g., majoritarian electoral systems for legislative elections, presidential systems, etc.) are likely to engender higher levels of military spending. 2 The reason for this is that, in order to boost their reelection efforts, politicians in candidate-centered electoral systems are associated with more generous defense spending outlays, as these funds can be targeted to a specific constituency in a relatively visible manner. 3 Conclusions derived from this logic and these studies, however, are limited in some nontrivial ways. For starters, the notion that military spending is intended to benefit a small group of elites 4 or a subset of the population is overly restrictive. In fact, while some scholars contend that defense expenditures fail to produce economic benefits, 5 various other studies show that there are in fact numerous spillover benefits that emanate from military spending; 6 therefore, classifying defense spending as being similar to a private good warrants additional consideration. As well, the operationalization and estimation of the models used to reach the aforementioned conclusions have yet to take full advantage of additional measures for both military spending and explanatory institutional variables. This is problematic for examining military spending, as there are numerous concerns with the calculation and reporting of defense funding. Thus, using a single metric for this expenditure is worrisome. 7 Likewise, earlier attempts to gauge majoritarianism in these works have neglected to include more detailed characterizations of electoral systems that could strengthen our understanding of this connection. 8
In this study, we assess the link between political institutions and military spending by focusing on electoral system components to determine whether the previous works’ general conclusions hold both over time and in a pooled cross-sectional analysis of roughly fifty democracies. Given that several studies have identified institutional incentives as a leading reason for defense funding, 9 we continue in this vein as well. Our main emphasis is disaggregating electoral system components through the use of Carey and Shugart’s “Personal Vote Index” (PVI). 10 We start by testing the notion that candidate-centered or majoritarian electoral systems engender greater levels of military spending. In contrast with other studies, we find that military spending is higher in party-based systems, not in candidate-based ones. We account for this incongruent finding by emphasizing the public good benefits derived from military spending. The notion that party-based electoral systems (e.g., proportional representation [PR] systems) allocate more funding to military spending may, at a glance, seem counterintuitive, but when we consider the voluminous literature linking PR systems with public goods spending, this discovery is rather intuitive. In fact, several recent studies have detailed the various positive externalities that flow from military spending 11 while others have crafted lengthy explanations as to why military spending, and its consequences, better resemble a public, rather than private good. 12 Consequently, while military spending resembles a private good in certain cases, this type of spending, and the manifold political, social, and economic dividends that it produces, actually present sound justification for categorizing it as a public good.
This work contributes to the literature on defense spending by highlighting more specific features of electoral systems (i.e., party- vs. candidate-centered) and their effects on military spending compared with previous works. That is, rather than assuming that electoral systems affect military expenditures based solely on the broad categories of PR and single member district (SMD) systems, we examine the more nuanced nature of electoral institutions. Along with the four separate military spending measures, this nuanced approach reveals important information regarding the link between electoral systems and defense spending.
Why should we expect military spending to be contingent upon electoral systems? For starters, electoral systems have a well-established link with various types of government spending. 13 In fact, in many instances, electoral systems emerge as the most consistent and powerful predictor of fiscal policy making. 14 Furthermore, the behavior incentivized by institutional milieu that impacts public goods, private, and total government spending is expected to shape military spending also. While military spending is clearly different from many other spending types, there is still ample reason to believe it will be impacted by institutional design as well. After all, politicians are likely to be evaluated based on this type of spending, and the implications derived from it, just as they are other types of spending. Consequently, we argue—and discover—that the PVI is an important predictor of defense spending. While we accept that military spending may provide narrow benefits to a small segment of a population at times (i.e., it may be viewed by some as being similar to a private good), our findings do not support the notion that military spending is positively related to candidate-centered electoral systems. Instead, we find that military spending is higher in party-centered systems where politicians have greater incentives to provide public goods to their constituents.
Domestic-level Determinants of Military Spending
Most previous research on military spending has emphasized international-level factors; however, there have been a few studies that have examined the importance of domestic-level characteristics in predicting military spending patterns. At the aggregate level, many of these studies have assessed to what degree regime type (i.e., democratic and nondemocratic states) shapes military and social spending. 15 Beyond these works, most studies have focused their efforts on other explanatory variables. For instance, economic development has a long-standing connection with defense spending. 16 Political business cycles, unemployment rates, and inflation have also been linked with these policies. 17 Furthermore, recent work shows how both international factors and democratic political traits impact military spending. 18 Other scholars contend that public opinion can influence defense spending 19 and defense contracting. 20 Additional scholars have found that the ideological leanings of legislators (i.e., hawkish or dovish) affect military spending patterns and the awarding of defense contracts. 21
Defense Spending as a Public Good
In examining the effects electoral systems have on military spending, we need to further discuss why this expenditure would resemble a public good. One of the central differences between private goods and public goods is that public goods cannot be divided among recipients and are nonexcludable, whereas private goods can be distributed to certain portions of the population, divided among recipients, or withheld from some individuals or groups. 22 Thus, public goods are services and resources that cannot be withheld from any segment of a population regardless of how much an individual pays for a good. “This conclusion follows directly from the definition of a collective good. A collective or pure public good is defined by two properties: jointness of supply and impossibility exclusion. Once a public good is supplied by one member of a group, it may be enjoyed by all.” 23 Furthermore, public goods are also considered to be nonrivalrous. 24 That is, consumption of a certain amount of a good by an individual or group does not diminish the value of the good for another individual or group, and this is certainly the case with military spending which translates into state security that is difficult to divvy up among a populace. Furthermore, the security attained by one group of individuals as a result of national defense does not detract from the security of another group. Not surprisingly, “National defense is the classic example of a public good. It is non-rival, non-excludable and widely accepted as a primary function, if not the primary function, of national governments.” 25
While national defense is generally viewed as a public good, some scholars argue that military spending is similar to a private good because it fails to produce economic benefits, 26 and is often used to profit a specific group of elites within a state, especially in less democratic states. 27 Other scholars contend that defense spending can have public as well as private benefits. Bueno de Mesquita et al. state that “National security policy if oriented toward providing actual security for the residents in a state, is consumed as a public good, though its production is a private benefit to manufacturers, soldiers and so on.” 28 Thus, we contend that while individuals within the military, legislature, or defense contracting agencies can benefit from military spending, the good produced by military spending is public in nature. “The rule of law and the provision of national security are close to being pure public goods, though lawyers and generals certainly enjoy private gains when the rule of law and the provision of national security are promoted.” 29
Electoral Systems and Military Spending
Although several studies have investigated the systemic-level determinants of military spending, these writings have typically overlooked integral domestic factors. To be sure, a number of domestic factors have been analyzed in these studies, but most scholars find that electoral systems are among the strongest predictors of government spending. Many studies have held that states with majoritarian, SMD electoral systems spend less than states with PR systems. 30 The working consensus has offered two reasons for this finding. The first reason rests on the need to form ruling coalitions, which are commonplace in PR systems. Given the Duvergerian notion that more PR systems result in more political parties, forming coalitions is often plagued by more collective action difficulties given the more populous political landscape. Those subscribing to this notion conjecture that bargaining among potential coalition members becomes more complex as the number of parties increases. Since there are more parties, there are numerous discrepancies in fiscal policy priorities and, hence, it forces lawmakers to allocate more funds to placate more parties in order to maintain the ruling coalition—thus, resulting in greater total spending.
The second explanation for increased government spending stems from the incentives derived from PR contests. In these races, there are greater incentives for candidates to secure costlier public goods that generally reach a larger segment of the populace. 31 In most PR systems, parties must compete in large, nationwide districts, and to win seats, parties must reach as many voters as possible. This requires policies that affect more people than the less expensive private goods. Typically, this takes the form of education, health care, and/or pension allotments. Alternatively, it has been argued that goods, such as agricultural subsidies and highway funding, are effective means of soliciting electoral support in majoritarian contests. 32 Yet, for those politicians competing in larger districts, this strategy is less effective; therefore, parties competing in PR races often pursue public goods.
Conversely, it is also necessary to elaborate on the relationship between majoritarian or SMD systems and total spending. One reason for this difference lies with the number of parties. As Duverger explains, SMD contests generally result in fewer parties. 33 Based on this, problems occurring because of coalition formation and collective action problems should be less prevalent in majoritarian systems and spending should decline as a result. Another reason relates to the demand or need to attract support from smaller constituencies. Unlike the large, often nationwide districts in PR races, candidates in SMD systems can succeed if they target particularistic interests through agricultural subsidies, 34 highway dollars, 35 or through other similar policies. 36
How does all of this pertain to defense funding? Because of the public good nature of military spending, which provides for a state’s security, we argue that electoral system design is an influential factor for policy makers. For those seeking to get reelected in a PR system, public goods are often more effective and, at times, more efficient means of securing electoral victory. As such, funding the military and fortifying the state’s defense can be effective ways to garner popular support. Not only that, but military spending may stimulate economic growth, 37 curtail unemployment rates, 38 and increase both investment and consumption in the private sector. 39 Other scholars have found that reductions in military spending lead to lower levels of economic growth and higher unemployment rates. 40 Thus, there are several actors and groups who benefit from this expenditure.
Hypotheses
In this section, we present our formal hypotheses. While our empirical models examine four separate measures of military spending, our stated hypotheses only mention military spending as a percentage of a country’s gross domestic product (GDP) for brevity’s sake. The logic used, however, applies to the remaining dependent variables as well.
Given the growing literature on electoral systems, we know that electoral systems predicated on parties and party reputations (i.e., those states with larger district magnitudes or where personal vote incentives are reduced) are expected to fund public goods more generously than more candidate-driven electoral systems. Due to the complicated nature of electoral systems, it is necessary to control for multiple components of these mechanisms. That is, while existing research has relied on dichotomous indicators of electoral systems, 41 such an approach is unable to capture wide variations in electoral laws. Admittedly, binary indicators (i.e., single-member plurality districts vs. PR systems) of electoral systems are appropriate in certain settings, especially if there is little variation within each category of electoral systems. For this study, however, this is not the case. Furthermore, there is certainly more to an electoral system than whether or not it is PR or SMD, and using the PVI (in conjunction with other electoral system indicators) can augment our knowledge of these institutions vis-à-vis military spending.
The first electoral system indicator is district magnitude or the average number of seats per electoral district (this is not part of the PVI). As mentioned previously, scholars have noted that larger district magnitudes produce more government spending due to the increased reliance upon public goods spending. Predictably, we suspect that district magnitude or size should correspond with greater defense funding. Specifically, we anticipate that:
As other scholars have found, there are several noteworthy electoral mechanisms outside of district magnitude. In their PVI, Carey and Shugart highlight the following electoral laws: whether parties control nominations or ballot listing (Ballot), whether or not votes are pooled among or across party nominees (Pool), and whether or not voters select parties or individual candidates when casting a vote (Vote). The coding scheme created by Carey and Shugart is provided in Table 1. Because the PVI has been previously linked with other spending types, it should be a significant predictor of military spending as well. Therefore, the incentives to cultivate a personal vote are also expected to influence military spending.
Personal Vote Index.
Source: Carey and Shugart, “Incentives to Cultivate a Personal Vote,” 1995.
The first measure, Ballot, reveals whether or not parties control ballot formation. This is crucial since party strategies fluctuate with respect to this mechanism. Greater public goods spending is likely to surface in states where parties control ballot formation or in closed-list systems. In contrast, under open-list systems, candidates are likely to pursue particularistic goods rather than public goods spending since candidates face both interparty and intraparty electoral competition. In open-list systems, such as Brazil’s, we find that particularistic spending is a necessity as it enables candidates to distinguish themselves from other competitors. Furthermore, in open-list systems, it is difficult, if not impossible, to take sole credit for passing public goods legislation and thus spending on public goods should decrease. In other states that use closed-list systems, party labels are key since partisans compete with politicians from other parties and are more likely to cling to “brand names” rather than individual reputations. As such, the benefits of increased public goods spending can be attributed to parties, rather than individual candidates. In addition, parties in these systems have greater motivation to pursue policies that provide benefits for the party as a whole and therefore increased education, health care or pension funding is preferable to allocating more money through particularistic policies. For these reasons, we expect ballot control and military spending to be inversely related:
The second component in the PVI is whether or not votes are pooled across candidates (Pool). In states where votes are pooled across candidates, public goods spending is preferred to private goods spending. The reason for this is akin to the logic used in the discussion of list type. Where the party as a whole may benefit from vote-pooling (i.e., where the leftover votes from candidates can be carried over to other rank-and-file members), there is a greater need to push for policies that can garner greater attention, such as public goods. But when votes are not pooled across candidates, then there is less to gain by candidates or parties in focusing on public goods legislation. In this case, individual candidates may need to secure enough votes for themselves since they cannot expect to receive help from vote-pooling; therefore, private goods spending that can target a smaller segment of the population and produce greater dividends for individual candidates will be favored over public goods spending. The same logic should hold true for military funding. Therefore,
The final component of the PVI measures whether citizens vote for parties or individual candidates (Vote). In candidate-centric systems, where voters cast ballots for politicians rather than parties, there is more of an incentive to solicit personal vote support which should increase the demand for policies targeting a narrower constituency. Conversely, in party-centric systems where voters choose among parties rather than among individual politicians, we should see a greater emphasis on public goods spending. After all, devoting more funds to public goods is a more efficient way to increase parties’ base of support. Thus,
While we test each individual component of the PVI, it is also useful to combine these individual elements and test the index as well. Here, we use the data from Johnson and Wallack’s data set
42
and create the PVI which combines the three elements discussed previously. Although we detail our coding subsequently, it is worth noting that higher values of the PVI reflect more candidate-centric systems that place greater demand on cultivating a personal vote. Thus,
Data and Methods
In order to test the stated hypotheses, we have collected data from multiple sources. Although some researchers have looked at how military spending varies between authoritarian and democratic states, we focus only on the latter since elections and electoral systems often have little value in nondemocratic settings. The countries under review have all scored at least a “6” or higher on the Polity2 measure in the Polity IV data set. 43 The data set generally spans 1985–2005 and the unit of observation is country-year. 44 The Appendix contains a full list of the countries and years used in our analyses.
Dependent Variables
We measure the defense burden of states with four separate variables: military spending as a percentage of a country’s GDP, military spending as a percentage of central government expenditures (CGEs), the natural log of total military spending in US dollars, and total military spending per capita. Military spending, here, includes all capital expenses devoted to a state’s armed forces including peacekeeping units, defense and other relevant ministries, paramilitary sectors, and so on. 45 The first measure compares military spending to a country’s GDP while the second measure is the percentage of a country’s total budget devoted to military spending, which reveals how much a government prioritizes military spending compared to other budgetary needs. The final two dependent variables cast military spending as the total amount of military spending and total spending per capita, respectively. Both figures are in US dollars and, given the sheer size of third measure, we have used the natural log of total spending to facilitate interpretation of the coefficients. Total military spending data have also been collected from the Correlates of War (COW) database. 46 Using these four dependent variables provides for a more robust analysis and also acts as a safeguard against the idea that our findings are simply a function of one particular calibration of military spending. 47
The PVI
The primary independent variable of interest is the PVI. This is a composite index that accounts for whether votes are pooled across candidates, if party leaders control ballot access, and if votes are cast for parties rather than individual candidates. Because none of these measures is truly continuous, we have made a few modifications to these data. Specifically, in order to improve the straightforwardness in interpreting the results, these components have been recoded such that 0 indicates states where parties have ballot control; voters cast a ballot for parties not candidates; and where votes are pooled by the party (denoted as Ballot = 0, Pool = 0, and Vote = 0, respectively). 48 Each component is tested separately due to collinearity concerns. Then, we combine these components to create the PVI, which should be inversely related to defense spending metrics. These data have been collected from the Johnson and Wallack data set (which was truncated at 2005), which is based on the aforementioned Carey and Shugart index.
Additional Controls
In addition to the PVI, we control for other political, institutional, economic, and conflict-related variables. These include measures for logged district magnitude, presidential regimes, election years, and the government’s ideological makeup (Liberal Government). 49 We also consider controls for economic conditions: GDP per capita, unemployment, inflation (the consumer price index), and the percentage of a state’s population over sixty-five years old. These data were gathered from the World Bank. As for the military controls, we use the following: military personnel as a percentage of the total workforce and whether a state is involved in an interstate conflict. 50 We also include a measure for the major power status of states in an attempt to capture the potential for major powers to be involved in a greater number of international disputes. 51 Data for these interstate indicators were taken from the COW Project. 52 Membership in the North Atlantic Treaty Organization (NATO) may also impact military spending levels and we have controlled for this as well since military alliances could engender higher allocations to a state’s war chest. 53 Finally, we employ two commonly used measures of alliances. A state that is allied with the system leader may face fewer international threats because it can rely on the assistance of the system leader in times of crisis, thus resulting in a lower defense burden. Therefore, we use two measures of alliances to capture the type of alliances a state may have with the leader in the international system—Systemic Hegemonic Alliance and Direct Hegemonic Alliance. 54 These data were generated using the EUGene software generator. 55 Summary statistics for all variables are provided in Table 2.
Summary Statistics.
Note: CGE = central government expenditure; NATO = North Atlantic Treaty Organization; GDP = gross domestic product.
Methodology
A few additional notes are necessary to fully explain how we test our hypotheses. Because we use a pooled time-series approach, there are two major concerns that must be addressed: heteroscedasticity and autocorrelation. We use an ordinary least squares (OLS) regression with panel-corrected standard errors (PCSEs) to estimate our models in order to mitigate heteroscedasticity. 56 To be sure, there are other approaches to resolve this issue. In some situations, fixed effects equations may be ideal or preferable. However, fixed effects models raise a number of concerns that could undermine this study. For starters, fixed effects models may eliminate stable or time invarying covariates (e.g., district magnitude, presidentialism, NATO members, etc.) from the analysis 57 ; hence, using PCSEs are better suited for this study in order to retain these necessary controls. And, while a fixed effects model is appropriate for numerous applications, such an approach is not well suited to address the between-country variation contained in our sample. 58 Furthermore, the results of a Hausman test also indicate that fixed effects estimation is inappropriate for our data. In addition, other studies on military spending have used random effects estimation as well. 59
Regarding autocorrelation, there are two common remedies for dealing with this pathology. One solution is to use a lagged dependent variable; however, there are concerns with such an approach. 60 The other solution (and the one used here) is to utilize an Autoregressive Order One (AR1) process to address the presence of autocorrelation. A series of Wooldridge tests affirm the presence of serial correlation (in all models) and, hence, the need for an AR1 process. 61 Finally, there are several ways to test for panel unit roots including methods advanced by Im et al. and Levin and Lin. 62 Unfortunately, many of these tests cannot be used with unbalanced data sets (which is inevitable in this study). Because of this, we have used a Fisher test, which is another method to detect panel unit roots in unbalanced panels. As a result, we are able to reject the null hypothesis of nonstationarity (χ2 = 294.65, prob. > χ2 = .00, for the first set of results in Table 3; the remaining χ2 values were also statistically significant though are unreported here). As a result of these additional diagnostics, we are confident that stationarity problems are not undermining the current study.
Time-series Regression on Military Spending as a Percentage of GDP.
Note: GDP = gross domestic product; NATO = North Atlantic Treaty Organization; PCSE = panel-corrected standard error.
*p < .10; **p < .05; ***p < .01 (two-tailed tests).
Results and Discussion
Tables 3 –6 provide the results from our time-series cross-sectional analyses. The following findings reveal a number of interesting discoveries that not only support our hypotheses but also hold implications for future work. In addition, much of the variance is explained in these models and each model is statistically significant. 63 Overall, the results provide strong support for the hypotheses mentioned in the previous sections.
Time-series Regression on Military Spending as a Percentage of CGEs.
Note: GDP = gross domestic product; NATO = North Atlantic Treaty Organization; PCSE = panel-corrected standard error; CGE = central government expenditure.
*p < .10; **p < .05; ***p < .01 (two-tailed tests).
Time-series Regression on Military Spending (US$).a
Note: NATO = North Atlantic Treaty Organization; PCSE = panel-corrected standard error; GDP = gross domestic product.
aThe dependent variable used in Table 5 is the natural log of military spending in US dollars.
*p < .10; **p < .05; ***p < .01 (two-tailed tests).
Time-series Regression on Military Spending (US$) Per Capitaa.
Note: NATO = North Atlantic Treaty Organization; GDP = gross domestic product; PCSE = panel-corrected standard error.
aThe dependent variable used in Table 6 is military spending divided by the total population.
*p < .10; **p < .05; ***p < .01 (two-tailed tests).
In line with our expectations, the primary independent variables are significantly related with the various military spending indicators. Much like other public goods, military spending is necessary to bolster a nation’s security and defense and is significantly influenced by the PVI. The findings show a statistically significant and negative correlation between the PVI and this relationship holds even when we deconstruct the index into its component parts. We see that the measures for the PVI as well as two of the three individual components (Ballot and Pool) are significant predictors of all four dependent variables and are in the expected direction. The third component, Vote, is significant in some models and correctly signed, but the relationship is not robust across all specifications. This suggests that vote pooling has less bearing on military spending in US dollars and spending per capita than the other components of the index (as well as the index as a whole). Nevertheless, the Vote measure was significant and correctly signed in the first two models, thereby implying that vote pooling is a significant predictor of defense spending though it is contingent upon the dependent variable chosen. In general, we find that where parties do not control ballot formation and where voters select candidates rather than parties, military spending is adversely affected. In other words, military spending is higher in those states where electoral systems place greater emphasis on parties or where there is little incentive to cultivate a personal vote. Aside from this, we see that the PVI remains a statistically significant predictor of military spending in Tables 3 –6.
But what does this mean in practice? We can evaluate this effect by looking at changes in PVI and the subsequent changes in our dependent variables. In Table 7, we provide estimated values for military spending based on varying levels of the PVI. In more substantive terms, we see that moving from the most party-centric (PVI = 0) to candidate-centric (PVI = 3) systems, there is a noticeable drop-off in military spending. Specifically, we see that there is a decrease of 2 percent and 3 percent for our first two dependent variables, respectively. That is, party-centric electoral systems allocate roughly 3 percent more of a state’s budget to defense when compared with electoral systems predicated on individual candidates. For countries such as Botswana, Colombia, Ireland, and Finland, where the PVI is quite large, less money is earmarked for military consumption; likewise, in those states where the PVI is lower, and where party reputations are significantly more important, such as in Albania, Bulgaria, Germany, South Africa, and Venezuela, more money is set aside for military expenditures. The change in military spending (in US dollars) and military spending per capita exhibited similarly large changes when moving from party-centric to candidate-centric systems.
Estimated Values for Military Spending at Various PVI Levels.
Note: GDP = gross domestic product; CGE = central government expenditure; PVI = Personal Vote Index; DV = dependent variable.
The remaining political institutional variables revealed other noteworthy findings. For instance, the other electoral system indicator, logged district magnitude, is hypothesized to exert a positive effect on defense spending. Our expectations regarding this electoral system measure were consistently confirmed in three of the four dependent variables. In Tables 4 –6, district magnitude was positively signed and statistically significant in most models; yet, the measure was an insignificant predictor of military spending as a percentage of a country’s GDP. 64 Altogether, the direction of the coefficients for the electoral system variables reveals that military spending is significantly higher under those systems that place a premium on public goods spending. 65 In addition, the results indicate that Presidential systems significantly increase the amount of money a country devotes to military funding in all but one set of results (i.e., Table 4). This finding dovetails with our broader theory in that military spending can be a vital cog in providing a valuable public good since presidents are accountable to the entire country and thus have a vested interest in increasing military strength to stave off threats. Elsewhere, we find that Major Powers exhibit a greater proclivity for military investment which is likely an integral part of becoming a major power state. Like the Presidential measure, the Major Power indicator was significant in all tables but Table 4, revealing a generally robust relationship with the various measures of military spending. It is also worth noting that both the Major Power and the Presidential indicators narrowly missed reaching statistical significance in Table 4 while the direction of the coefficient was consistent with the results in the other tables.
Aside from these findings, there are several other valuable revelations from these models. For instance, the percentage of a country’s population employed by the military was an accurate predictor of military spending and, not surprisingly, it was positively aligned with our dependent variables. This is in the expected direction since employing larger military forces should inevitably require greater funding. In Tables 3 and 4, the results also suggest that alliances do in fact play a role in military funding. Namely, systemic alliances often result in reduced military expenditures. Thus, we find lower military spending levels in those states that possess alliance portfolios more closely aligned with the leader in the international system. 66 The direct alliance variable, however, was significant only in Table 3. The findings also reveal that macroeconomic conditions impact defense funding in numerous important ways. The most consistent economic predictor of our dependent variables was unemployment, which was statistically significant in all but two models (models 12 and 16).
Conclusion
While defense spending can benefit private groups associated with the military—generally—security is a good that cannot be excluded from segments of the citizenry and should be considered a public good. This finding challenges earlier assertions that military spending is a private good and not a public good. In addition, our study presents a more robust test of the link between electoral systems and military spending that offers greater leverage for this finding. As the results indicate, we believe that electoral systems which promote party-centered incentives correspond with increased defense funding because of the various benefits that arise from it (e.g., security, lower unemployment levels, higher investment, etc.). In this respect, military spending resembles a public good more than extant work has acknowledged. In a broader sense, this finding comports with the voluminous electoral system literature which has repeatedly linked institutional design with policy incentives. In our study, though, we learn that MPs who must appeal to a national (rather than local) constituency are inclined to allocate more money to defense funding in order to appease their given constituencies. Thus, we argue that examining the political incentives democratic leaders have to allocate more or less funds for defense, based on the design of their electoral systems, is an important avenue of inquiry and one this study addresses.
Footnotes
Appendix
| Country | Years | Country | Years |
|---|---|---|---|
| Albania | 1996–2004 | Latvia | 1991–2004 |
| Argentina | 1985–2004 | Lesotho | 1994–2004 |
| Australia | 1988–2004 | Lithuania | 1991–2004 |
| Austria | 1988–2004 | Macedonia | 1994–2004 |
| Bahamas | 1988–2004 | Madagascar | 1992–2004 |
| Belgium | 1988–2004 | Mali | 1992–2004 |
| Belize | 1987–2004 | Malta | 1988–2004 |
| Bolivia | 1985–2004 | Mauritius | 1988–2004 |
| Botswana | 1988–2004 | Mexico | 1997–2004 |
| Brazil | 1985–2004 | Moldova | 1994–2004 |
| Bulgaria | 1990–2004 | Namibia | 1999–2004 |
| Canada | 1988–2004 | Netherlands | 1988–2004 |
| Cape Verde | 1993–2004 | New Zealand | 1988–2004 |
| Chile | 1989–2004 | Nicaragua | 1990–2004 |
| Columbia | 1988–2004 | Norway | 1988–2004 |
| Croatia | 2000–2004 | Panama | 1989–1999 |
| Cyprus | 1988–2004 | Papua New Guinea | 1988–2004 |
| Czech Republic | 1994–2004 | Peru | 1989–2004 |
| Denmark | 1988–2004 | Philippines | 1980–2004 |
| Dominican Republic | 1988–2004 | Poland | 1991–2004 |
| Ecuador | 1988–2004 | Portugal | 1991–2004 |
| El Salvador | 1984–2004 | Romania | 1988–2004 |
| Estonia | 1991–2004 | Russia | 1993–2004 |
| Finland | 1988–2004 | Senegal | 2000–2004 |
| France | 1988–2004 | Slovakia | 1983–2004 |
| Germany | 1990–2004 | Slovenia | 1991–2004 |
| Ghana | 2001–2004 | South Africa | 1993–2004 |
| Greece | 1988–2004 | South Korea | 1988–2004 |
| Guatemala | 1996–2004 | Spain | 1988–1999 |
| Guyana | 1992–1997 | Sweden | 1988–2004 |
| Honduras | 2000–2004 | Thailand | 1992–2004 |
| Hungary | 1990–2004 | Turkey | 1988–2004 |
| India | 1988–2004 | Ukraine | 1995–2004 |
| Indonesia | 1999–2004 | United Kingdom | 1988–2004 |
| Ireland | 1988–2004 | Uruguay | 1988–2004 |
| Israel | 1988–2004 | United States | 1985–2004 |
| Italy | 1988–2004 | Venezuela | 1991–2004 |
| Jamaica | 1990–2004 | ||
| Japan | 1988–2004 | ||
| Kenya | 2002–2004 |
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.