Advertisement

International Journal of Economic Policy Studies

, Volume 13, Issue 1, pp 119–146 | Cite as

Conscription and the developing countries

  • Nguyen Dinh Tuan Vuong
  • David FlathEmail author
Research article
  • 33 Downloads

Abstract

Many countries, mostly ones with developing economies, still have conscription. This study explains why, based on the costs of different recruitment systems. A conscription system can be costly to set up, and misallocates labor compared to all-volunteer recruitment, but entails a smaller wage bill. In a developing country with a highly distorting tax system, the burden of taxes needed to finance the higher wage bill of an all-volunteer military can be larger than the burdens that are unique to a conscription system. The highly burdensome taxes that the government of a developing country with a large informal sector must rely upon mean that its variable cost of staffing a government labor force or army by an all-volunteer system is greater than that of a developed country that can levy less distorting taxes. The minimum force size for which conscription has lower cost than voluntary recruitment is higher, the greater the tax distortion. Furthermore, to economize on training cost, a country with a more distorting tax system is more likely to have a lottery, with longer service obligation than if it did not have a lottery. The upshot is that a country with a distorting tax system is more likely to have conscription, and if it does have conscription, is more likely to have a lottery system, with longer service obligation than if it did not have a lottery and drafted an entire age cohort. If our reasoning is correct, countries with less efficient tax systems, wishing to recruit larger fractions of their labor forces, would be more likely to have conscription with longer service obligations. We show that such patterns are indeed present in the data—a cross section of 104 countries observed around 2012. In this way, we explain why many developing countries continue to impose conscription even as the developed countries are abandoning it.

Keywords

Conscription Tax systems in developing countries 
JEL classification H20 H21 O17 

Introduction

This is an original contribution on the economics of military conscription, with special attention to the design of tax systems of developing countries. The persistence of conscription in the developing countries, even though the developed countries seem to have one-by-one been abandoning it in favor of all-volunteer systems, has been remarked by many but never yet fully explained. (See however: Hall [11], Economist magazine [7], and Tarabar and Hall [25]). We offer an explanation, based on the cross-country differences in cost of a conscription system compared to an all-volunteer recruitment system, as represented in the model of Mulligan and Shleifer [17]. We will argue that within the Mulligan and Shleifer model, the less efficient tax systems of developing countries incline them to favor conscription with longer service obligations. This is a line of argument previously advanced by Ross [21] but not fully confirmed by his statistical estimates. The data related to the tax systems of the different countries now enable a more complete investigation. We estimate a Tobit regression equation with cross-country data that supports the claim that countries whose abilities to collect taxes are constrained by their citizens’ non-use of banks, are more likely to have conscription with longer required lengths of service.

Figure 1 shows the inverse relationship between GDP per capita and the length, in months, of conscript service obligation. For many of the countries, the length of obligation is zero; they are not drafting anyone. Countries with higher income are less likely to have military conscription. As shown in Table 1, about one-third of the high-income countries have conscription, but more than half of the low-income and middle-income countries have conscription.
Fig. 1

Length of conscript service obligation and GDP per capita

Table 1

Conscription across countries, by per-capita income, 2012

Sources: Conscription and length of service: Authors’ calculation based on CIA Factbook, Military Balance, CIA Factbook, War Resisters’ International, and Toronto [26]. Country group: World Bank, World Development Indicators [28]

Country group

Number of countries

Number of countries with conscription (and  % of country group)

Required length of service in months, averaged across countries with conscription

Low Income

20

12 (60.0%)

20.4

Lower Middle Income

31

16 (51.6%)

18.9

Upper Middle Income

30

19 (63.3%)

17.6

High Income

35

10 (28.6%)

12.5

Conscription is not a strictly dichotomous variable. Shorter length of service obligation in a universal conscription system means less reliance on conscription. It means that the number of conscripts in service at any given time are fewer. Exemptions from conscription, for example student deferments, represent a margin for reducing the reliance on conscription. Unfortunately we do not have adequate information about the deferment and exemption practices of the different countries that use conscription, and so our empirical analysis will focus on length of service obligation. The high-income countries that do have conscription tend to require fewer months of service than do low and middle income countries. This is quite evident in Fig. 1. It is also shown in Table 1. The developing countries are more likely than the high-income countries to use conscription, and to require more months of service by each draftee. The question we aim to answer is, why?

The model of Mulligan and Shleifer [17] is simple yet general enough to accommodate key differences between rich countries and poor ones. As will become evident, our argument rests within their model. A conscription system can be costly to set up, and misallocates labor compared to an all-volunteer system, but a voluntary enlistment system entails a higher wage bill. The taxes needed to fund the higher wage bill of a voluntary enlistment system are more burdensome, the less efficient the tax system. For a poor country with an inefficient tax system, the tax burden of a voluntary enlistment system can be greater than the tax burden of a conscription system. That is the crux of our explanation as to why poor countries are more likely to rely on conscription. It is also the case that the particular distortions of military conscription, arising from the misallocation of labor, are likely to be small in a developing country where labor skills are relatively homogeneous compared to most developed countries. But we are unable to confirm this with data. We do find empirical support for the link between the inefficient tax systems of poor countries and their greater reliance on conscription—presumably because it makes the higher wage bill of a voluntarily recruited force more burdensome for them.

Poor countries collect taxes where they can, which often means the taxes are highly distorting and have high excess burdens. In the rich countries, taxes are broad-based and so entail smaller excess burdens than in the poor countries. The rich countries have gradually come to rely less on distorting tax sources including customs duties, government monopolies, and inflation of the currency. Military conscription—a tax in-kind on labor—can be added to the list of distorting taxes that persist mostly in the developing countries, a reflection of the constraints facing developing countries. Simply put, in a developing country, where the informal sector of the economy is large, taxes that might be less distorting are easier to evade and harder to collect.

A recent paper by Tarabar and Hall [25], which came to our attention only after we completed our own analysis, overlaps a bit with ours, but also differs from ours. They estimate a single equation with panel data at 5-year intervals, 1970–2010, for about 100 countries, an equation that explains length of conscript service obligation. They argue that the long-term decline in conscription over their period of observation is related to the secular increase in primary schooling among young males, and to the changing age profiles of country populations. This is consistent with a view that the misallocation of labor caused by conscription is becoming more costly over time, prompting the high-income countries in particular to shift toward voluntary recruitment. We think this is a correct understanding of the time trend. But, our focus is on the cross-sectional variation in the recent period, and why conscription persists in the developing countries more than in the high-income countries. Our reasoning and our empirical analysis point to the less efficient tax systems of the developing countries as an important underlying factor. In our empirical estimate we find that some of the same control variables also used by Tarabar and Hall are significant. For both their study and ours, countries with British legal origin (common law) rely less on conscription, and countries with Socialist origin rely more on it. Also countries facing hostile neighbors have larger military forces and so rely more on conscription (we model this with structural equations, which Tarabar and Hall do not—they put it all in a single equation). And finally, we show that the governments of countries whose citizens and businesses use banks less—countries with large informal sectors, mostly developing countries—tend to collect less tax revenue, and to just that extent, rely more on conscription. Again, we model this with structural equations. Tarabar and Hall, following the example of Ross [21], include government consumption spending relative to GDP as a variable in their equation, on the premise that the marginal excess burden of taxes rises with size of government, but the variable is generally not significant and has the wrong sign for their interpretation. The efficiency of the tax system, the main focus of our analysis, is absent from their findings.

We begin by describing our theoretical framework and how it frames our understanding of the differing relative costs of conscription compared to voluntary enlistment. Then we present empirical estimates showing that patterns actually present in the data support our reasoning. Countries with less efficient tax systems, wishing to recruit larger fractions of their labor forces, are more likely to have conscription with longer service obligations.

Theoretical framework

Tax implications of conscription

There are two opposing ways that the fiscal resources needed to fund a conscription system differ from that of an all-volunteer system. First, conscription has specific costs of administration, especially for enforcement measures, and the costs of such administration, including set-up costs, can be large. But a second and opposing factor is that, with conscription, the wage bill is less than for an all-volunteer system. Draftees are paid a lower wage than would have been needed to elicit their voluntary enlistments. If a large fraction of the population of a set age are to be drafted, then the (mostly fixed) cost of administration and enforcement of a conscription system can be outweighed by its lower wage bill compared to an all-volunteer system of recruitment. In this case, conscription imposes a smaller drain on the fiscal resources of the government than would an all-volunteer system of recruitment. Similarly, because many of the costs of administering and enforcing a conscription system are fixed costs, if a small fraction of the eligible population is to be drafted, then conscription probably requires more fiscal resources than an all-volunteer system. This explains why the rich countries resorted to conscription during the World Wars but reverted to all-voluntary recruitment after the wars ended and they downsized their armed forces.

Besides its effect on the fiscal position of the government, conscription also differs in the nature of the excess burden it imposes compared to an all-volunteer system of recruitment. Conscription is a tax-in-kind. Its immediate burden, equal to the difference in wage actually paid and the lowest wage needed to elicit voluntary enlistment (reservation wage), falls on the draftee. But the costs of administering a conscription system and the stipends that must be paid even to draftees, require still further taxes. As described in the previous paragraph, the fiscal resources needed to maintain a conscription system can be either greater or less than are needed to maintain an all-volunteer system that recruits the same labor services. The total tax burden of a conscription system compared to an all-volunteer system equals the burden that is borne by the draftees plus the difference between the two systems in burdens of the other taxes needed to cover the wage bill and the costs of administration and enforcement (and that difference can be positive or negative—more likely to be negative the larger the government staff being recruited).

The burden that is borne by draftees arises as a misallocation of their labor. Reluctant draftees—whose reservation wages are higher than the wages they are paid once drafted—have alternatives to military service that they value more. The alternatives can be leisure, investment in education, or supplying labor in other markets. Conscription harms draftees’ professional careers by depreciating existing skills or by preempting the acquisition of new skills. Empirical estimates of the costs of conscription-induced misallocation of labor vary widely.1 The excess burden of the labor tax-in-kind that is conscription is the waste from employing the wrong persons in military service. The budding professional is plucked out of school or out of a promising job where his or her efforts are productive, and wastefully set to work peeling potatoes and mopping floors (or firing a bazooka)—tasks that another could do at lower cost.

In spite of the excess burden that is specific to a conscription system (the misallocation of labor that it induces), the total burden of taxes needed to recruit and maintain an all-volunteer force can be even greater than the total tax burden of a conscription system, particularly if the size of the force to be recruited is large. We expect conscription to be more likely in countries where its total burden is less than that of an all-volunteer system. That is the principle behind the Mulligan and Shleifer [17] model that is also the framework of this paper. We will have more to say about that model in the next section, for our aim is to extend the model to accommodate differences in tax systems across countries. We will argue that within the Mulligan and Shleifer model, differences in tax systems between developing and developed countries affect the relative likelihood that conscription would have a lower total tax burden than an all-volunteer system of recruitment. In general, the less efficient the tax system, the greater the excess burden of the higher wage bill of an all-volunteer force compared to a conscripted one. Poor countries have less efficient tax systems with greater excess burdens relative to the revenue they collect. That is why, for poor countries, voluntary recruitment systems are often more costly compared to conscription systems.

Tax systems differ across countries because the constraints on the design of a tax system matter. The costs of administering and enforcing tax rules are the important constraints, and these differ between developing and developed countries. Gordon and Li [9] describe and model the differences in tax systems between developing and developed countries. Governments of developing countries rely more on forms of taxation that are relatively distorting than do the governments of developed countries, and generally collect less tax revenue in relation to GDP than do the governments of developed countries. These two facts are obviously related to one another.

As shown in Table 2, high per-capita-income countries collect more tax revenue as a percentage of GDP than do low per-capita-income countries. Also, more of the taxes of poor countries are taxes on goods and services, or taxes on international trade, and less are taxes on personal income or wages. The governments of poor countries derive more seigniorage than do those of the rich countries—their average inflation rates are rather high. Gordon and Li argue that the inflation tax and border taxes (taxes on international trade) have high excess burdens compared to the other kinds of taxes, but are easier to collect when fewer of a nations’ citizens or businesses have bank accounts.
Table 2

Tax systems across countries, by per capita income, 2005–2013

Sources: Financial Inclusion Index: Park and Mercado [19]. Tax revenues as percentages of GDP: Prichard et al. [20]. Inflation rate: based on CPI, from World Bank, World Development Indicators [28]

Country group

Financial inclusion index

Tax revenue

Social security wage tax

Personal income tax

Corporate income tax

Tax on goods and services

Tax on international trade

Inflation

(CPI)

Low income

9.04

15.73

0.04

1.03

0.80

5.25

2.24

10.31

Lower middle income

20.79

16.42

0.21

1.81

2.27

14.54

1.58

8.60

Upper middle income

35.43

22.28

5.33

2.22

1.10

20.37

0.70

4.83

High income

71.21

40.74

8.77

8.49

2.89

8.28

0.20

2.03

Numbers of observations

Low income

16

16

11

10

16

15

15

16

Lower middle income

28

28

18

16

24

22

22

27

Upper middle income

25

25

16

14

21

21

21

24

High income

35

35

34

33

35

26

26

35

Total

104

104

79

73

96

84

84

102

In the table, financial inclusion index and inflation are simple averages across countries in each group. All others–tax revenues as percentages of GDP–are weighted averages for countries in each group with weights equal to GDP. Numbers of observations vary for each item as shown in the lower half of the table.

The first column of Table 2 shows the average value across countries in each income class of the “financial inclusion index” constructed by Park and Mercado [19]. The components of the index are: (1) the borrowers from commercial banks per 1000 adults, (2) depositors with commercial banks, per 1000 adults, (3) domestic-credit-to-GDP ratio, (4) ATMs per 100,000 adults, and (5) commercial bank branches per 100,000 adults. The index ranges from 0 to 100, with higher values implying wider access to banking services. Countries for which the financial inclusion index is lower, may be presumed to have more extensive informal economies, beyond the purview of taxing authorities. As shown in the table, this is characteristic of the developing countries, not the advanced countries.

The relatively large informal sectors of the developing countries make it easier for their citizens to evade taxes on consumption or personal income. The developed countries are better able to collect revenue from taxing personal income because more of their citizens and businesses have bank accounts and are officially registered with government authorities. That is why the governments of developed countries have largely abandoned sources of revenue with large excess burdens such as customs duties, seigniorage and government monopolies, but are still able to collect more revenue in relation to their GDP than is characteristic of the governments of developing countries. Other scholars have made a similar argument. Slemrod [23] and Besley and Persson [3] also describe the difficulty of collecting taxes in developing countries because of their large informal sectors and inadequate institutions, including banks, securities markets with listing requirements, civil bureaucracies and so on. This is a key point in our explanation why developing counties are more likely to use conscription with long required lengths of service.

The costs of different recruitment systems

Mulligan and Shleifer [17] provide a simple algebraic model for comparing the social cost—that is to say, the total tax burden—of each different system of recruitment into government (military) service. We will confine our attention to the costs of two types of recruitment system only: an all-volunteer system and a universal conscription system. A universal conscription system places an entire age cohort at risk of being drafted, possibly by lottery, but without allowing exemptions or replacements. 2 In an all-volunteer system, no one is subject to a draft.

In the Mulligan and Shleifer framework, the total cost of recruitment has both a fixed cost, \(\rho_{i}\), including cost of adoption, administration and enforcement, and a variable cost, \(\delta_{i} mNc\left( m \right)\), including any tax burden associated with the costs of maintaining the force once it is recruited.

$$total \cos t = \rho_{i} + \delta_{i} mNc\left( m \right),$$
$$i = v \left( {voluntary} \right), i = u\left( {universal\;conscription} \right),$$
where N is the total population, and m is the fraction of soldiers in the total population, so mN is the number of troops. Here, \(c(m)\) denotes the rising average opportunity cost per soldier of recruiting the fraction m of the whole population—a cost that is common to all recruitment systems—and \(\delta_{i} c\left( m \right)\) is the average variable cost for each soldier in the type i recruitment system. In this formulation, total cost of each system of recruitment is a linear function of \(mNc\left( m \right)\), an expression which is monotonically increasing in number of troops and in fraction of the population recruited.
Mulligan and Shleifer argue that a universal conscription system has significant fixed costs but that a voluntary recruitment system does not:
$$0 \approx \rho_{v} < \rho_{u}$$

They reason that conscription requires special government monitoring and enforcement. For example, every young person has to register with the government when he or she reaches a certain age. The health of all potential draftees has to be checked. Highlighting the importance of these fixed costs of establishing and operating a conscription system is an original contribution of Mulligan and Shleifer. They argue that the countries with a French heritage are likely to have a system of government administration that is better adapted to the implementation of a conscription system, compared to other countries, and in particular compared to countries with a British heritage. They argue that the countries with British colonial heritage are less likely to have conscription systems. They confirm this with empirical estimates, as do we.

Next consider the variable cost of each type of recruitment system, which is the nub of our main line of argument. The variable cost per soldier of universal conscription is generally less than that of voluntary recruitment:
$$\delta_{u} < \delta_{v} .$$

Universal conscription misallocates labor compared to an all-volunteer recruitment system but has a smaller wage bill, and the marginal difference in the wage bill between the two systems as the size of the force to be recruited is increased is likely to more than offset the marginal change in misallocation of labor under universal conscription. That is why the variable cost per soldier of universal conscription is generally less than that of voluntary recruitment.

As evident in Fig. 2, if as just argued, universal conscription has a lower average variable cost than voluntary recruitment, \(\delta_{u} < \delta_{v}\), but a higher fixed cost, \(\rho\), then universal conscription has lower total cost than voluntary recruitment only if the size of force to be recruited is large. To just that extent, a country is more likely to introduce conscription the greater the fraction of the population it wishes to recruit. As we will next explain, it is also more likely to introduce conscription, the greater the excess burden of its taxes in relation to tax revenue, and the smaller the excess burden that would be borne by draftees.
Fig. 2

In the framework of Mulligan and Shleifer [17], universal conscription has lower total cost than voluntary recruitment only if the size of force to be recruited is large. This is because conscription has a lower average variable cost than voluntary recruitment, \(\delta_{u} < \delta_{v}\), but a higher fixed cost, \(\rho\)

The highly burdensome taxes that governments in developing countries with large informal sectors must rely upon—customs duties, seigniorage, profits from government monopolies, and so on—mean that their variable cost of staffing a government labor force or army by an all-volunteer system is greater than that of a developed country that can levy less distorting taxes such as a broad-based consumption tax or personal income tax with low marginal tax rates but high yields. To put it another way, \(\delta_{v}\) in developing countries must be larger than in rich countries. But the same consideration would also increase the fixed cost of establishing a conscription system, \(\rho_{u}\). In the Mulligan and Shleifer framework, the minimum staffing requirement at which a conscription system has lower cost than an all-volunteer system is
$$mNc\left( m \right) = \frac{{\rho_{u} }}{{\delta_{v} - \delta_{u} }}$$
If the average variable cost of a voluntary system, \(\delta_{v}\), and fixed cost of establishing a universal conscription system, \(\rho_{u}\), are both increased by the same factor α > 1, then \(\frac{{\rho_{u} }}{{\delta_{v} - \delta_{u} }}\) falls, meaning that conscription is more likely to be adopted. This is shown in Figs. 3, 4.
Fig. 3

An increase in the excess burden of general taxes entails an equiproportionate increase in both variable costs and fixed costs of voluntary recruitment and conscription, and lowers the minimum size of force to be recruited for which conscription has lower total cost than voluntary recruitment

Fig. 4

The minimum size of force to be recruited for which universal conscription has a lower total cost than voluntary recruitment depends on the excess burden of general taxes. It is less, the greater the distortion

Length of service obligation and the costs of different systems

The focus of our empirical analysis is not only on which countries have conscription systems and which do not, but also on length of service obligation of conscripts. Consider now a simple extension of the Mulligan and Shleifer framework that accommodates length of service obligation. First, let us follow their lead and abstract from the effect of size of military force itself on the opportunity cost of staffing that are common to all recruitment systems, represented by their expression, \(c\left( m \right)\). Let us suppose that these costs are constant, \(c\left( m \right) = 1\), and therefore do not enter our analysis. Furthermore, in systems with conscription, let us suppose that the length of service obligation \(\gamma\) is proportionate to the number of conscripts, and let us further suppose that the number of conscripts is exogenously determined. The idea here is that the staff comprises a cadre who serve voluntarily, while the line positions are filled either by volunteers only, or by conscripts only. The number of line positions, \(\bar{N}\) say, is determined before deciding on the recruitment system. If the recruitment system is universal conscription—in which the entire age cohort is called to serve—then, to raise the set number of conscripts would require a length of service obligation, \(\bar{\gamma } = \bar{N}\), where units of measurement of γ are defined implicitly. In a conscription system with a lottery, the length of service could be adjusted along with the fraction of the age cohort that is selected, to attain the given number of conscripts while varying the length of service obligation. The given number of conscripts could be raised by selecting the fraction of the age cohort, \(\theta\), where \(0 < \theta \le 1,\) and requiring service obligation, \(\gamma\), such that \(\theta \gamma = \bar{N}\).

Now let us consider how length of service obligation relates to the costs of a conscription system. In a system with conscription, the variable cost per conscript \(\delta_{u}\) is
$$\delta_{u} = f\left( \gamma \right) + g\left( \gamma \right) + w_{u} .$$

This includes training cost \(f\) plus the cost \(g\) borne by the conscript (difference between reservation wage and actual wage \(w_{u}\)), and both are related to length of service obligation, \(\gamma\).

Training cost per conscript is inversely related to length of service obligation:
$$f = f\left( \gamma \right)\,\,{\text{and}}\,f^{\prime} \left( \gamma \right) < 0,\,f^{\prime \prime } \left( \gamma \right) > 0.$$

This reflects the higher turnover associated with shorter service obligation, which implies a higher frequency of training.

The tax burden borne by each conscript is positively related to length of service obligation and is concave:
$$g = g\left( \gamma \right){\text{and}}\,g^{\prime} \left( \gamma \right) > 0,g^{\prime \prime } \left( \gamma \right) > 0.$$
This reflects the upward-sloping supply of labor of each conscript.
There is some length of service \(\gamma^{*}\) that minimizes variable cost per conscript:
$$f^{\prime} \left( {\gamma^{*} } \right) + g^{\prime} \left( {\gamma^{*} } \right) = 0,$$
$$f^{\prime \prime } \left( {\gamma^{*} } \right) + g^{\prime \prime } \left( {\gamma^{*} } \right) > 0.$$
If the number of positions to fill with conscripts, \(\bar{N}\), is below \(N^{*} = \gamma^{*}\), then rather than shortening the length of service below \(\gamma^{*}\), it would be economical to keep the length of service at \(\gamma^{*}\) and use a lottery (with \(\theta = \bar{N}/\gamma^{*}\)) to choose the \(\bar{N} < N^{*}\). The reduction in training cost by doing this more than offsets the increase in burden it places on the conscripts. With this in mind, let us characterize the length of service obligation in a system with conscription. It is the solution to a constrained minimization problem.
$$\mathop {\hbox{min} }\limits_{\gamma ,\theta } \delta_{u} = f\left( \gamma \right) + g\left( \gamma \right) + w_{u}$$
$${\text{such that}} \gamma = N/\theta ,$$
$$0 < \theta \le 1,$$
$$N \ge \bar{N}.$$
The length of service obligation and lottery parameter that minimize the variable cost per conscript are
$$\gamma = \left\{ {\begin{array}{*{20}ll} \gamma^{*}, &\quad \bar{N} < N^{*} \\ \bar{N}, &\quad \bar{N} \ge N^{*} \\ \end{array} } \right.$$
$$\theta = \left\{ {\begin{array}{*{20}ll} \bar{N}/\gamma^{*} ,&\quad \bar{N} < N^{*} \\ 1 ,&\quad \bar{N} \ge N^{*} \\ \end{array} } \right.$$
Let us return to the effect of a distorting tax system on the costs of a conscription system compared to a voluntary recruitment system. As we argued before, tax inefficiency may be presumed to inflate the total burden of a voluntary recruitment system by a factor \(\alpha \ge 1\), that reflects the excess burden of taxes needed to pay the wage bill of voluntary recruits. Taxpayers must also bear the burden of funding the costs of training conscripts. Accordingly, the same factor \(\alpha\) should be applied to these training costs. In a system with conscription, the variable cost per conscript \(\delta_{u}\) is
$$\delta_{u} = \alpha f\left( \gamma \right) + g\left( \gamma \right) + w_{u} .$$

Filtering through our results in the previous paragraph, the effect of a more distorting tax system, meaning a higher value of \(\alpha\), is to increase \(N^{*}\), the maximum number of conscripts for which a lottery system would be instituted. With \(\alpha\)  > 1, the first-order condition determining \(N^{*} = \gamma^{*}\), the length of service obligation that minimizes variable cost per conscript, becomes \(\alpha f^{\prime} \left( {\gamma^{*} } \right) + g^{\prime} \left( {\gamma^{*} } \right) = 0.\)

By total differentiation,
$$\frac{{{\text{d}}N^{*} }}{{{\text{d}}\alpha }} = \frac{{{\text{d}}\gamma^{*} }}{{{\text{d}}\alpha }} = \frac{{ - f^{\prime} \left( {\gamma^{*} } \right)}}{{\alpha f^{\prime \prime } \left( {\gamma^{*} } \right) + g^{\prime \prime } \left( {\gamma^{*} } \right)}} > 0.$$
To economize on training cost, a country with a more distorting tax system is more likely to have a lottery, with longer service obligation than if it did not have a lottery. And as argued in the previous section—still true in this expanded framework with training costs of conscripts subject to tax distortion—the minimum force size for which conscription has lower cost than voluntary recruitment is higher, the greater the tax distortion. The upshot is that a country with a distorting tax system is more likely to have conscription, and if it does have conscription, is more likely to have a lottery system with longer service obligation than if it did not have a lottery and drafted an entire age cohort.
The Fig. 5 compares a country with a distorting tax system and another without. For the country with the distorting tax system, conscription has lower cost than voluntary recruitment for a wider range of staffing requirements, and conscription also entails a lottery system with longer service obligation for a wider range of staffing requirements.
Fig. 5

Effects of tax distortion on conscription and on length of conscript service obligation. The figure illustrates how an increase in the distorting effect of the tax system expands the range of staffing needs for which conscription has a lower cost than voluntary enlistment (a), and expands the range of staffing for which conscription would entail a lottery system with a longer service obligation (b)

Our argument relating length of service obligation to training costs assumed that the number of persons to be called, either by conscription or by voluntary enlistment, is set and equal to \(\bar{N}\). More generally, conscripts and volunteers are imperfect substitutes. Perhaps volunteers are more responsive to incentives, more apt to follow orders, and more trainable than conscripts, even though conscripts may have a lower cost. Then, with a draft, the desired number of conscripts \(\bar{N}\) would itself depend on the cost per conscript compared to the cost per volunteer. If the cost per volunteer is higher—because of the inefficiency of the tax system—then the target number of conscripts, for any given size of army, would be greater, which beyond some point could be achieved only by lengthening the service obligation. So there are two ways that a more distorting tax system favors longer length of service obligation. One is that, for a given number of conscripts, if the tax system is more distorting, conscription is more likely to entail a lottery, with higher average length of service obligation. The other is that, for a given military force size, if the tax system is more distorting, more of the force will be conscripts, and more conscripts—with no lottery, i.e., universal conscription—will mean longer service obligation.

In short, the inefficient tax systems of developing countries would, all else equal, incline them toward conscription with longer service obligations. Here, the “other things” being equal include the staffing need of the government in relation to the labor force, which we previously argued is likely to be smaller in a developing country—and which inclines developing countries to rely less on conscription. There is still another important difference between developing and developed countries.

For any given government staffing requirement, developing countries are relatively more likely to impose conscription. However, we have also suggested that developing countries are likely to have smaller government sectors and smaller government staffing requirements. Empirical investigation is needed to determine whether and how well the model explains the cross-country variation in recruitment systems.

Empirical analysis of cross-country variation in recruitment systems

Data and specification

If our reasoning is correct, countries with less efficient tax systems, wishing to recruit larger fractions of their labor forces, would be more likely to have conscription with longer service obligations. We will show that such patterns are indeed present in the data. To do this requires some attention to reverse causality.

The variables in our dataset are defined in Table 3, with descriptive statistics in Table 4 and correlation matrix in Table 5. The “Appendix A.1” describes sources in more detail. The dataset has observations for 104 countries, focused around one year—2012. It is a cross-section, not a panel dataset, and we do not invoke panel data techniques in our estimation. We include high-income countries in the dataset even though our interest is in explaining why developing countries are inclined toward conscription for the same reason that experiments include a control group as well as a treatment group. The ways that developing countries differ from rich countries matter for our explanation, not only the ways that developing countries differ from one another.
Table 3

Variables and Sources

Variable

Definitions

Sources

Length of service obligation

Months of service required of those conscripted (2012)

IISS (The Military Balance) [15], CIA (CIA Factbook) [6], War Resisters’ International (website) [27], and Toronto [26]

Non-resource government revenue

Non-resource government revenue as percentage of GDP (average, 2005–2013)

Prichard et al. [20]

Size of military

The number of armed force personnel as percentage of labor force (average, 2008–2013)

IISS (The Military Balance) [15]; secondary source: World Bank, World Development Indicators [28]

British legal origin

Dummy variable, Socialist legal origin: 1, otherwise: 0

La Porta et al. [16] and Siems [22]

Socialist legal origin

Dummy variable, British legal origin: 1, otherwise: 0

Financial inclusion index

An index on the ease of access, availability, and usage of formal financial system for all members of an economy

Park and Mercado [19]

Relationship with neighbor countries

Scale from 1 = good relations to 5 = hostile relations (average, 2005-–2013)

Economist Intelligence Unit; secondary source: Institute for Economics and Peace which uses it as a component of its “Global Peace Index,” at this link: http://www.visionofhumanity.org/#/page/indexes/global-peace-index [14]

Table 4

Descriptive statistics

Variable

Units of measurement

Observations

Mean

Standard deviation

Minimum

Maximum

Length of service obligation

Months

104

8.4

9.7

0

30

Non-resource government revenue

Percentage of GDP

104

24.6

10.9

4.0

49.6

Size of military

Percentage of labor force

104

1.2

1.2

0.1

6.9

British legal origin

Dummy

104

0.22

0.42

0

1

Socialist legal origin

Dummy

104

0.14

0.35

0

1

Financial Inclusion Index

Index,

(0 to 100)

104

33.2

23.4

2.4

91.0

Relationship with neighbor countries

Index

(1 to 5)

104

2.3

1.0

1

5

Table 5

Correlation matrix

Variables

Length of service obligation

Non-resource government revenue

Size of military

British legal origin

Socialist legal origin

Financial inclusion index

Relationship with neighbor countries

Non-resource government revenue

− 0.32***

      

Size of military

0.22**

0.04

     

British legal origin

− 0.23**

− 0.23**

− 0.12

    

Socialist legal origin

0.40***

− 0.01

0.15

− 0.22**

   

Financial Inclusion Index

− 0.32***

0.66***

0.04

− 0.04

− 0.17*

  

Relationship

with neighbor countries

0.44***

− 0.39***

0.33***

− 0.06

0.40***

− 0.41***

 

GNI per capita

− 0.28***

0.62***

− 0.09

0.02

− 0.22**

0.74***

− 0.50***

Number of observations = 104, except for GNI per capita (n = 98)

The coefficient estimates are in bold to distinguish them from the standard errors and other statistics

*p < 0.1, **p < 0.05, ***p < 0.01

Our measure of conscription is the length of conscript service obligation in months, which is zero for about half of the countries. Longer service obligation means that more draftees are potentially in service. Unfortunately, we do not observe the number of draftees in service, nor, if the conscription system is by lottery, what fraction are called up. Nor do we have any information about exemptions or deferments. Length of service obligation is an imperfect measure of each country’s reliance on conscription, but it is the best measure available, for now.

We observe for each country, size of military as a fraction of the labor force. As evident from the Table 5 correlation matrix, this is negatively correlated with length of service obligation. This negative correlation is consistent with our reasoning, but not conclusive, because longer service obligation could be the cause of a larger military force rather than the reverse. Our estimating strategy is to find an instrumental variable correlated with size of military but not directly influencing system of recruitment. That instrument is an index for peacefulness of relationships with neighboring countries, ranging from 1 = peaceful, to 5 = very aggressive with frequent invasions by neighboring countries.3 This index is constructed by the Economist Intelligence Unit and revised annually. Our secondary source for it is the website of the Institute for Economics and Peace.4 We have averaged the measure for each country, 2008–2103. Our presumption is that the government of a country with less peaceful relationships with neighboring countries would wish to recruit a larger military force, and the larger the force to be recruited, the lower the cost of conscription compared to voluntary enlistment. As shown in Table 5, relationship with neighbors is indeed correlated with size of military, which is consistent with this reasoning.

Our claim is that relations with neighboring countries affect the demand for armed forces but have no direct bearing on the choice of recruitment system—the larger armed force precipitated by deteriorating relations with neighbors could be staffed either by voluntary enlistment or by conscription. If this is true, then in a regression in which size of military influences choice of recruitment system, relations with neighbors is a valid instrument for the size of the military. Readers that might be skeptical of our claim should consider the recent example of Sweden. According to The Guardian (“Sweden to reintroduce conscription amid rising Baltic tensions,” March 2, 2017), Sweden is reintroducing conscription to increase the size of its armed forces in response to deteriorating relations with Russia [24]. That fits our reasoning.

Countries with less efficient tax systems will tend to collect less tax revenue. As shown in the Table 5 correlation matrix, non-resource government revenue—that is government revenue other than arising from the exploitation of oil and mineral resources, as a percentage of GDP—is negatively correlated with length of service obligation. This is consistent with our reasoning that countries with less efficient tax systems would be more likely to have conscription. But, again, reverse causality is a concern. Perhaps the smaller military wage bill under a conscription system compared to a voluntary system is a cause for reduced tax revenue rather than merely a result of it. We require an instrument that is correlated with constricted tax revenue but not directly influencing the choice of recruitment system. Our instrument is the “financial inclusion index” already described earlier, that indicates on a scale from 0 to 100 the extent that a country’s citizens and businesses use banks for their economic transactions. A country with less banking service has a larger underground economy, beyond the purview of government taxing authorities, and its government can raise revenue only by imposing taxes with greater excess burdens. As shown in Table 5, the financial inclusion index is indeed positively correlated with non-resource government revenue relative to GDP. Our presumption is that a country with less financial inclusion has a less efficient tax system, and for that reason the cost of the higher wage bill of a voluntary force compared to a conscripted force is also greater, which makes the country more likely to have conscription with longer service obligation.

A comment is in order on our use of non-resource government revenue—that is government revenue other than arising from the exploitation of oil and mineral resources, as a percentage of GDP—as indicative of the efficiency of the tax system. Why not also include government revenue arising from the exploitation of oil and mineral resources? Our line of argument is that absence of bank records induces greater reliance on distorting and burdensome taxes, that both limit government spending and raise the excess burden of the taxes needed to fund such spending, including the greater spending associated with voluntary recruitment systems compared to conscription. As argued by Prichard et al. [20], our source for the non-resource government revenue variable, the resource-based revenue reported by many governments of developing countries is inconsistent and unreliable. Government revenue arising from the exploitation of oil and mineral resources is poorly measured, and that is reason enough to exclude it and use instead the narrower variable non-resource government revenue. Our claim is that non-resource government revenue is an accurate indicator of the extractive capacity of the state, as indicated by its high correlation with the financial inclusion index. Skeptics of our claim should also note that government revenue arising from the exploitation of oil and mineral resources has a complicated relation to the excess burden of taxation. Government-owned natural resources are not necessarily a windfall that relaxes the fiscal constraint and diminishes the excess burden of taxes. Exploitation of natural resources through government enterprise or through opaque arrangements with foreign investors are themselves examples of distorting measures that the governments of rich countries with sophisticated banking systems are apt to avoid but that developing countries embrace. It is not at all clear whether governments with ample revenue from the exploitation of oil and mineral resources should be regarded as having a less distorting tax system—which would favor voluntary enlistment, or a more distorting one—which would favor conscription. Ours is not the only way of modeling the way that constraints on design of a tax system influence the public choice between voluntary enlistment and conscription, but it is coherent and logical and is a defensible way of confronting the data.

Our estimating model includes control variables. The legacies of the Colonial Age of the 19th century and Cold War of the 20th century still affect the world in the 21st century. We include in our dataset two dummy variables that code those countries whose legal systems are of British heritage and of Socialist heritage, i.e., communist. As the correlation coefficients of Table 5 show, countries with British legal origins are more likely to have all-volunteer armies while countries which still keep a Socialist legal system are more likely to have conscription. Mulligan and Shleifer [17] argue that countries with British legal origins (common law) have higher fixed cost for conscription than countries following civil law. However, if classifying legal origins into five groups: British, French, German, Scandinavian and Socialist, the fraction of countries that have all-volunteer recruitment in the British group and German group are nearly equal (84.9% and 77.8%). Many countries following the German legal system are in Eastern Europe. Most of these former communist nations abolished conscription after 1991, but most of the other countries with Socialist legal origins still have military conscription. Cambodia after the 1993 Peace Accords had an all-voluntary army since, but in 2006 they again adopted the draft. We follow Mulligan and Shleifer in presuming that the nature of the legal system directly affects the cost of administering and enforcing conscription.

Empirical model and result

The model we estimate has three structural Eqs. (1), (2), and (3), below. We suppose that length of service obligation in months, \(y_{1i}\), is the censored value of a latent variable \(y_{1i}^{*}\).
$$y_{1i} = \left\{ {\begin{array}{*{20}ll} {y_{1i}^{*} } &\quad if \;{y_{1i}^{*} > 0} \\ 0 &\quad if \; {y_{1i}^{*} \le 0.} \\ \end{array} } \right.$$

In the first structural Eq. (1), the latent variable, \(y_{1i}^{*}\), depends on: \(y_{2i}\) ≡ Non-resource government revenue as a fraction of GDP, \(y_{3i}\) ≡ Size of military force as a fraction of the labor force, \(x_{1i}\) ≡ Dummy equal to one if British legal origin, \(x_{2i}\) ≡ Dummy equal to one if Socialist legal origin.

The equation is
$$y_{1i}^{*} = \beta_{0} + \beta_{1} y_{2i} + \beta_{2} y_{3i} + \beta_{3} x_{1i} + \beta_{4} x_{2i} + u_{i} ,$$
(1)
where \(\beta_{i}\) are the coefficients to be estimated, and \(u_{i}\) is an error term.

In the second structural Eq. (2), Non-resource government revenue as a fraction of GDP, \(y_{2i}\), depends on: \(z_{1}\) ≡ Financial inclusion index (ranging from 0 = no banking, to 100 = full recourse to banks).

The equation is
$$y_{2i} = \gamma_{0} + \gamma_{1} z_{1} + v_{i} ,$$
(2)
where \(\gamma_{i}\) are the coefficients to be estimated, and \(v_{i}\) is an error term.

And in the third and final structural Eq. (3), size of military force as a fraction of the labor force, \(y_{3i}\), depends on: \(z_{2}\) ≡ Relationship with neighbor countries (ranging from 1 = peaceful, to 5 = very aggressive).

The equation is
$$y_{3i} = \delta_{0} + \delta_{1} z_{2} + \varepsilon_{i} ,$$
(3)
where \(\delta_{i}\) are coefficients to be estimated, and \(\varepsilon_{i}\) is an error term.
We assume that the vector of errors of the three structural equations \(\left( {u_{i} ,v_{i} ,\varepsilon_{i} } \right)\) is distributed multivariate normal with mean zero and covariance matrix
$$\varSigma = \left[ {\begin{array}{*{20}c} {\sigma_{u}^{2} } \quad{\varSigma_{21}^{\prime} } \quad{\varSigma_{31}^{\prime} } \\ {\varSigma_{21} } \quad{\sigma_{v}^{2} } \quad{\varSigma_{32}^{\prime} } \\ {\varSigma_{31} } \quad{\varSigma_{32} } \quad{\sigma_{\varepsilon }^{2} } \\ \end{array} } \right].$$
(4)
As represented by Eq. (1), we suppose that \(z_{1}\) and \(z_{2}\), Financial inclusion index and Relations with neighbors, are independent of \(u_{i}\); i.e., they are valid instruments for \(y_{2}\) and \(y_{3}\), Non-resource government revenue relative to GDP and Size of military relative to the labor force.

The maximum likelihood estimator of the censored regression Eq. (1) is the Tobit estimator. Notice that if \(\varSigma_{21} \ne 0\), then \(y_{2i}\) is correlated with \(u_{i}\), and if \(\varSigma_{31} \ne 0\), then \(y_{3i}\) is correlated with \(u_{i}\). The Tobit estimates of coefficients in Eq. (1) will be biased unless these non-zero correlations are taken into account in forming the likelihood function. The estimates that maximize the correctly specified likelihood function are called full-information maximum likelihood estimates, in effect 3SLS estimates—they use the excluded exogenous variables to instrument the right-hand-side endogenous variables, and they weight observations according to the sample estimates of the correlation between error terms of the different equations. This estimation was accomplished with SAS software (Proc QLIM).

Limited-information maximum likelihood estimates of Eq. (1) would use \(z_{1}\) and \(z_{2}\), financial inclusion index and relations with neighbors, as instruments for \(y_{2}\) and \(y_{3}\), non-resource government revenue relative to GDP and size of military relative to the labor force, but without weighting observations according to the estimated covariance of errors across all three equations. This limited-information maximum likelihood Tobit estimate, shown in the second column of Table 6, was computed using STATA software (ivtobit command).
Table 6

Tobit estimates, length of service obligation

 

(1) Naïve specification

(2) Limited-information

maximum likelihood

(3) Reduced form

(4) Structural equations (Full-information maximum likelihood)

Dependent variable:

Length of service obligation

Length of service obligation

Length of service obligation

Length of service obligation

Non − resource government revenue

Size of military

British legal origin

− 13.10***

4.55

− 14.12***

4.97

− 12.20***

4.67

− 13.03***

4.54

  

Socialist legal origin

14.00**

4.20

12.17***

4.55

9.03**

4.60

10.63**

4.54

  

Non-resource government revenue

− 0.56***

0.15

− 0.76***

0.23

 

− 0.57**

0.26

  

Financial Inclusion Index

  

− 0.15*

0.08

 

0.31***

0.04

 

Size of military

2.26*

1.29

5.20

3.47

 

8.50*

4.97

  

Relationship with neighbor countries

  

4.42**

1.81

  

0.39***

0.11

Intercept

13.57

4.33

15.68

7.40

− 2.18

5.76

7.17

9.79

14.35

1.42

0.28

0.27

Correlation between predicted value of dependent variable and actual value

0.554

0.554

0.526

0.395

0.710

0.319

Correlation of errors across structural equations: Length of service obligation × non-resource government revenue, \(\varSigma_{21} /\left( {\sigma_{u} \sigma_{v} } \right)\)

    

0.09

0.18

 

Correlation of errors across structural equations: Length of service obligation × Size of military, \(\varSigma_{31} /\left( {\sigma_{u} \sigma_{\varepsilon } } \right)\)

     

− 0.47*

0.29

Observations

104

104

104

104

104

104

The coefficient estimates are in bold to distinguish them from the standard errors and other statistics

Standard errors beneath coefficients: *p < 0.1, **p < 0.05, ***p < 0.01

The full-information maximum likelihood estimates of the three structural Eq. (1), (2) and (3) are reported in the last columns of Table 6. The third column of Table 6 reports the estimates of the reduced-form equation for the three-equation system:
$$y_{1i}^{*} = \beta_{0} + \beta_{1}^{*} z_{1} + \beta_{2}^{*} z_{2} + \beta_{4} x_{1i} + \beta_{5} x_{2i} + u_{i} ,$$
(5)
where \(\beta_{1}^{*} = \beta_{1} \left( {\gamma_{0} + \gamma_{1} } \right)\) and \(\beta_{2}^{*} = \beta_{2} \left( {\delta_{0} + \delta_{1} } \right).\)

The first column of Table 6 reports a naïve specification that presumes non-resource government revenue relative to GDP and size of military relative to the labor force are exogenous variables.

Before turning to the results of estimation, let us briefly comment on the relation between the various estimating equations. Our basic framework is a three-equation system that jointly determines length of service obligation, non-resource government revenue relative to GDP, and size of military relative to the labor force. This system may be estimated using full-information maximum likelihood (FIML) or limited-information maximum likelihood (LIML). We use both, and report the FIML estimates of all three structural equations and the LIML estimate of the structural equation explaining length of service obligation. We also report the estimate of the reduced form of this system for explaining length of service obligation to show that we have included relevant variables in the model. The FIML estimates are our preferred specification for the structural equations because, within our assumed framework, it results in more efficient estimates. We report the LIML estimate of the structural equation explaining length of service obligation as a robustness check—LIML can be the more efficient estimator if the assumptions about the cross-equation error structure that underlie the FIML estimator are incorrect. Both the LIML and FIML estimates are aimed at overcoming the possible bias resulting from correlation between included variables and unobserved variables in the structural equation explaining length of service obligation. The naïve specification ignores this possibility, and models non-resource government revenue relative to GDP and size of military relative to the labor force as exogenous determinants of length of service obligation. We report estimates of the naïve specification to show, by comparison with the other estimates, the size of endogeneity bias. We label it the ‘naïve’ specification because it ignores the possibility of endogeneity bias—it is not our preferred specification. We next turn to the estimates themselves.

The Table 6 reports for each specification, coefficient estimates and their standard errors, estimated correlation between errors in different structural equations, and correlation between the censored predicted value of each equation and observed dependent variable. This last item indicates goodness of fit.

The coefficient estimates of Table 6 are consistent with our reasoning. As shown by the full-information-maximum likelihood estimates of the three structural equations, financial inclusion increases non-resource government revenue, and to just that extent, contributes to reduced length of service obligation. Worse relations with neighbors increase size of military, and to just that extent, contributes to greater length of service obligation. The estimated coefficients of the reduced form equation explaining length of service obligation also comports with this logic. The limited-information maximum likelihood estimate of the equation explaining length of service obligation resembles the full-information maximum likelihood estimate that is our preferred specification.

The estimated coefficients of the naïve specification differ from those of the LIML and FIML estimates, an indication of the bias that results from ignoring the reverse causation running from length of service obligation toward size of military and toward non-resource government revenue. The true direct effect of size of military on length of service obligation appears to be much greater than the estimated coefficient in the naïve specification—three or four times greater. The estimated coefficient on non-resource government revenue in the naïve specification also seems to reflect bias though not as much as the coefficient on size of military.

The full-information maximum likelihood estimates (fourth column of Table 6), can be used to predict the censored variable length of service obligation and its corresponding latent variable in the Tobit estimating equation. These predictions, averaged over high-income countries and over developing countries, with standard errors, are shown in Table 7. The country classifications are those used by the World Bank, with “developing” defined as low income, low middle income and high middle income.
Table 7

Predicted values of latent variable and censored variable based on full − information maximum likelihood estimate (Table 6, col. 4)

 

Number of countries

Latent variable: \(y_{1i}^{*}\)

Censored variable: length of service obligation in months, \(y_{1i}\)

Developing

69

5.5

14.3

8.1

11.6

High income

35

−5.9

10.0

2.0

4.9

Means and standard deviations, for developing countries and high − income countries

The coefficient estimates are in bold to distinguish them from the standard errors and other statistics

These predictions are a useful point of reference for interpreting the coefficient estimates in Table 6. The predicted value of the latent variable is positive for developing countries, but negative for high-income countries, and with sizeable standard deviations for both. In other words—and unsurprisingly—the prediction matches the data on which the estimating equation is based: the modal developing country has conscription and the modal high-income country does not, but some countries in both groups have conscription and some not. More to the point, changes in independent variables sufficient to lower the latent variable by more than 5.5 months would predict a switch by the average developing country, from conscription to voluntary enlistment. Given this, are the estimated effects on conscription of size of military and of non-resource government revenue big enough to matter?

The Table 8 shows the differences in independent variables between developing countries and high-income countries. The difference between developing countries and high-income countries in financial inclusion index is about 1.6 standard deviations—roughly 38. Based on the estimated coefficient of the financial inclusion index in the reduced form equation (= − 0.15), an increase in the index of 38 would increase the latent variable by 5.7 months, enough to shift the average developing country from conscription to voluntary enlistment.
Table 8

Means and standard deviations of selected variables, for developing countries and high-income countries

 

Number of countries

Non-resource government revenue, (percent of GDP)

Financial inclusion index

Size of military, (percent of labor force)

Relationship with neighboring countries

Developing

69

19.6

(8.9)

20.3

(14.5)

1.2

(1.3)

2.6

(0.9)

High income

35

34.5

(7.1)

58.6

(15.3)

1.1

(1.0)

1.6

(0.8)

The coefficient estimates are in bold to distinguish them from the standard errors and other statistics

Developing countries and high-income countries do not differ so much from one another in size of military or in relations with neighbors. For either, a one-standard deviation deterioration in relations with neighbors—from 2 = “generally good”, to 3 = “serious tensions”—based on the estimated coefficient on relations with neighbors in the reduced form equation (= 4.42), would increase the latent variable by about 4 months.

The difference between British legal origin and Socialist legal origin, based on the estimated coefficients of the reduced form equation, accounts for a 21-month difference in latent variable. Legal origin matters greatly, in predicting which countries have conscription. Former or current communist regimes, with command and control systems and traditions from the Cold War, have lower costs of administering and enforcing conscription. In Vietnam and China, people have to register their households with the local authorities, which note the name, address, and birthday of each household member. In this case, administering and enforcing conscription would not have as high a cost as otherwise.

We have shown that our reasoning does account for some of the variation across countries in conscription, and does partially account for why developing countries are more likely than high-income countries to have conscription with longer service obligations. But there is much variation that our estimating equation does not capture. The correlation between length of service obligation predicted by our reduced form equation and its actual value is 0.526. The square of this correlation, 0.28 (= 0.532), indicates that the predicted values share about 28% of their variance with the observed values. There is much variation that is unexplained by our equation.

Our reasoning presumes that in public choice between conscription and voluntary enlistment, the system with lower tax burden is more likely to be chosen. Differences in tax burden related to the misallocation of labor are missing from our equation. We did attempt to include such variables but none were significant. The exemptions and deferments that differentiate the conscription systems of different countries are probably more sensitive than length of service obligation to variables correlated with possible misallocation of labor in conscription systems. But we could not assemble data related to exemptions and deferments.

Conclusion

Since the Cold War ended, the world has become relatively peaceful and the possibility for a total war between countries is low, but there are still many countries which enforce military conscription. Most of them are developing countries. We have argued that within the framework of Mulligan and Shleifer [17], developing countries, compared to the developed countries, are likely to have lower costs of conscription relative to voluntary recruitment. We statistically analyzed cross-country data of 104 countries. The empirical estimates suggest that the higher wage bill of an all-volunteer system is an important reason for military conscription. Governments that must resort to distorting and burdensome taxes and are constrained in their spending, are more likely to have military conscription. For them, the lower wage bill that accompanies conscription outweighs the higher fixed cost of administering and enforcing a conscription system compared to a voluntary recruitment system.

Technological advance will eventually cause the developed countries to become more like the advanced countries of today. They will have more efficient tax systems for funding the higher wage bill of an all-volunteer force and face greater costs of misallocating labor if using conscription. The developing countries are then likely to abandon conscription for the same reasons that the developed countries of today are already doing so. On the other hand, if IT advances enable developing countries to lower their costs of setting up and administering conscription systems, conscription may persist for a while longer.

Footnotes

  1. 1.

    [Imbens and van der Klauuw [13]] find that former draftees in the Netherlands have 5% lower salaries than their birth cohorts, on average. Bingley [4] find that high-ability former draftees faced a 7% earning penalty in Denmark. Grenet et al. [10] and [Bauer et al. [2]] find no effect of military conscription on wages—in Britain and in Germany. [Card and Cardoso [5]] find a positive effect of conscription on the wage of the low-educated in Portugal. Galiani et al. [8], and Hjalmarsson and Lindquist [12] find that those having been conscripted were more likely to later commit crimes—in Argentina and in Sweden, which further indicates how conscription preempts the formation of human capital. However, there is apparently no relationship between conscription and crime in Denmark [Albæk[1]].

  2. 2.

    Mulligan and Shleifer [17] also include conscription with “exemption” and with “replacement” in their analysis. Conscription with exemption means that students, parents, workers in designated industries, and such, are exempted from service. And conscription with replacement means that a draftee may pay another person to serve in his or her place, or may pay a monetary tax in lieu of service. Conscription with replacement can be found in the American Civil War (1861–1865), in which 81% of draftees made a monetary payment as a commutation fee or to hire a substitute (Mulligan [18]). Lack of data prevents us from extending our cross-country empirical analysis to take into account which countries have conscription systems with exemption or replacement.

  3. 3.

    From: Institute for Economics and Peace, Global Peace Index, 2016, p. 99:

    “Assessment of the intensity of contentiousness of neighbors, ranked from 1–5 (peaceful to very aggressive) by the EIU’s Country Analysis team. Country analysts are asked to assess this indicator on an annual basis, for the period March to March.

    Scoring Criteria: basis, for the period March to March.

    Scoring Criteria:

    1 = Peaceful: None of the neighbors has attacked the country since 1950.

    2 = Low: The relationship with neighbors is generally good, but aggressiveness is manifest in politicians’ speeches or in protectionist measures.

    3 = Moderate: There are serious tensions and consequent economic and diplomatic restrictions from other countries.

    4 = Aggressive: Open conflicts with violence and protests.

    5 = Very aggressive: Frequent invasions by neighboring countries.

  4. 4.

    The raw observations on relationships with neighboring countries are taken from the website of the Institute for Economics and Peace which uses it as a component of its “Global Peace Index,” at this link: http://www.visionofhumanity.org/#/page/indexes/global-peace-index. After accessing this link, click on “specify indicator”, and select “neighboring countries relations,” from the drop-down menu. Then click on “global rankings” .

References

  1. 1.
    Albæk, K., Leth-Petersen, S., Le Maire, D., & Tranæs, T. (2013). Does peacetime military service affect crime?. IZA Discussion Paper No. 7528.Google Scholar
  2. 2.
    Bauer, T. K., Bender, S., Paloyo, A. R., & Schmidt, C. M. (2012). Evaluating the labour-market effects of compulsory military service. European Economic Review, 56(4), 814–829.CrossRefGoogle Scholar
  3. 3.
    Besley, Timothy, & Persson, Torsten. (2014). Why do developing countries tax so little? Journal of Economic Perspectives, 28(4), 99–120.CrossRefGoogle Scholar
  4. 4.
    Bingley, P., Lundborg, P., & Lyk-Jensen, S. Vincent. (2014). Opportunity cost and the incidence of a draft lottery. IZA Discussion Paper No. 8057.Google Scholar
  5. 5.
    Card, D., & Cardoso, A. R. (2012). Can compulsory military service raise civilian wages? Evidence from the Peacetime draft in Portugal. American Economic Journal: Applied Economics, 4(4), 57–93.Google Scholar
  6. 6.
    Central Intelligence Agency. (annual). The CIA Factbook. Virginia.Google Scholar
  7. 7.
    Economist magazine. (2011). Conscription a farewell to arms: Military service lingers in countries that are poor or small, but elsewhere it is on the way out, September 10, 2011.Google Scholar
  8. 8.
    Galiani, Sebastian, Rossi, Martín A., & Schargrodsky, Ernesto. (2011). Conscription and crime: Evidence from the Argentine draft lottery. American Economic Journal: Applied Economics, 3(2), 119–136.Google Scholar
  9. 9.
    Gordon, Roger, & Li, Wei. (2009). Tax structures in developing countries: Many puzzles and a possible explanation. Journal of Public Economics, 93((7–8)), pp. 855–866, Aug.Google Scholar
  10. 10.
    Grenet, J., Hart, R., & Roberts, E. (2011). Above and beyond the call: Long-term real earnings effects of British male military conscription in the post-war years. Labour Economics, 18(2), 194–204.CrossRefGoogle Scholar
  11. 11.
    Hall, Joshua C. (2011). The Worldwide decline in conscription: A victory for Economics? October 3, 2011. Library of Economics and Liberty. Retrieved July 15, 2016 from the World Wide Web: http://www.econlib.org/library/Columns/y2011/Hallconscription.html.
  12. 12.
    Hjalmarsson, R., & Lindquist, Matthew J. (2016). The causal effect of military conscription on crime and the labour market, CEPR Discussion Paper No. 11110.Google Scholar
  13. 13.
    Imbens, Guido, & van der Klauuw, Wilbert. (1995). Evaluating the cost of conscription in the Netherlands. Journal of Business and Economic Statistics, Vol. 13(2), 207–215.Google Scholar
  14. 14.
    Institute for Economics and Peace (website). Global Peace Index. http://www.visionofhumanity.org/#page/indexes/global-peace-index/. Accessed 12 Dec 2015.
  15. 15.
    International Institute for Strategic Studies (annual). The Military Balance. London, various issues.Google Scholar
  16. 16.
    La Porta, Rafael, Lopez-de-Silanes, Florencio, Shleifer, Andrei, & Vishny, Robert. (1999). The quality of government. Journal of Law Economics and Organization, 15(1), 222–279.CrossRefGoogle Scholar
  17. 17.
    Mulligan, Casey, & Shleifer, Andrei. (2005). Conscription as regulation. American Law and Economics Review, 7(1), 85–111.CrossRefGoogle Scholar
  18. 18.
    Mulligan, C. (2015). In-kind taxes, behavior, and comparative advantage, NBER Working Paper 21586.Google Scholar
  19. 19.
    Park, C-Y. & Mercado, R. Jr. (2015). Financial inclusion, poverty, and income inequality in developing Asia, ADB working paper series no. 426/2015, Asian Development Bank, Singapore Economic Review (forthcoming). http://hdl.handle.net/11540/2272. Accessed 12 Dec 2015.
  20. 20.
    Prichard, Wilson, Cobham, Alex C., & Goodall, Andrew. (2014). The ICTD Government Revenue Dataset (September 1, 2014). ICTD Working Paper 19.Google Scholar
  21. 21.
    Ross, Thomas. (1994). Raising an army: A positive theory of military recruitment. Journal of Law and Economics, 37, 109–131.CrossRefGoogle Scholar
  22. 22.
    Siems, Mathias M. (2007). Legal origins: Reconciling law and finance and comparative law. McGill Law Journal, 52(1), 2007.Google Scholar
  23. 23.
    Slemrod, Joel. (1990). Optimal taxation and optimal tax systems. Journal of Economic Perspectives, 4(1), 157–178.CrossRefGoogle Scholar
  24. 24.
    Sweden to reintroduce conscription amid rising Baltic tensions, The Guardian, March 2, 2017.Google Scholar
  25. 25.
    Tarabar, Danko, & Hall, Joshua, C. (2016). Explaining the worldwide decline in the length of mandatory military service, 1970–2010. Public Choice, 168(1), 55–74.CrossRefGoogle Scholar
  26. 26.
    Toronto, N. (2005). Military Recruitment Data Set, version 2005.1. Available from the author at nathan.toronto@us.army.mil.Google Scholar
  27. 27.
    War Resisters’ International (website). World survey of conscription and conscientious objection to military service. http://www.wri-irg.org/co/rtba/.
  28. 28.

Copyright information

© Japan Economic Policy Association (JEPA) 2018

Authors and Affiliations

  1. 1.Department of Agricultural and Applied EconomicsUniversity of Wisconsin-MadisonMadisonUSA
  2. 2.Faculty of Economics, Ritsumeikan UniversityRitsumeikan UniversityShigaJapan

Personalised recommendations