Political Accountability and Military Spending

ABSTRACT

Over the past two decades, evidence on the relationship between democratization and reductions in military spending has accumulated. This association has proven to be robust to a wide variety of specifications and samples. Nevertheless, there remain two important gaps in our understanding of this relationship. Firstly, while most studies argue that democratic political institutions constrain the incumbents’ power to allocate resources to the military, there is still disagreement about which institution. Second, empirical evidence on causality is very scarce. Endogeneity remains a difficult problem to grapple with. To address these issues, I specify a clear set of channels of interaction linking democratic political institutions to military spending. Following previous work, I also argue that political accountability constrains allocations to the military. However, I distinguish three types of accountability: horizontal, vertical and diagonal. I find that diagonal accountability is the strongest mechanism. Furthermore, I use a country’s accumulated experience with political accountability as an instrumental variable to tackle the endogeneity between political accountability and military spending. I find evidence suggesting that political accountability causes military spending as a share of GDP to fall.

KEYWORDS:

• Military spending
• political accountability
• democracy

JEL CLASSIFICATION:

• D72
• H56
• C36
• C23

Introduction

In 2020, the Chilean Congress shifted resources away from arms procurement to cover the costs associated with the COVID-19 pandemic. Until recently, however, Congress had no control over this part of the budget. A legislation in force since 1958 earmarked a percentage of copper export revenues to arms procurement and maintenance, irrespective of Congress’ consent. After many attempts, this legislation was finally replaced in 2020 for a system whereby Congress must approve allocations to arms procurement, significantly improving civil society’s control over public resources. That same year, Brazil’s President Jair Bolsonaro presented a budget proposal to Congress calling for a 49% increase in the Defence budget for 2021. Bolsonaro’s budget proposal, if approved, would have raised the Ministry of Defence’s budget above to that of Education. Several media outlets condemned the proposal, echoing a widespread disapproval among citizens. Raising military spending above Education did not represent a preference of Brazilian society at that point, something that became evident to Bolsonaro soon after his proposal was announced. After the backlash, the President revised the Ministry of Defence’s budget downwards, below Education.

These anecdotes illustrate how military spending can be controlled democratically. In the case of Chile, the empowerment of Congress led to a use of public resources more in line with society’s most urgent needs. As for Brazil, the engagement of civil society, amplified by a free press, signalled that raising military spending was undesired, and that Education rather than Defence should be prioritized. Both cases exemplify much of what has been found in the literature on the relationship between political institutions and military spending. Over the past two decades, accumulated evidence has consistently found that democracies spend less on their military (Yildirim and Sezgin 2005; Fordham and Walker 2005; Dunne and Perlo-Freeman 2003).

At the same time, the anecdotes also illustrate one of the main gaps in the literature. Most studies argue that democratic political institutions constrain the incumbents’ power to allocate resources to the military (Bove and Nisticò 2014; Fordham and Walker 2005), but there is disagreement about which one. The compound quality of democracy ultimately leads to a variety of theories on its effects, each underscoring specific attributes. Brauner (2015), for instance, stresses the role of electoral accountability, arguing that elected politicians have incentives to favour social expenditure over military spending. Lebovic (2001) and Fordham and Walker (2005) put forward similar arguments. On the other hand, Rosh (1988) also posits that deliberation and public debate about resource allocation is the mechanism linking democracies to lower military spending. These explanations point to different accountability mechanisms and extant evidence is yet to identify which one has the most significant effect.

The difficulty in ascertaining which type of accountability has the strongest effect on military spending is mainly due to data limitations. Most studies used the same measurement of political regimes, namely Polity IV (Marshall, Jaggers, and Gurr 2017), to test different hypotheses. Polity IV measures a very specific trait of political regimes, namely the contestation of office and constitutional constraints on the head of state. Other critical aspects to democracy, such as suffrage, are not included (Boese 2019). Ultimately, available evidence on the relationship between democratic political institutions and military spending is limited to a very narrow conceptualization of democracy. It was not until recently that finer-grained measurements of political institutions were produced (Coppedge et al. 2021), allowing us to revisit its relationship with military spending with better data.

In what follows, I tackle two problems with our current understanding of how political institutions and military spending intertwine. First, I build upon recent theoretical refinements of Democratic Peace introduced in Hegre, Bernhard, and Teorell (2020) and Fjelde, Knutsen, and Nygård (2020) to specify a clear a set of channels of interaction linking democratic political institutions to military spending. Following much of previous scholarship, I also argue that mechanisms of political accountability are associated with lower military spending. However, I distinguish between three types: horizontal, vertical, and diagonal accountability. I rely on the Varieties of Democracy (V-Dem) dataset (Coppedge et al. 2021) for measures of political accountability and its components. This dataset has an unprecedented level of disaggregation that allows me to unbundle the relationship between military spending and political accountability into its smaller components, shedding light on its nuances.

A second problem is the direction of this relationship. Although the association between democracy and military spending is consistent, there is very limited evidence about whether democracy causes lower military spending or if increasing the latter weakens the former. Only Kimenyi and Mbaku (1995) and Brauner (2015) dealt with this issue directly. In this paper, I leverage recent improvements in measuring institutional legacies (Edgell et al. 2020) to build a measure of political accountability stock that I use as an instrument to tackle the problem of endogeneity. I find that endogeneity is indeed an issue, and that political accountability causes military spending as a share of gross domestic product to fall by as much as 49%.

This paper contributes to a growing literature on the role of institutions and military spending. For the most part, scholarship on the determinants of military spending has accounted for institutional factors merely as additional controls, but rarely at the forefront of the analysis. Partly influenced by the surge in institutional explanations for economic development in mainstream Economics (Acemoglu and Robinson 2006; Besley and Persson 2011), this has been changing. Scholarly work on the matter is taking political power more and more seriously (Bove and Nisticò 2014; Leon 2014; Brauner 2012, 2015; Töngür, Hsu, and Elveren 2015), approaching military spending as the outcome of a particular constellation of interests that effect change while at the same time are constrained by institutions. This paper seeks to take another step forward in this direction.

Furthermore, this paper also contributes to ongoing policy and academic debates on the consequences of autocratization. Understanding the intricacies of military spending and political institutions became even more relevant with the proliferation of autocracies around the world in recent years. Autocracies, for the first time since 2001, are the majority in the world. Scholarship has identified a ‘third wave of autocratization’ (Lührmann et al. 2020a), as the state of democracy in countries like Hungary, India, Brazil, and Turkey continues to decline. The well-established link between democracy and military spending sheds a concerning light on these developments, suggesting that as autocratization surges, military spending may follow suit.

Extant Literature and Knowledge Gaps

Research on the relationship between military spending and political institutions has consistently found that democracies have lower military burden, i.e. military spending as a share of gross domestic product (Kollias et al. 2018; Dunne and Perlo-Freeman 2003; Fordham and Walker 2005; Yildirim and Sezgin 2005). Brauner (2015) finds that as democracy improves – measured as a 1-point increase in the Polity IV scale – military burden falls by 2%. This result ultimately means that full democracies spend as much as 40% less on the military than full autocracies. The literature has also addressed, albeit less frequently, the variation in military spending within democracies. Albalate, Bel, and Elias (2012) find that presidential democracies have higher military spending than parliamentary systems. There is also evidence that social democracies spend less on their militaries compared to the other regime types (Töngür, Hsu, and Elveren 2015).

Explanations for such a consistent finding vary, each stressing different mechanisms. Fordham and Walker (2005) make the case that liberal democracies allocate fewer resources to their militaries 1) as a means to curb spiralling security dilemmas, and 2) to avoid the deleterious economic effects of preparing for war, such as debt and fewer resources available to other social goods. This argument draws heavily on Immanuel Kant’s writings on a Perpetual Peace (Kant 1991). Democracies are peace prone mainly because of the role of civil society: individuals are hesitant to support warmongering incumbents as the costs of conflict, human and economic, will largely fall upon them. Liberal democratic institutions allow civil society to express its aversion to conflict as well as removing incumbents that act in disagreement with this preference.

Liberal approaches to the relationship between democracies and military spending underscore the role of institutional constraints on incumbents, namely free and fair elections, and the legal protection of individual freedoms (Albalate, Bel, and Elias 2012; Brauner 2015). In that vein, Rosh (1988) argues that the rule of law, deliberation and competitive elections are associated with lower military spending, and Yildirim and Sezgin (2005) stress the responsiveness of elect incumbents. Also, on the effect of elections, Bove, Efthyvoulou, and Navas (2017) find that governments tend to bias outlays towards social expenditure and away from military expenditure at election times.

Rota (2016) introduces an important nuance to the liberal strand, arguing that some democratic features exert stronger effects on constraining military spending than others. Democracy, it argues, should not be treated as a monolithic black box, but rather unpacked into its components. Not every facet of democracy should be equally associated with lower military spending. In that sense, Rota (2016) finds that between 1880 and 1938, constraints on the Executive were less important in reducing military burden than effective public participation. Similarly, Elbadawi and Keefer (2014) argue that not every process of democratization leads to lower military spending – they find that only deeper democratic reforms have that effect.

The works of Bueno de Mesquita et al. (2003); Bueno de Mesquita and Smith (2009, 2010) on Selectorate Theory have also been used to assess the relationship between democracy and military spending. Selectorate Theory hypothesizes that the main feature of a political regime is the size of the incumbent’s winning coalition – the subgroup of the constituents who maintain the incumbent’s tenure in exchange of special privileges. Regardless of the formal aspects of a regime, the incumbent must appease its winning coalition to remain in power; this relationship determines both revenue-extraction and resource allocation. Bove and Brauner (2016) partly rely on Selectorate Theory to explain why military regimes display the highest levels of military burden among autocratic regimes.

Whilst extant evidence converges on a negative relationship between democracy and military spending, a mismatch between theoretical claims, data and research design weakens the validity of the results. The most widely used data source for political regimes, Polity IV’s 21-point democracy index (Marshall, Jaggers, and Gurr 2017), measures only a specific type of democratic accountability. The index aggregates five different components measuring a polity’s executive recruitment, constitutional constraints, and political participation. Other sorts of constraints on the incumbent, such as an active civil society and suffrage, are not covered (Boese 2019). As a result, the only theoretical claim possibly tested with Polity IV is whether contestation and checks and balances on the incumbent are associated with military spending.

Because different theoretical claims are tested using the same measurement of democracy, one can hardly evaluate the explanatory reach of each theory, nor if its mechanisms are sound. An example is the conflation of political regimes and the size of winning coalitions (Fordham and Walker 2005; Bove and Brauner 2016). While there is certainly a correlation, these two elements must be taken separately, that is, the association of political regimes and military spending must be assessed while controlling for the size of winning coalitions.

Finally, the theoretical claim that political institutions cause variations in military spending is yet to be sufficiently addressed. Only two papers tackled causality empirically. Kimenyi and Mbaku (1995) use the predicted values of the Political Democracy Index (PDI) (Bollen 1980) as an instrument to military spending in a cross-section of 87 countries for 1980. It finds that military spending hinders democratization in developing countries. Brauner (2015) revisits the issue using the lagged value of democracy as an instrument to democracy at $t$. It finds that endogeneity is not an issue: the coefficients of the two-stage least square regressions were very close to those of the fixed effects specifications. However, Brauner (2015) is aware of the limitations of this instrument, cautioning that the results should not be overstated. On the other hand, Brauner does find that democracy causes military spending to fall using a Granger causality test. Both papers have made important contributions, but the results are still far from conclusive.

Research Design

Political Accountability and Military Spending

Political accountability leads to lower military spending because the de facto constraints on government incentivize incumbents to choose policies in tandem with civil society’s preferences. In case of defection, incumbents are faced with potential sanctions from citizens and oversight institutions (Lührmann, Marquardt, and Mechkova 2020b). Accountability mechanisms lead to diligence in military spending, assuring decisions are taken with discretion and resources are allocated authoritatively (Mechkova, Lührmann, and Lindberg 2019). If unconstrained, incumbents may allocate resources to the military in exchange for political support or to repress internal dissent violently as a means to extend their political survival (Bueno de Mesquita et al. 2003; Acemoglu et al. 2010). Whilst public-spirited individuals can occasionally ascend to govern and act in society’s best interests, mechanisms of political accountability must be in place to constrain the behaviour of those who lack these virtues (Przeworski, Stokes, and Manin 1999).

Hypothesis 1: Mechanisms of political accountability constrain military spending.

Political accountability stems from different, yet converging, directions. Horizontal accountability depends on the existence of state agencies that are legally empowered to oversee, sanction or impeach incumbents in case of unlawful actions (O’Donnell 1998). Usually, horizontal accountability takes form as the separation of state institutions into Executive, Legislative and Judicial branches, all counter-posing each other. Parliament, for instance, binds Executive power to allocate resources to the military if budgets must be approved by vote. Furthermore, the Parliament can also oversee the execution of military budgets. In many countries, the Minister of Defence must report back to Congress on arms acquisitions and other related expenses. On the other hand, if military spending is mismanaged or allocated on the basis of personal political gain (Bueno de Mesquita et al. 2003), a strong and independent Judiciary can prosecute incumbents. In the shadow of sanctions, incumbents have incentives to comply to their constituent’s preferences rather than defecting (Rose-Ackerman 1996). This system of checks and balances ensures that decisions to allocate resources to the military are not wasteful, also pondering the opportunity costs involved.

Hypothesis 2: Mechanisms of horizontal accountability constrain military spending.

Vertical accountability is the result of constituents enforcing their preferences via voting. Elections are a powerful instrument to hold defective politicians accountable (Przeworski, Stokes, and Manin 1999). To act as a constraining mechanism, elections must encompass an electorate that decides whether an incumbent should remain in office or not and an incumbent who can reasonably anticipate the electorates decision (Ashworth 2012). Free and fair elections assure that incumbents ponder the political costs of elevating military spending above what is socially acceptable; if this threshold is surpassed, chances of re-election grow slimmer. Since public goods such as Health and Education are more likely to appeal to voters during peacetime (Mintz 1988), incumbents favour the former over military spending. An obvious yet important caveat is that the effects of vertical accountability become more evident during election periods. Bove, Efthyvoulou, and Navas (2017) present some evidence on this effect, finding that governments shift resources away from military spending at election times.

Hypothesis 3: Mechanisms of vertical accountability constrain military spending.

Elections are seasonal; civil society can only express its preferences through vote when elections are held. Yet engagement in political processes in between electoral cycles can push incumbents to take civil society preferences into account continuously (Hegre, Bernhard, and Teorell 2020). Mechanisms of diagonal accountability concern the role of non-state actors, such as civil society and the media, in constraining political decisions. Civil society can increase the costs of non-preferential policies over time through various means, such as protests, monitoring and oversight of policy decisions and engagement with bureaucracies. It can demand transparency and responsiveness from the government as well as a reasonable justification for policies (Sharma 2009). For its part, a free press is critical to bringing important issues to the forefront of public debate and informing citizens, so they can make better choices. Diagonal accountability is a powerful channel, with the potential to supplant other types of accountability (Bernhard and Anna 2019).

Hypothesis 4: Mechanisms of diagonal accountability constrain military spending.

Empirical Strategy and Data

My benchmark specification follows a well-established demand function for military spending (Bove and Brauner 2016; Brauner 2015; Dunne and Perlo-Freeman 2003). I assess the association of the various measures of political accountability and military spending as a share of GDP, estimating a linear model using ordinary least squares (OLS) estimation with country (${\alpha }_i$) and year (${\gamma }_t)$ fixed effects to control for unobservable factors that may affect military spending, as well as country-specific time trends to account for unobserved heterogeneity across units. Year-fixed effects address the concomitant reduction in military spending and the growing pace of democratization around the world starting in the mid-1970s and following through the 1990s. The ‘third wave of democratization’ (Huntington 1991) swept away authoritarian regimes in several countries across Europe, Asia, and Latin America. Around the same time, global levels of military spending started to fall as the Cold War waned. Certainly, these two processes were not independent of each other. However, the mitigation of international tensions must be considered as a different element from the multiplication of democracies around the world. The year-fixed effects address these wider time trends. I report clustered standard errors by country in all specifications to account for panel-specific heteroskedasticity.

(1) $\mathrm{l}\mathrm{n}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}\mathrm{y}\mathrm{b}\mathrm{u}\mathrm{r}\mathrm{d}\mathrm{e}{\mathrm{n}}_{\mathrm{i}\mathrm{t}}={\beta }_0+{\beta }_1\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}{\mathrm{y}}_{\mathrm{i}\mathrm{t}}+{\beta }_{\mathrm{n}}{\mathrm{X}}_{\mathrm{n}}+{\alpha }_{\mathrm{i}}+{\gamma }_{\mathrm{t}}+{\mathrm{u}}_{\mathrm{i}\mathrm{t}}$(1)

Data on military burden – military spending as a share of GDP – are from the Stockholm International Peace Research Institute (SIPRI). The dataset covers 173 countries from 1949 until 2020. I chose to exclude observations for 2020 because a significant part of these figures is based on budget projections instead of actual expenditure. As such, they tend to overestimate military spending. SIPRI military expenditure data include all current and capital expenditure on the armed forces, including peacekeeping forces; defence ministries and other government agencies engaged in defence projects; paramilitary forces, when judged to be trained and equipped for military operations; and military space activities. I also use military spending as a share of government expenditure as an alternative dependent variable in the main specifications. The data is also from SIPRI and it covers the period from 1988 until 2020(Table 1).

Table 1. Descriptive statistics.

Variable	Obs.	Mean	Std. Dev.	Min.	Max.
Military burden	7477	0.028	0.029	2.97 10^−^7	1.173
Conflict	12,283	0.135	0.341	0	1
Military spending as a share of government expenditure	4029	0.076	0.065	0.000	0.574
Political accountability	10,824	0.55	0.31	0.025	0.982
Vertical accountability	10,824	0.592	0.282	0.059	0.965
Diagonal accountability	10,824	0.55	0.317	0.016	0.982
Horizontal accountability	10,824	0.52	0.308	0.014	0.99
Clientelism	10,824	0.497	0.264	0.018	0.984
Reg. support group size	10,824	2.9	1.03	0.000	3.968
GDP per capita	7997	10,420.81	15,912.36	132.077	116,232.8
Oil rents (% of GDP)	6865	4.324	10.533	0.000	87.369
Population	10,009	3.13 10^7	1.15 10^8	41,700	1.40 10^9
Tax effort (% of GDP)	3694	17.28	8.025	0.000	149.283
Trade (% of GDP)	7475	73.354	48.285	0.020	437.326
Total debt service (% of GDP)	4474	4.864	5.554	0.000	102.220
Political accountability stock (δ 0.99)	11,534	0.141	0.195	0.001	0.819
Political accountability stock (δ0.95)	11,534	0.069	0.083	0.001	0.797
Political accountability stock (δ 0.90)	11,534	0.043	0.049	0.001	0.763

My main variable of interest is accountability index (v2x_accountability_osp) (Sigman and Lindberg 2019) and it is embedded into the Varieties of Democracy dataset (Coppedge et al. 2021). It measures to what extent is the ideal of government accountability achieved, that is, if the government’s use of political power is constrained through requirements for justification for its actions and potential sanctions (Lührmann, Marquardt, and Mechkova 2020b). I will use this variable to test my first hypothesis.

The accountability index is an aggregate measure composed of three sub-indices, namely horizontal (v2x_horacc_osp), vertical (v2x_veracc_osp), and diagonal accountability (v2x_diagacc_osp) (Coppedge et al. 2021). Horizontal accountability concerns the power of state institutions to oversee the government by demanding information, questioning, and punishing improper behavior. This form of accountability ensures checks between institutions and prevents the abuse of power. Vertical accountability measures the extent to which citizens have the power to hold the government accountable through formal political participation, such as being able to freely organize in political parties, and to participate in free and fair elections, including for the chief executive. Diagonal accountability covers a range of actions and mechanisms that citizens, civil society organizations, and an independent media can use to hold the government accountable. These mechanisms include using informal tools such as social mobilization and investigative journalism to enhance vertical and horizontal accountabilities. I will regress military burden against each of these components to address hypotheses 2 to 4.

Figure 1 shows the correlation between all measures of accountability in 2015. It also includes military burden as a color. The darkest observations – those with higher military burden – are closer to the origin of the axes, indicating that lower levels of political accountability are correlated with higher military spending as a share of GDP. It is also evident that accountability measures are correlated. This may create a problem to discern their effects independently. Of course, all three sub-indices are strongly correlated with the accountability index because they are part of its composition. The correlation between vertical, horizontal, and diagonal accountability is noticeably weaker, but still significant. I will deal with this issue regressing each component separately.

Figure 1. Correlation of accountability measures and military burden in 2015.

The covariates ($X_{it}$) follow much of what has been done before in previous studies, such as gross domestic product per capita, international trade and population size, all from the World Development Indicators (WDI). As sources of revenue, I include taxation (Rota 2016) and oil rents (Perlo-freeman and Brauner 2012), both measured as a share of GDP. I also use the Uppsala Conflict Data Programme and International Peace Research Institute Oslo (UCDP/PRIO) database (Gleditsch et al. 2002) on armed conflict. This variable is coded as a dummy where 1 marks conflict years and 0 peaceful years.

I include possible confounding factors to ascertain the robustness of my results. In tandem with Bueno de Mesquita et al. (2003, 2017), I control for winning coalition size. Yet, I do not use the original measures proposed in Bueno de Mesquita et al. (2003). Gallagher and Hanson (2015) provide a comprehensive discussion on the weaknesses of Bueno de Mesquita et al. (2003) measures of winning coalition size. Instead, I use size of the regimes support coalition, i.e. groups that are supportive of the regime, and, if it/they were to retract support, would substantially increase the chance that the regime would lose power (Coppedge et al. 2021, 132). Regime support coalition measures how large is the percentage share of the domestic adult population that belongs to the political regime’s supporting groups.

A second confounding factor I include is the political power of the military. Acemoglu, Ticchi and Vindigni (2010) have theorized how the de facto power of the military affects the allocation of resources even if a de jure democratic regime is in place. Besley and Robinson (2010) reach a similar conclusion theorizing about the consequences of a weak civilian control over the military. To account for the political power of the military, I will use V-Dem’s military dimension of executive power (v2x_ex_military). This variable measures the extent to which the appointment or dismissal of the executive, or the head of state, is based on the threat or actual use of force. Additionally, I include V-Dem’s clientelism index to control for the exchange of goods and services for political support (Sigman and Lindberg 2017).

Results

Main Specifications

Table 2 presents the results of my main specifications – based on a Hausman test, I only report results from models with fixed effects. All of the specifications show negative and statistically significant coefficients for the association between different types of accountability and military spending. It is noteworthy that time seems to exert an important effect: the year dummies in all models become statistically significant from the 1990s onward, pointing to the effect of the end of the Cold War on military burden.

Table 2. Results of main specifications.

	Dependent variable: log of military spending as a share of GDP
	(1)	(2)	(3)	(4)
Political accountability	−0.242***
	(0.078)
Horizontal accountability		−0.184**
		(0.070)
Vertical accountability			−0.164**
			(0.077)
Diagonal accountability				−0.267***
				(0.078)
Conflict	0.064**	0.066**	0.065**	0.059*
	(0.032)	(0.032)	(0.032)	(0.032)
GDP per capita	0.037	0.032	0.037	0.041
	(0.096)	(0.096)	(0.095)	(0.097)
Trade (% of GDP)	−0.0006	−0.0006	−0.0006	−0.0006
	(0.0007)	(0.0007)	(0.0007)	(0.0007)
Tax (% of GDP)	0.008**	0.008**	0.008**	0.008**
	(0.004)	(0.004)	(0.004)	(0.004)
Oil rents (% of GDP)	−0.007	−0.001	−0.0008	−0.0005
	(0.004)	(0.004)	(0.004)	(0.004)
Population	−0.779**	−0.840**	−0.826**	−0.763**
	(0.345)	(0.338)	(0.343)	(0.346)
Country fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Country-specific time trends	Yes	Yes	Yes	Yes
Observations	3289	3289	3289	3289
Time period	1973-2019	1973-2019	1973-2019	1973-2019
R^2 within	0.727	0.726	0.725	0.727

*p < 0.1; **p < 0.05; ***p < 0.01

Model 1 uses the main measure of political accountability and finds that if a country with virtually no constraints on the incumbent reforms its political system to impose strong mechanisms of political accountability, military spending as a share of GDP is expected to fall as much as 21%. That is, if a country moves from the lowest possible value of political accountability to the highest. This finding largely corroborates what has been previously found in the literature and confirms my first hypothesis.

Models 2 to 4 unbundle political accountability into its smaller components to ascertain which particular types of accountability are the most strongly associated with lower military spending. A one decimal in horizontal accountability is related to a 1.7% reduction in military spending as a share of GDP. This finding points to the importance of Parliaments, for example, in the military budgeting processes. It is reasonable to assume that as Parliaments exercise oversight over the military budget, wasteful spending is reduced. Vertical accountability also seems to contribute to reducing military spending. For every decimal increase in the vertical accountability index, military burden is expected to fall by 1.5%. The expectation that unwanted increases in military spending will be sanctioned by voters seems to exert a constraining effect on incumbents.

The coefficient of diagonal accountability is by far the largest, and it suggests that the engagement of civil society in political processes can have a significant effect on military spending. Societies with a free press and an active civil society can spend as much as 23% less compared to repressive societies, with no freedom of press and public engagement. This finding sheds light onto an unexplored mechanism hitherto. Whereas the literature has primarily focused on horizontal and vertical constraints, the role of diagonal forms of accountability has gone largely unaddressed. This finding supports the idea that popular pressure, akin to the example given in the introduction about Brazil, is an effective instrument to assure that military spending is not used in exchange for political support but rather used according to society’s demands.

Robustness

To assess the sensitivity of my results, I include another measurement of military spending as well as key confounders that could provide alternative explanations for my findings (Table 3). Models 5 to 8 replicate the main specifications but with military spending as a share of government expenditure as a dependent variable (Töngür, Hsu, and Elveren 2015; Do 2021). I find that results are not dependent on measurement: all types of political accountability remain negative and statistically significant, and diagonal accountability still has the largest coefficient. As a matter of fact, all four coefficients are much larger than those obtained from models 1 to 4, suggesting that the effect of political accountability is even stronger on government expenditures.

Table 3. Results – robustness tests – alternative measure of military spending.

	Dependent variable: log of military spending as a share of government expenditure
	(5)	(6)	(7)	(8)
Political accountability	−0.594***
	(0.202)
Horizontal accountability		−0.355*
		(0.180)
Vertical accountability			−0.390***
			(0.128)
Diagonal accountability				−0.588***
				(0.199)
Conflict	0.077**	0.082**	0.076*	0.071**
	(0.035)	(0.037)	(0.038)	(0.034)
GDP per capita	0.137	0.141	0.185	0.133
	(0.154)	(0.158)	(0.157)	(0.155)
Trade (% of GDP)	0.0004	0.0002	0.0004	0.0004
	(0.0008)	(0.0008)	(0.0008)	(0.0008)
Tax (% of GDP)	0.0005	0.0003	0.0006	0.0003
	(0.004)	(0.004)	(0.004)	(0.004)
Oil rents (% of GDP)	−0.002	−0.002	−0.001	−0.001
	(0.003)	(0.003)	(0.003)	(0.003)
Population	−0.641**	−0.771	−0.613*	−0.676
	(0.757)	(0.748)	(0.777)	(0.743)
Country fixed effects	Yes	Yes	Yes	Yes
Time fixed effects	Yes	Yes	Yes	Yes
Country-specific time trends	Yes	Yes	Yes	Yes
Observations	2424	2424	2424	2424
Time period	1989-2019	1989-2019	1989-2019	1989-2019
R^2 within	0.576	0.571	0.578	0.577

*p < 0.1; **p < 0.05; ***p < 0.01

The results so far can be threatened by several confounders. I explore their potential effects in Table 4. The first one is the size of winning coalitions. As briefly discussed in the literature review, Selectorate Theory posits that the size of a winning coalition largely determines the expenditure choices of incumbents. Formal aspects of political regimes are thus secondary to the symbiotic relationship between leaders and their support base. However, as Bueno de Mesquita et al. (2003) points out, democracies do have larger winning coalitions. Because membership in a winning coalition is open to any citizen, the incumbent must consider the preferences of the majority. This correlation between institutional aspects of democracy, such as its various accountability mechanisms, and the size of winning coalitions creates a problem of specification because it is difficult to isolate the effects of each. Model 9 disentangles their respective effects and finds that political accountability remains statistically significant, while the regime support group size is not. This finding suggests that while winning coalitions might be important in determining policy decisions, formal institutions still bear a significant weight.

Table 4. Results – robustness tests – confounders and outliers.

	Dependent variable: log of military burden
	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)
Political accountability	−0.275***	−0.175*	−0.253***	−0.154*	−0.069	−0.241***	−0.239***	−0.191***
	(0.099)	(0.093)	(0.090)	(0.088)	(0.081)	(0.078)	(0.078)	(0.071)
Regime support group size	0.021
	(0.033)
Military dimension index		0.148*
		(0.085)
Clientelism			−0.047
			(0.266)
Debt service (% of GDP)				−0.001
				(0.003)
Lag of military burden					0.958***
					(0.067)
Conflict	0.064**	0.062**	0.064*	0.059	0.038**	0.064**	0.073**	0.069**
	(0.032)	(0.031)	(0.032)	(0.039)	(0.017)	(0.032)	(0.032)	(0.029)
GDP per capita	0.041	0.042	0.037	0.064	0.007	0.036	0.066	0.054
	(0.095)	(0.097)	(0.096)	(0.132)	(0.010)	(0.096)	(0.089)	0.081
Trade (% of GDP)	−0.0006	−0.0006	−0.0006	−0.001	−0.000	−0.0006	−0.0006	−0.0004
	(0.0007)	(0.0007)	(0.0007)	(0.001)	(0.0001)	(0.0006)	(0.0006)	0.0006
Tax (% of GDP)	0.008*	0.008*	0.008*	0.019**	0.000	0.008*	0.006*	0.005
	(0.004)	(0.004)	(0.004)	(0.008)	(0.0009)	(0.004)	(0.004)	(0.004)
Oil rents (% of GDP)	−0.0005	−0.0005	−0.0005	−0.004	−0.0002	−0.0007	−0.0005	−0.001
	(0.004)	(0.004)	(0.004)	(0.005)	(0.001)	(0.004)	(0.003)	(0.003)
Population	−0.790**	−0.748**	−0.775**	−1.338	0.0000	−0.789**	−0.495	−0.274
	(0.345)	(0.333)	(0.345)	(0.863)	(0.004)	(0.345)	(0.313)	(0.310)
country fixed effects	Yes	Yes	Yes	Yes	-	Yes	Yes	Yes
time fixed effects	Yes	Yes	Yes	Yes	-	Yes	Yes	Yes
country-specific time trends	Yes	Yes	Yes	Yes	-	Yes	Yes	Yes
Observations	3289	3289	3289	1655	3263	3289	3289	3289
Time period	1973-2019	1973-2019	1973-2019	1973-2019	-	1973-2019	1973-2019	1973-2019
R^2 within	0.727	0.727	0.727	0.657		0.727	0.730	0.731
AR(1)					0.000
AR(2)					0.055

*p < 0.1; **p < 0.05; ***p < 0.01

A second confounder is the political power of the military, which draws heavily from Acemoglu, Ticchi and Vindigni (2010) argument on the role of the military in democracies. The basic intuition is that if a democracy inherits a powerful military from a previous nondemocratic regime, the incumbent faces a choice between making concessions to the military in the form of ‘efficiency wages’ – like higher military spending – or being threatened by a coup. Besley and Robinson (2010) make a similar argument about the consequences of a weak civilian control over the military. Leon (2014), Albrecht and Eibl (2018) and Bove and Nisticò (2014, 2014b) provide empirical support for these theories. Model 10 accounts for the political power of the military and finds that if the armed forces can appoint or dismiss the incumbent, the military burden is expected to increase by 16%. Even adjusting for the variation explained by the political power of the military, political accountability is still statistically significant.

Model 11 investigates the role of clientelism. Clientelistic practices can coexist with accountability institutions and are to a large extent the result of the latter’s weaknesses, but they should not be conflated. The directionality of V-Dem’s clientelism index is inverse to the others, which means that lower scores indicate more clientelism and higher scores less. Akin to regime support group size, clientelism has no significant effect. Political accountability, on the other hand, has a large and statistically significant coefficient.

I also included debt service as a share of GDP in model 12 an alternative source of funds to finance military spending. There has been some evidence on the relationship between debt and military spending (Kollias, Manolas, and Paleologou 2004; DiGiuseppe 2015). Alexander and Robert (2013), for instance, find that military burden is an important determinant of public debt among high-income members of the North Atlantic Treaty Organization (NATO). Ahmed (2012) observed something similar among African countries. Political accountability remains above under the threshold for statistical significance, but debt does not seem to explain variations in military burden in this specification.

Expenditure is path dependent. Choices made in previous years are likely to weigh heavily on subsequent ones. If unaccounted for, this autocorrelation will appear in the error term, leading to inefficient estimators. For this reason, I include a time lag of military spending. This inclusion, however, leads to a correlation between the lag and the error term that bias the results (Nickell 1981). To account for that, I use a System-Generalized Method of Moments (System-GMM) estimator in model 13 (Arellano and Bover 1995, Roodman 2009). The suitability of System-GMM is confirmed by the results of an AR(1) test.

The coefficient for political accountability loses its statistical significance in the short run with the inclusion of an autoregressive term. Achen (2000) documents this effect quite well: autoregressive terms often take on very strongly significant coefficients that, although improve the overall fit of the model, obfuscate the effects of other variables. Achen (2000) shows that autoregressive terms can be dominant in regressions even when they have little or no real explanatory power. This happens when high serial correlation and heavy trending in the exogenous variables are combined. Wilkins (2018) offers the counterpoint that the inclusion of lags can actually improve the accuracy of estimators. This debate, at the very least, shows that the interpretation of the loss of statistical significance of political accountability with the inclusion of the lag of military spending should not be overstated. Moreover, while the coefficient for political accountability loses its statistical significance only in the short run; in the long run – where political accountability and military burden reach a steady-state equilibrium – the coefficient jumps to −1.027 and is significant at 1%.

The effect of outliers driving the results is a relevant concern. It may well be the case that my results are driven by a few observations of exceptionally high military burden coupled with low scores of political accountability. A potential case would be Oman, which had the highest military spending as a share of GDP in the world in 2019 at 8.6% and a 0.413 accountability index score. Observations with higher values of military burden were also evident in Figure 1. In models 14 to 16, I run the benchmark specification (model 1) once again but after winsorizing military spending at 1%, 5% and 10% to limit extreme values. This does not affect the size, direction or statistical significance of the coefficients.

In Table 5, I split the sample into relevant groups to explore heterogeneity in the results. First, I split the sample into income groups (Clements, Gupta, and Khamidova 2019) using the World Bank classification. In model 17, I limit the sample to high and higher-middle-income countries and find that political accountability has a large and statistically significant coefficient of −0.322. The association between political accountability and military burden is even stronger among low and lower-middle-income countries (model 18).

Table 5. Results – Robustness tests – heterogeneity.

	Dependent variable: log of military spending as a share of GDP
	High income (17)	Low income (18)	Closed autocracy (19)	Minimally constrained regimes (20)
Political accountability	−0.322***	−0.415**	−0.345	−0.211*
	(0.111)	(0.167)	(0.325)	(0.118)
Conflict	0.065	0.068	−0.026	0.044
	(0.043)	(0.043)	(0.073)	(0.030)
GDP per capita	−0.147	0.355	−0.170	0.066
	(0.098)	(0.236)	(0.424)	(0.108)
Trade (% of GDP)	−0.0001	0.0002	0.005	−0.0005
	(0.0006)	(0.001)	(0.002)	(0.0007)
Tax (% of GDP)	0.004	0.019*	−0.023	0.008
	(0.003)	(0.010)	(0.012)	(0.004)
Oil rents (% of GDP)	0.003	−0.006	0.011	−0.006
	(0.007)	(0.006)	(0.008)	(0.005)
Population	−0.125	−0.870	−2.121	−1.013
	(0.397)	(1.052)	(0.481)	(0.384)
country fixed effects	Yes	Yes	Yes	Yes
time fixed effects	Yes	Yes	Yes	Yes
country-specific time trends	Yes	Yes	Yes	Yes
Observations	2167	1122	264	3025
Time period	1973-2019	1988-2019	1973-2019	1973-2019
R^2 within	0.803	0.581	0.766	0.726

*p < 0.1; **p < 0.05; ***p < 0.01

In model 19 and 20, I split the sample into political regimes based on the Regimes of the World (ROW) dataset (Lührmann, Tannenberg, and Lindberg 2018). I group regimes based on the institutional constraints: first, I replicate the baseline specification using only closed autocracies. In this type of regime, the chief executive is either not subjected to elections or there is no meaningful, de-facto competition in elections (Lührmann, Tannenberg, and Lindberg 2018). As such, incumbents are virtually unconstrained. In closed autocracies, political accountability does not seem to lead to a lower military burden – while the coefficient is negative, it is not statistically significant. This suggests that improvements in political accountability among closed autocracies may be too incipient to yield results in terms of reducing military spending. In minimally constrained regimes – electoral autocracies, electoral democracies and liberal democracies – political accountability is statistically significant at the 10% threshold.

Endogeneity

The results so far lend support to the relationship between political accountability and military spending. However, the issue of endogeneity remains. The argument put forward so far posits that political accountability constrains military burden. Yet it could well be the case that an unobserved factor is driving both variables, or even that military burden weakens political accountability. If military spending is used in exchange for political support, this can in turn strengthen the incumbent against any institutional constraints. Disentangling political accountability and military spending requires the use of a source of variation for political accountability that is exogenous. For this purpose, I leverage the experience a given country has with accountability institutions.

Decisions are historically informed. The theoretical proposition that political accountability affects the decision to increase military spending rests to a large extent upon the incumbent’s own perception about the effectiveness of such mechanisms. For its part, this assessment is based on past experiences. The threat of sanctions from these accountability mechanisms needs to be credible, and their historical record creates an expectation that they may or may not work. For instance, it is conceivable that the enlargement of the voting franchise in a given year strengthens electoral accountability: if the incumbent increases military spending beyond what is socially accepted, it will likely face sanctions in the form of non-re-election, for example. Yet the decision of the incumbent to follow through with its original intent to increase military spending is largely based on past experiences with electoral sanctions. If sanctions did not work in the past, there may be an expectation that they will not work now or in the future, despite the recent improvement in electoral accountability. The incumbent’s attitude towards military spending is shaped by experience, and this cognition is resilient to sudden shifts in political institutions (Simpser, Slater, and Wittenberg 2018).

To capture the weight of the past, I build on Edgell et al. (2020) to calculate a country’s political accountability stock ($pas$). This variable measures a country’s accumulated experience with political accountability. The variable is the weighted sum of V-Dem’s accountability index for previous years until $t-1$ with a baseline depreciation rate $\delta$, as seen in equation 2(2) $\mathrm{p}\mathrm{a}{\mathrm{s}}_{\mathrm{t}}={\delta }^{\prime }\mathrm{p}\mathrm{a}{\mathrm{s}}_{\mathrm{t}-1}+\mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{\_}\mathrm{o}\mathrm{s}{\mathrm{p}}_{\mathrm{t}-1}$(2) . Edgell et al. (2020) sets $\delta$ at 1%, as done before in Gerring et al. (2005). However, Gerring et al. (2005) admits that the 1% depreciation rate is arbitrary and based on trial and error. I adopt the same strategy and iterate the same specification with measures of political accountability stock with different baseline depreciation rates, namely at 1%, 5% and 10%. The political accountability stock covers the period between 1900 and 2019. Following Edgell et al. (2020) and Persson and Tabellini (2009), I normalize the values between 0 and 1 and assume that cases enter the sample with zero accumulated stock in the first year they appear in the dataset, i.e. $pa{s}_0=0$.

One possible criticism of the aforementioned design is that a constant baseline rate of depreciation $\delta$, for all countries at all times, is too unrealistic. Idiosyncrasies and shifts in the political landscape certainly shape the way the past weighs on the present. To address this issue, I discount the rate of depreciation based on the political power of the military at $t-1$. This gives ${\delta }^{\prime }$ in equation 3(3) ${\delta }^{\prime }=\left[\delta -\left(\delta \ast \mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}{\mathrm{y}}_{\mathrm{t}-1}\right)\right]$(3) . The intuition behind this is that if the military can appoint and overthrow the government, the incumbent will pay ‘efficiency wages’ (Acemoglu et al. 2010; Besley and Robinson 2010) to the military to remain in power, heavily discounting the weight of the past. Military spending then becomes a tool of political survival (Bueno de Mesquita et al. 2003).

Bernhard and Edgell (2019) have done something similar before. In their study on the role of social organization and mobilization on democracy, Bernhard and Edgell use a stock of civil society organizational capacity that depreciates at a rate given by the level of state repression. I use something similar: I discount the depreciation rate with V-Dem-s military dimension index at $t-1$ ($\mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{e}\mathrm{x}\mathrm{\_}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}{\mathrm{y}}_{\mathrm{t}-1}$). If $\mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{e}\mathrm{x}\mathrm{\_}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}{\mathrm{y}}_{\mathrm{t}-1}$ is 0, then the political accountability stock depreciates at the baseline levels of 1%, 5% or 10%, because $\delta$ is then equal to $\delta$’. If $\mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{e}\mathrm{x}\mathrm{\_}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}{\mathrm{y}}_{\mathrm{t}-1}$ is 1 – the military both appointed and can remove the incumbent by force – the political accountability stock bears no weight on the incumbent’s decision to allocate resources to the military. Instead, political survival will be the driving force shaping allocations.

(2) $\mathrm{p}\mathrm{a}{\mathrm{s}}_{\mathrm{t}}={\delta }^{\prime }\mathrm{p}\mathrm{a}{\mathrm{s}}_{\mathrm{t}-1}+\mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{a}\mathrm{c}\mathrm{c}\mathrm{o}\mathrm{u}\mathrm{n}\mathrm{t}\mathrm{a}\mathrm{b}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{y}\mathrm{\_}\mathrm{o}\mathrm{s}{\mathrm{p}}_{\mathrm{t}-1}$(2)

(3) ${\delta }^{\prime }=\left[\delta -\left(\delta \ast \mathrm{v}2\mathrm{x}\mathrm{\_}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{i}\mathrm{t}\mathrm{a}\mathrm{r}{\mathrm{y}}_{\mathrm{t}-1}\right)\right]$(3)

I use the political accountability stock as an instrumental variable to political accountability. Past experiences with accountability largely determine political accountability at $t$. On the other hand, military spending at $t$ cannot affect a country’s historical record of political accountability. Under this assumption, Table 6 shows the results of a two-stage least squares (2SLS) estimation with fixed effects and robust standard errors using the political accountability stock with different baseline depreciation rates. I report the results of both the first and second stages.

Table 6. Results of two-stage least-squares regressions.

	Dependent variable: political accountability
	(21)	(22)	(23)
Political accountability stock (δ 1%)	1.347***
	(0.097)
Political accountability stock (δ 5%)		2.346***
		(0.131)
Political accountability stock (δ 10%)			4.447***
			(0.225)
Conflict	−0.045***	−0.036***	−0.031***
	(0.010)	(0.009)	(0.009)
GDP per capita	−0.012	0. 013	0.023***
	(0.010)	(0.009)	(0.008)
Trade (% of GDP)	−6.15 10^−7	−0.0001**	0.0002***
	(0.000)	(0.000)	0.000
Tax (% of GDP)	0.0005	0.0004	0.0003
	(0.0005)	(0.0005)	(0.0005)
Oil rents (% of GDP)	−0.0006	−0.0007	−0.0008
	(0.0006)	(0.0006)	(0.0006)
Population	0.328***	0.320***	0.300***
	(0.019)	(0.017)	(0.016)
	Second stage: log of military spending as a share of GDP
	(21)	(22)	(23)
Political accountabilityt	−0.690***	−0.108	−0.104
	(0.137)	(0.137)	(0.087)
Conflict	0.138***	0.167***	0.167***
	(0.025)	(0.025)	(0.025)
GDP per capita	−0.210***	−0.234***	−0.235***
	(0.029)	(0.028)	(0.028)
Trade (% of GDP)	−0.002***	−0.002***	−0.002***
	(0.0002)	(0.0002)	(0.0002)
Tax (% of GDP)	0.011***	0.010***	0.010***
	(0.002)	(0.002)	(0.002)
Oil rents (% of GDP)	0.004	0.005*	0.005*
	(0.002)	(0.003)	(0.003)
Population	−0.608***	−0.763***	−0.765***
	(0.067)	(0.062)	(0.060)
Observations	3244	3244	3244
Centered R^2	0.377	0.389	0.389
Uncentered R^2	0.377	0.389	0.389

*p < 0.1; **p < 0.05; ***p < 0.01

Model 21 uses the political accountability stock at a 1% baseline depreciation rate as an instrument. It is the only 2SLS with a statistically significant coefficient for political accountability. It shows a local average treatment effect of −0.690. This coefficient is considerably larger than those of regular OLS estimations. For instance, the baseline specification in model 1 resulted in a − 0.242 coefficient for political accountability, a little over a third of the 2SLS. The instrumental variable design allows me to make a stronger statement about the relationship between political accountability and military spending. Because the political accountability measure used in the second-stage regression consists of only the variation explained by political accountability stock, causality can be inferred. The results from model 21 suggest that a one decimal increase in a country’s political accountability causes a 4.9% reduction in military burden. This is much larger than the coefficient I found in model 1. This finding suggests that endogeneity leads to an underestimation of the causal effect of political accountability on military spending.

I perform multiple tests on the strength of my instruments, as well as if my specifications are weakly or under-identified. A Kleibergen-Paap rk Wald test shows that the F statistic of the instrument used in model 21 is 189.808, well above the critical value. This confirms the explanatory power of the instrument. Subsequent tests also confirm that the equations are not under-identified, and that endogeneity is indeed an issue to be dealt with.

Conclusion

Research on the relationship between political institutions and military spending has advanced a great deal in the last two decades. The finding that democracies spend less than autocracies turned out to be robust across studies. Explanations for such a finding, however, diverged, putting forward different mechanisms. Empirical work, on the other hand, has not yet tested these competing explanations satisfactorily. As a result, our knowledge on the imbrication of political institutions and military spending still has much to advance.

In this article, I outline a theoretical argument that builds upon much of what has been done before, both in the literature on military spending and that on Democratic Peace. Like previous work, I stress the centrality of political accountability in reducing military spending. Yet I specify clear mechanisms that are subsequently tested. To test the hypotheses, I run a series of specifications to explore the different channels linking political accountability and military spending, which are subsequently complemented with two-stage least squares (2SLS) estimations that instrumentalize a country’s political accountability stock. In that sense, one of the contributions of this paper is this to address a rather common mismatch between theoretical claims and empirical design. This was made possible with recent developments in the measurement of political regimes. The V-Dem dataset not only has a theory-based measurement method of political regimes and institutions, but it also has an unprecedented level of disaggregation. The possibility of unpacking measures into its smaller components allows researchers to test as well as refine extant theories.

Political accountability shows a robust negative correlation with military spending as a share of GDP. The next step was to unbundle political accountability into its components, namely horizontal, vertical, and diagonal accountability. Unexpectedly, I found that diagonal accountability has the strongest association with lower military spending. This finding is novel, as this mechanism has been largely ignored in the literature. As the example of Brazil given in the introduction illustrates, the engagement of civil society and an independent media can act as a powerful mechanism to signal dissatisfaction with increasing military spending. Another important contribution of this paper is the use of political accountability stock as an instrument to tackle endogeneity and infer causality. In doing so, I found evidence suggesting that political accountability causes military burden to fall, and the effect is bigger than previously assumed.

As a broader sub-field of inquiry, research on the relationship between military spending and political institutions can benefit from a focus on two avenues of investigation. Firstly, exploring the heterogeneity across cases can lead to interesting findings on the limits of this relationship. Large-N studies are useful to provide average effects, however this comes at the cost of concealing how different cases may be from each other. For example, identifying deviant cases would be particularly helpful in refining theoretical statements or even generating new hypotheses (Levy 2008). Second, the area has much to benefit from qualitative case studies. Methods such as Process Tracing can potentially outline causal chains linking political institutions to military spending in ways large-N statistical studies can hardly do (Beach 2016; Gerring 2006). In combination, qualitative case studies exploring country-specific features can lead to novel results, significantly improving our understanding.

Disclaimer

The views and opinions expressed in this paper are those of the author alone and do not necessarily reflect the views of any institution.

Acknowledgments

The author is grateful for comments from two anonymous referees, Nan Tian, Lucie Béraud-Sudreau, Raphael Camargo Lima and assistance from Ana Cecília Ribeiro Guimarães.

Disclosure statement

No potential conflict of interest was reported by the author(s).