U.S. War Costs: Two Parts Temporary, One Part Permanent

Abstract

Military spending, fatalities, and the destruction of capital, all of which are immediately felt and are often large, are the most overt costs of war. They are also relatively short-lived. But the costs of war borne by combatants and their caretakers, which includes families, communities, and the modern welfare state, tend instead to be lifelong. In this paper I show that a significant component of the budgetary costs associated with U.S. wars is long-lived. One third to one half of the total present value of historical war costs are benefits distributed over the remaining life spans of veterans and their dependents. Even thirty years after the end of hostilities, typically half of all benefits remain to be paid. Estimates of the costs of injuries and deaths suggest that the private burden of war borne by survivors, namely the uncompensated costs of service-related injuries, are also large and long-lived.

1. Introduction

It is well understood that governments of modern nation-states spend very large amounts on military activities during periods of major warfare. In earlier history, warfare between looser agglomerations could last for many years and even decades, but in the modern era, major wars have most often been relatively brief. Direct military spending has typically been financed through deficits, and the impacts of short-lived but massive amounts of military spending and borrowing on GDP and interest rates are perennial topics in macroeconomics (Barro, 1981, 1987; Evans, 1987; Seater, 1993; Wang, 2005; Barro, 2006; Hall, 2009; Ramey, 2011a,b).

Large costs of military spending during wartime are also part of one side of the equation in the cost-benefit analysis of whether or not to go to war. Research into the economics of warfare is broad-based and vibrant (Garfinkel and Skaperdas, 2010), and cost forecasts and estimates are important elements in our understanding of the behavior of governments (Clark, 1931; Goldin and Lewis, 1975; Edelstein, 2000; Nordhaus, 2002; Wallsten and Kosec, 2005; Stiglitz and Bilmes, 2008; Davis et al., 2009; Glick and Taylor, 2010).

My contribution in this paper, which is empirically focused, provides new insights into both of these areas. I show that the budgetary costs, or the payments by the federal government, associated with major U.S. wars have throughout its history been larger and much longer-lived than is commonly understood. The budgetary costs of war include the direct military spending that is relatively short-lived, but they also include transfer payments and in-kind benefits given to surviving veterans, their spouses, and their survivors. These latter components of budgetary costs, which I will collectively call veterans’ benefits, are significantly smaller when measured on an annual basis but are also very long-lived. I find that veterans’ benefits typically account for between one third to one half of the present value of all war-related government spending.

I currently restrict my attention to U.S. war costs, but similar patterns and challenges are likely to exist in other industrialized countries. The generosity of public pensions, their work disincentives, and their fiscal impacts tend to vary across the OECD (Gruber and Wise, 1998, 2005), with U.S. pension systems often appearing ungenerous, benign, and small by comparison. To the extent that other advanced countries have ceded major military operations to the U.S. since World War II, veterans’ benefits may become less of an issue in those countries over time. But the long shadows of the major European wars of the 20th century could easily have been more important in fiscal terms for the countries that were directly affected.

The implication for macroeconomic studies is that at least in the modern era in the U.S., war spending broadly defined is not all that temporary after all. Under the assumption that it was, previous researchers have attributed the lack of increase in U.S. interest rates during wartime either to Ricardian equivalence (Evans, 1987; Seater, 1993), to patriotism (Mulligan, 1998), or to heightened risk of future disasters (Barro, 2006). By contrast, Barro (1987) had found that in Britain, direct military spending up through World War I (1914–1918) did raise interest rates, as most macroeconomic models would predict if the spending were in fact temporary. The insight that U.S. war-related spending has been relatively long-lived due to veterans’ benefits may somewhat reduce the size of this puzzle. As discussed by Costa (1998), Linares (2001), and by Gerber (2000), the permanent system of broad-based veterans’ benefits that took hold in the U.S. several decades after its Civil War ended in 1865 represented a major innovation in the development of old-age support that was unique in the world at the time. Although I restrict my focus in this study to U.S. data alone, it seems plausible that the long right tail of veterans’ benefits did not exist in Britain or elsewhere much prior to World War I,¹ and that may explain the results of Barro (1987), which do not extend past 1918 to British data (Barro, 2006).

The implications of my results for the cost side of the cost-benefit analysis of warfare remain somewhat murky, because it is difficult to rigorously identify an economic cost as opposed to a budgetary cost. One difficulty is that as the historical record reveals, veterans’ benefits in the U.S. have always been a mixture of compensation for war-related wounds and for service. Only the former is a clear economic cost of war, by which I mean an additional burden caused by the choice to go to war. Transfers that are unrelated to war wounds may be an economic cost if they represent a deferred part of the marginal product of labor diverted for war purposes. But if instead they are a pure transfer, perhaps they do not belong in a cost-benefit calculation. A second difficulty is that because U.S. government compensation for war-related wounds is designed only to replace lost earnings, it may undercompensate for the harm and produce downward bias in the measure of economic cost. In addition to measuring the budgetary costs associated with war veterans’ benefits, which could be larger or smaller than the economic costs, I also directly estimate the economic costs for recent war cohorts using micro-level data. These estimates, which are structurally similar to those of Stiglitz and Bilmes (2008) for veterans of the wars in Iraq and Afghanistan, suggest that the economic costs of war wounds may have been much larger than the budgetary costs.

Other types of government activities may trigger long-lived obligations as well, but I do not explicitly consider them. Civil service pensions would be a long-term cost associated with an expansion in the bureaucracy, for example. Because the typical counterfactual scenario we have in mind is not going to war and not spending the money at all, rather than spending it on an expanded bureaucracy, I leave a comparative analysis of long-lived government obligations to future work.

A final and significant shortcoming is that my analysis focuses only on the costs of waging war and offers no insights at all about the benefits. While I show that war costs associated with veterans are large and long-lived, it is certainly possible that the benefits of war, whether they may include freedom and democracy or territorial control, are also large and long-lived. I merely intend my results to help better inform future cost-benefit analysis, which must also take into account the benefits.

2. The scope and aftermath of U.S. conflicts

Wars are costly because personnel and matériel must be deployed to combat zones, because hostilities result in deaths and wounded, and because surviving veterans and survivors of deceased veterans require medical care and are entitled to compensation. All of these costs tend to vary with the scope of the conflict, with offensive and defensive military technology, with medical technology, and with the general mortality environment faced by veterans and their survivors.

Table 1 lists statistics detailing several of these dimensions for each major U.S. war. The source here and throughout this and the next section is the Millennial Edition of the Historical Statistics of the United States. Details are described in the appendix. The left panel in the table shows estimates of military personnel involved, military fatalities, the number of service members experiencing wounds that did not result in death, and the number of surviving veterans, calculated as total personnel minus deaths.² The right panel displays several crude incidence indicators: the number of wounded per participating personnel, wounded per killed, and wounded per surviving veteran.

Table 1.

Participants, deaths, and wounded in major U.S. wars

					Wounded as a share of:
Conflict	Personnel	Killed	Wounds not mortal	Surviving veterans	Personnel	Killed	Survivors
Revolutionary Wars (1775-1783)	217,000	4,435	6,188	212,565	0.029	1.395	0.029
War of 1812 (1812-1815)	286,730	2,260	4,505	284,470	0.016	1.993	0.016
Mexican War (1846-1848)	78,718	13,283	4,152	65,435	0.053	0.313	0.063
Civil War (1861-65)	3,277,556	622,511	478,968	2,655,045	0.146	0.769	0.180
Confederate	1,064,193	258,000	197,087	806,193	0.185	0.764	0.244
Union	2,213,363	364,511	281,881	1,848,852	0.127	0.773	0.152
Spanish American War (1898)	306,760	2,446	1,662	304,314	0.005	0.679	0.005
World War I (1917-1918)	4,734,991	116,516	204,002	4,618,475	0.043	1.751	0.044
World War II (1941-1945)	16,112,566	405,399	671,846	15,707,167	0.042	1.657	0.043
Korea (1950-1953)	5,720,000	36,576	103,284	5,683,424	0.018	2.824	0.018
Vietnam (1964-1972)	8,744,000	58,200	153,303	8,685,800	0.018	2.634	0.018
First Gulf War (1990-1991)	2,225,000	383	467	2,223,038	0.000	1.219	0.000
Iraq and Afghanistan (OEF/OIF) (2001-)	2,100,000	5,376	36,906	2,094,624	0.018	6.865	0.018

Sources: For the Korean War and earlier, Historical Statistics of the United States (Carter et al., 2006); for Vietnam and later, the Statistical Information Analysis Division (SIAD) of the Defense Manpower Data Center. Personnel are the number serving worldwide during the conflict. For the First Gulf War, deaths exclude 1,565 non-theater deaths. Surviving veterans are calculated as the difference between personnel and killed.

Conflicts have varied widely in terms of overall scope, with World War II (1941–1945) the largest conflict to date in terms of U.S. participants.³ The most deaths occurred during the American Civil War (1861–1865), if fatalities on both sides are counted; otherwise World War II was also the deadliest. Recent conflicts, especially the two following the Vietnam War (1964–1972), have been more limited in scope.

Soldiers serving during the Civil War (1861–1865), especially on the Confederate side, were the most likely in history to have died or been wounded. The indicators in Table 1 reveal that nearly 20 percent of surviving veterans had physical war wounds. Because mental health trauma appears to have been a signature combat ailment in each historical era (Institute of Medicine, 2010), the share of surviving veterans with either physical or mental wounds was probably higher still. In other conflicts, the proportion of survivors with war wounds has fluctuated between 2 and 6 percent, averaging 2.5 percent.

In recent conflicts, most notably the wars in Afghanistan (2001–present) and Iraq (2003–2011), the share of wounded soldiers per fatality has risen. This statistic measures roughly how likely it is that a service member will survive his or her wounds. While relatively high during the Revolutionary War (1775–1983) and the War of 1812 (1812–1815), this measure halved during the early mechanization of war in the 19th century. It rose during the World Wars and rose further during the Korean (1950–1953) and Vietnam (1964–1972) conflicts before temporarily dropping during the brief and largely airborne First Gulf War (1991). In Iraq and Afghanistan to date, the statistic is nearly 7 wounded per death. Improvements in emergency medical care and more rapid evacuation by air to trauma centers are responsible for the improvements in survival probability (Tanielian and Jaycox, 2008).

Despite the growth in the probability of surviving wounds, the number of wounded as a share of surviving veterans has fallen recently, from about 4.4 percent during the World Wars to 1.8 percent in Korea (1950–1953), Vietnam (1964–1972), and Iraq (2003–2011) and Afghanistan (2001–present). This may reflect an increasing mechanization or automation of warfare, or a force reconfiguration toward more support units and fewer combat units, or both. Other things equal, a more limited share of the wounded among surviving veterans should reduce per-capita need and compensation. But a greater intensity of nonfatal harm among wounded survivors, as implied by their increased survival probability, would clearly work against that.

The severity of wounds is no doubt a major determinant of the future survival of veteran cohorts, but the latter is foremost the product of the prevailing mortality environment, which has improved markedly over time. Figure 1 plots survivorship curves for veterans cohorts starting with the War of 1812 (1812–1815), where the horizontal axis shows years from the end of the conflict. As mortality rates have fallen in the U.S. in general, survivorship for each successive cohort of veterans has improved. While it took only about 20 years for survivorship to dwindle to half of all veterans from the War of 1812 and the Mexican War (1846–1848), the half-life, or the time it took for the number to halve, was 40 years for the Civil War (1861–1865) cohort, more than 45 years for veterans of World War I (1917–1918), and higher still for later cohorts. Today the median veteran is expected to live for more than 60 years following the end of hostilities. These survivorship projections roughly match cohort mortality forecasts provided by the Social Security Administration (Bell and Miller, 2005).

Figure 1.

The life cycles of veterans of major U.S. wars

Notes: Data are the proportion of veterans surviving each year by period of service, constructed as the ratio of veterans in the given year divided by the maximum over all years. Some survivorships are less than unity because conflicts are ongoing. For the Spanish American War, veteran population counts are volatile for unknown reasons. The sources are Tables Ed 245–261 in the Historical Statistics of the United States (Carter et al., 2006), and projections from the VetPop 2007 model of the U.S. Department of Veteran Affairs.

3. The budgetary costs of wars and veterans

In this section, I present data on the budgetary costs of war associated with total spending on veterans’ benefits in several formats in order to reveal key elements of their nature. In addition to veteran population statistics, the Historical Statistics of the United States also contain data compiled by the U.S. Department of Veterans Affairs (VA) and its predecessor agencies that reveal VA spending on particular cohorts of veterans, sometimes broken down by type of spending.

3.1. Annual benefits in 2008 dollars

Figures 2, 3, and 4 plot real inflation-adjusted veterans’ benefits in 2008 dollars for ten major U.S. wars in each calendar year. Patterns in annual spending are interesting because they reveal elements of political economy that have played important roles in veterans’ benefits through U.S. history.

Figure 2.

Real spending on veterans of major U.S. wars, 1783-1860

Notes: Data are nominal spending on veteran cohorts deflated or inflated by the CPI to produce 2008 dollars. The sources are Tables Ed 324–336 and Cc 1–2 in the Historical Statistics of the United States (Carter et al., 2006).

Figure 3.

Real spending on veterans of major U.S. wars, 1861-1917

Notes: Data are nominal spending on veteran cohorts deflated or inflated by the CPI to produce 2008 dollars. The sources are Tables Ed 324–336 and Cc 1–2 in the Historical Statistics of the United States (Carter et al., 2006).

Figure 4.

Real spending on veterans of major U.S. wars since 1918

Notes: Data are nominal spending on veteran cohorts deflated or inflated by the CPI to produce 2008 dollars. The sources are Tables Ed 311–323, Ed 324–336, Ed 351–371 and Cc 1–2 in the Historical Statistics of the United States (Carter et al., 2006), and author’s calculations and forecasts as described in the text.

Figure 2 shows real costs for the American Revolution (1775–1783), the War of 1812 (1812–1815), and the Mexican War (1846–1848), none of which ever exceeded $80 million in 2008 prices in a year. As discussed by Glasson (1918), pensions for Revolutionary War veterans at first were relatively small, limited primarily to officers but also some others with service-related disabilities. The service-pension act of 1818, originally suggested by the Monroe Administration as a response to perceived need, expanded benefits massively and unexpectedly. Federal budget surpluses following the War of 1812 prompted their actions, but Monroe and others vastly underestimated the number of surviving Revolutionary War veterans. The rapid growth in spending visible in Figure 2 triggered a string of legislative fixes that first restricted and then again expanded benefits. Spending on Revolutionary War veterans and their survivors fluctuated but remained high until diminishing rapidly after 1850, some 67 years after the end of the conflict. The wars with Britain and Mexico similarly resulted in large outlays with long right tails, but it was not until the General Law pension system of 1862 and subsequent legislation was passed that spending expanded greatly (Glasson, 1918; Linares, 2001).

The U.S. Civil War (1861–1865) was a watershed event in the development of the modern military compensation system in the U.S., a fact to which Figure 3 attests. The Disability Act of 1890 actually extended benefits to veterans based on length of service, rather than only on specific service-related disabilities (Costa, 1998; Linares, 2001). Under that law, annual spending on Civil War veterans’ compensation swelled to almost $4 billion in today’s dollars, and all compensation represented almost 30 percent of the federal budget at the time (Costa, 1998, p. 197). Again, these benefits were extremely long-lived, dropping under $1 billion only after 1935, fully 70 years after the conflict.

Benefits for veterans of the Spanish-American War (1898) were limited until expansions in the 1920s and 1930s following the creation of the Veterans’ Bureau in 1921.⁴ The downward spike in benefits and its reversal in the mid 1930s was the result of wrangling between Congress and FDR,⁵ presumably over the massive budget deficits of the Great Depression. Spending diminished gradually from a peak of $2 billion in 1939 to less than a quarter billion by 1970.

Figure 4 shows real spending on veterans of World War I (1917–1918) and later conflicts. The picture is dominated by the massive spike in benefits on the World War II (1941–1945) cohort immediately after hostilities ceased, reaching $60 billion in today’s prices in 1947. These short-lived flows, which totaled roughly $6.7 billion in current dollars, reflect the unprecedented educational transfers of the midcentury G.I. Bill (Bound and Turner, 2002; Stanley, 2003). These alone accounted for $3.6 billion or about 54 percent of all spending on the World War II cohort in 1947 (OMB, 2009, Table 11.3). Benefits dropped rapidly afterward, reaching a nadir of $16.5 billion in 1956 before climbing to another peak of $37.3 billion in 1975, thirty years after the end of hostilities. The Korean War (1950–1953) cohort also saw a massive spike in their benefits around 1956, probably the last year of coverage under the G.I. Bill, which lapsed in 1955 (Institute of Medicine, 2010). Like World War II veterans, their real transfers peaked again around 1975, 23 years after hostilities.

The World War I (1917–1918) cohort’s peak year of real benefits was probably 1962, some 44 years after the end of the war. The graph depicts another peak in 1971, another 9 years later, but data after 1970 are estimates based only on compensation and pensions by cohort. Still, the uptick measured in 1970 was real.

For the Vietnam (1964–1972) cohort, benefits were monotonically increasing almost universally throughout the historical period. I forecast they will peak at just under $30 billion per year around 2020, 45 years after the end of the war. This is because according to the VA’s VetPop forecasts, the annual mortality rate of the cohort only then reaches the 3.5 percent average annual rate of growth in real costs per capita found in data for that cohort since 1980. First Gulf War (1991) costs, which are considerably more uncertain, are expected to peak around 2045 for the same reason.

3.2. Veterans’ benefits in present value

The very long-lived nature of veterans’ budgetary costs implies that the right way to compare them to short-lived costs of war like military spending is by calculating their present discounted value. For each major war, I deflate spending on veterans in each year past the start of the conflict by a discount factor that reflects the cumulative force of compounded interest measured from the middle year of the conflict. I take as my nominal discount rate the average interest rate on long-term government bonds. This produces a series of present discounted values of future spending on veterans for each conflict in current dollars, which are comparable to the current dollar totals of military spending during the conflict. When comparing different conflicts, I can inflate both of these using the CPI to recover real values in 2008 dollars. Details are described in Appendix A.1.

3.2.1. The life cycle of veterans’ benefits

First, it is useful to examine the time path or life cycle of veterans’ benefits by constructing the cumulative present value of benefits remaining to be spent as of each year past the end of each conflict. This number can be expressed as a proportion of the total present value of veterans’ benefits for that conflict. The result is a statistic similar to the survivorship probability measure in a life table. Like survivorship, this statistic reveals how long-lived the costs are, in a present value sense, of treating and compensating veterans of wars. As with cohort survival, a natural focal point is the year at which the statistic reaches 0.5, which is the half-life of the benefits.

Figure 5 plots this survivorship series for costs against years since the end of the conflict for each of the 10 major U.S. wars for which I have data and forecasts. There is similarity in that all curves slope upward toward the asymptote at unity, but there is also a considerable amount of heterogeneity in the slopes themselves. The schedules for World War II (1941–1945) and the Korean War (1950–1953), at the far left of the figure, are very steeply sloped at first, reflecting the large G.I. Bill disbursements early in the life spans of those cohorts. In both cases, their half-lives, where 50 percent of spending has been disbursed, occur relatively quickly, after about 20 years have elapsed. At the other end, spending on veterans from the War of 1812 (1812–1815), the Mexican War (1846–1848), and the Spanish American War (1898) was all considerably delayed due to the vagaries in development of military pension laws. For these cohorts, half-lives were more like 50 or more years. In between these extremes lie the other 5 conflicts, with no clear pattern governing their relative positions. Vietnam (1964–1972) and the First Gulf War (1991) reach their half-lives at 32 and 41 years respectively; World War I (1917–1918) reached its at 38 years. Although veterans’ survivorship schedules are advancing rightward, as remaining life expectancies increase with mortality declines, the life cycles of costs have not necessarily followed this pattern at all.⁶ For all 10 conflicts, the average half-life of costs is 37.5 years from the end of hostilities; among the 6 cohorts prior to World War II subject to minimal cost forecasting, the average was 44.2 years.

Figure 5.

The life cycles of present-value veterans’ benefits for major U.S. wars

Notes: The data show at each year following the end of hostilities the percent of the total present value of past, present and future veterans’ benefits associated with the conflict that remain to be paid out. For example, a figure of 0.5 in year 20 means that 20 years after the conflict, half of the total present value of costs have been paid out. Present values are calculated from the perspective of the end of the conflict. For sources, see the text.

Compared to the patterns in veterans’ survivorship seen in Figure 1, trends in the longevity of veterans’ benefits follow no clear pattern. This seems odd at first given how the two should be closely related. The differences between them must reflect changes in veteran compensation regimes, which apparently are large enough to overwhelm the effects of monotonically increasing survivorship.⁷ In light of this, the shorter life span of benefits following World War II (1941–1945) and Korea (1950–1953), while real, is also somewhat misleading. Half-lives are shorter because these benefits were front-loaded with massive educational outlays under the G.I. Bill. It would be a mistake to construe such a trend, which may be repeated now under the Post-9/11 G.I. Bill expansion, as indicating a reduction in the net fiscal burden of caring for veterans. Rather, it would represent a net increase in the fiscal burden that also moves its center of gravity forward in time.

3.2.2. Veterans’ benefits as a share of total budgetary costs

Second, it is revealing to compare the present discounted value of veterans costs associated with military conflict to the direct military costs. Because I am using the same deflator to translate current dollars into real dollars, the results will be independent of whether I inflate historical statistics or not.⁸ I present results in billions of 2008 dollars, in order to provide easy comparability to the estimates of Stiglitz and Bilmes (2008) for the wars in Iraq (2003–2011) and Afghanistan (2001–present).

Table 2 presents historical war costs for the 10 conflicts shown in Figure 5 plus estimates produced by Stiglitz and Bilmes. The left side of the table is similar to data presented by Nordhaus (2002), from whom I take the datum on the direct military costs of the First Gulf War (1991). The differences result from my using a later version of the Historical Statistics, in which some updated estimates are provided by Goldin (1980).

Table 2.

Budgetary costs of major U.S. wars

	Direct military costs		Historical and projected veterans’ benefits
Conflict	Millions of current dollars	Billions of 2008 dollars	Millions of current dollars	Billions of 2008 dollars	Present value in billions of 2008 dollars	Total war costs in billions of 2008 dollars	Veterans’ benefits share
Revolutionary Wars (1775-1783)	100	1.8	70	762	0.1	1.8	0.032
War of 1812 (1812-1815)	89	1.1	49	483	0.0	1.1	0.022
Mexican War (1846-1848)	82	2.1	64	662	0.2	2.4	0.091
Civil War (1861-65)	3,334	57.1
Confederate	1,032	17.7
Union	2,302	39.5	8,576	75,962	30.2	69.6	0.433
Spanish American War (1898)	270	7.0	5,767	27,895	25.5	32.5	0.785
World War I (1917-1918)	32,700	466.3	96,181	259,279	305.7	771.9	0.396
World War II (1941-1945)	360,000	4,480.3	542,308	742,833	1,373.1	5,853.4	0.235
Korea (1950-1953)	50,000	406.2	201,822	178,189	215.6	621.8	0.347
Vietnam (1964-1972)	140,600	869.9	1,806,737	648,245	554.8	1,424.7	0.389
First Gulf War (1990-1991)	61,000	96.4	2,259,350	407,165	371.9	468.3	0.794
Iraq and Afghanistan (OEF/OIF) (2001-)	1,559,000	1,559.0			673.3	2,232.3	0.302

Sources: Historical Statistics of the United States (Carter et al., 2006), Nordhaus (2002), Stiglitz and Bilmes (2008), and author’s calculations. For the last, see the text. The present values of veterans’ benefits are calculated from the perspective of the midpoint of the conflict. From that point, future nominal veterans’ benefits are deflated by the cumulative force of nominal discounting. Then the total present value at the midpoint is inflated to 2008 dollars using the historical CPI from the Historical Statistics. Nominal and real dollar totals for veterans’ benefits paid by the former Confederate states following the Civil War exist in principle but are currently unavailable. Those for OEF/OIF are not reported by Stiglitz and Bilmes.

The center panel in Table 2 reveals the total and present value of veterans’ benefits in a variety of formats. In current dollars, veterans’ benefits literally explode over time, to no great surprise. In real terms, the sum total rises and dips along with the direct costs of military activities, proxying the scope of conflicts. The most useful comparison is the present value of veterans’ benefits measured from the midpoint of the war in 2008 dollars, in the third column of the center panel. This number is extremely small for early conflicts, for which any appreciable veterans’ benefits were a long way off, and considerably larger for more recent conflicts. It reaches $1.4 trillion for World War II (1941–1945), $555 billion for Vietnam (1964–1972), and $372 billion for the First Gulf War (1991). Out of the $3 trillion in costs estimated by Stiglitz and Bilmes (2008), $673 billion represents the budgetary share of veterans’ benefits.

The rightmost panel in Table 2 shows the total budgetary costs of these eleven major U.S. wars, and the share of those total present-value costs attributable to veterans’ benefits. The total cost of World War II (1941–1945) is almost $6 trillion dollars in today’s prices, while the share attributable to veterans’ benefits is 23.5 percent. The First Gulf War (1991) stands out as a very inexpensive military operation that is also associated with a very large amount of veterans’ benefits, which account for 80 percent of total costs. This was not unprecedented; the Spanish-American War (1898) was also relatively cheap when fought but very expensive in terms of veterans’ benefits, which account for 78.5 percent of all costs. Across all eleven conflicts, the average share attributable to veterans’ benefits is 35 percent; for conflicts since World War I (1917–1918), the average is 41 percent.

3.2.3. Relative costs of war and veterans and implicit debt

Finally, it is useful to examine the magnitudes of war costs and participation relative to population, involved personnel, and to GDP. Table 3 reports the share of the resident population involved in each war, the direct military costs and the present value of veterans’ benefits per involved military personnel, and the ratios to GDP of costs and initial federal debt held by the public.

Table 3.

Relative budgetary costs and impacts of major U.S. wars

		Ratio to personnel:			Ratio to GDP: (measured at)
Conflict	Personnel per resident population	Direct military costs in 2008 dollars	PDV of veterans’ benefits in 2008 dollars	Direct military costs (start)	PDV of veterans’ benefits (end)	Initial federal debt (start)	Final federal debt (end)
Revolutionary Wars (1775-1783)	0.077	8,113	267	0.589	0.014
War of 1812 (1812-1815)	0.035	3,807	87	0.107	0.002	0.061	0.137
Mexican War (1846-1848)	0.004	27,232	2,723	0.040	0.003	0.008	0.018
Civil War (1861-65)	0.096	17,434		0.714	0.211	0.017	0.239
Confederate	0.114	16,613		0.221
Union	0.090	17,828	13,632	0.493	0.211
Spanish American War (1898)	0.004	22,779	83,182	0.016	0.059	0.074	0.074
World War I (1917-1918)	0.045	98,469	64,559	0.628	0.296	0.040	0.115
World War II (1941-1945)	0.117	278,062	85,218	2.841	0.499	0.359	0.942
Korea (1950-1953)	0.036	71,020	37,695	0.170	0.069	0.736	0.570
Vietnam (1964-1972)	0.044	99,482	63,450	0.212	0.074	0.384	0.252
First Gulf War (1990-1991)	0.009	43,338	167,134	0.011	0.039	0.387	0.414
Iraq and Afghanistan (OEF/OIF) (2001-)	0.007	742,381	320,619	0.152	0.047	0.329

Sources: See the notes to Tables 1 and 2. Prior to 1940, debt is federal public debt, from Historical Table Ea587; since 1940, it is federal debt held by the public, from Historical Table Ea728. GDP and debt are measured either at the beginning or end of each conflict. The Spanish American War began and ended in the same year.

Participation has varied enormously over these eleven conflicts, ranging from a low of 0.4 percent involved during the Mexican War (1846–1848), to 11.7 percent during World War II (1941–1945). Real costs per soldier have also varied widely and loosely track participation, as one would expect if both measures index the scope of war. But the real costs of the wars in Iraq (2003–2011) and Afghanistan (2001–present) per service member are vastly higher than in historical conflicts. Direct military spending per soldier in Iraq or Afghanistan is near the one million dollar mark at three times the quarter million dollars spent per World War II service member. The present value of veterans’ benefits per service member in Iraq or Afghanistan is more than $300,000.

Costs as a share of pre-conflict GDP track the participation rate more closely and provide a better index of scope than do costs per participant. Another natural comparison is between the ratio of war costs per GDP and the ratio of debt held by the public to GDP, which is measured at the beginning and the end of each conflict in the last two columns in Table 3. Other things equal, war costs raise indebtedness, but not necessarily one-for-one.⁹ For example, the unprecedented level of direct military spending during World War II (1941–1945), equal to almost three times initial GDP but spread over a period of five years, raised debt held by the public from 36 to 94 percent of GDP.

By contrast, the present value of future veterans’ benefits as a share of GDP is directly comparable to the debt-to-GDP ratio at the end of each conflict. The former is a measure of implicit debt, the latter explicit debt, both scaled to income. An analogous measure is the unfunded future liabilities of Social Security or Medicare, which respectively are on the order of $5.3 and $13.4 trillion today,¹⁰ or roughly 40 and 90 percent of GDP. The implicit debt associated with veterans’ benefits has tended to be much smaller, on the order of 4 percent of GDP, except in the cases of World War I (1917–1918), World War II (1941–1945), and the Civil War (1861–1865), after which implicit debts were between 20 and 50 percent of GDP. This analogy is also appropriate for another reason. Like Social Security and Medicare spending, compensation and medical care for veterans are more like transfers of resources rather than additions to income.¹¹ In this regard, the share of total war costs attributable to veterans is likely to understate their net fiscal burden relative to that of direct war costs, because the latter are believed to add to GDP (Barro, 1981; Hall, 2009; Ramey, 2011a,b).¹²

4. Estimating the economic costs of war-related injury and death

The budgetary costs of war include the costs of veterans’ benefits: compensation, pensions, medical treatment for service-related trauma, and survivors’ benefits. But these budgetary costs may or may not reflect the economic costs of war-related injuries and deaths. The modern system of VA disability compensation, in which service-related disability is scored as a percentage of total disability, is based on assessments of wounded veterans’ ability to work. Payments are calibrated to replace lost labor market earnings (Institute of Medicine, 2007; Stiglitz and Bilmes, 2008).

But a prevailing view in health economics is that the full cost to an individual of disability probably exceeds the value of lost wages alone. If the cost of a harm is defined as the amount an individual is willing to pay in order to avoid the harm, then the true cost will include not only the foregone earnings but any additional cost in reduced quality of life that remains after recovering the earnings lost to disability. The loss of a limb, for example, reduces future earnings but also probably reduces enjoyment of leisure activities, a harm that would persist even after lost earnings were recouped. In modern periods, estimates of the willingness to pay to avoid death or a particular injury, called the value of a statistical life (VSL) and the value of a statistical injury (VSI), are much larger than is implied by the capitalized value of lost future wages (Viscusi, 1993; Viscusi and Aldy, 2003). From this perspective, the economic costs of deaths and injuries depend on the VSL and VSI.

Given enough information, it might be possible to characterize and cost out each service-related injury or death for every conflict.¹³ Data on specific conditions related to service are no doubt available in some form for many cohorts of veterans from the VA. But a more expeditious strategy is instead to use survey-level measures of the VA disability rating.

The VA disability rating is a percentage running from 0 to 100 that is meant to reflect the amount of usual activity the veteran cannot perform due to service-related disability.¹⁴ In principle, 100 minus the disability rating measures something akin to what health economists call a quality-adjusted life year: what the veteran has left to enjoy. If one is willing to assume that a life year spent with a VA disability rating of x percent is worth 100 – x percent of a disability-free life year, and if the VA disability rating averages x throughout life, then the total gross cost of service-related disability borne by that veteran is x times the value of a statistical life (Stiglitz and Bilmes, 2008). In remainder of this section, I construct estimates of the costs of injury and death associated with major U.S. wars using this methodology.

4.1. Cohort-specific trends in VA disability ratings

The primary components of the economic costs of disability and death are the relative magnitudes of each among veterans. Deaths are easy to measure while disability is more difficult. Past studies have assumed that all service members who were reported as having “wounds not mortal” in the official military statistics were 50 percent disabled, while no other veterans suffered any disability (Goldin and Lewis, 1975; Glick and Taylor, 2010). For conflicts in the distant past, this assumption is difficult to assess. For recent wars, I can measure self-reported VA disability ratings for veteran cohorts using publicly available survey data and directly compare those statistics to numbers wounded in action.¹⁵

Table 4 presents weighted sample statistics from four cross-sectional snapshots of war veterans taken since 1980. Details about the data are provided in Appendix A.2. For each conflict, the first three rows report the VA disability rating averaged over all surviving veterans, the average VA disability rating among veterans with a positive rating, and the share of the cohort with a positive rating. The next four rows list weighted headcounts, the average age of the cohort, and the sample size. There is a limited degree of uniformity across veteran cohorts in their average VA disability level in these data, which has hovered between 2 and 6 percent disabled. Veterans with a positive rating typically experience about 35 percent disability, while they have accounted for 10–15 percent of their cohort, sometimes less as in the case of the Korean War (1950–1953). But behind this rough average that is immediately evident there is considerable variation over ages and cohorts.¹⁶

Table 4.

VA disability ratings among wartime veterans

		Dataset and year
Conflict	Measure	SAV 1983	NSV 1992	NSV 2001	ACS 2008
World War I (1917-1918)	Average VA disability, all survivors	4.1
	Average VA disability, positive rating	57.3
	Share with a positive rating	7.1
	Veterans with a positive rating	4
	All surviving veterans	56
	Average age	87.1
	Sample size	56
World War II (1941-1945)	Average VA disability, all survivors	4.4	3.6	4.0	4.4
	Average VA disability, positive rating	34.0	32.1	32.8	35.5
	Share with a positive rating	12.8	11.3	12.2	12.4
	Veterans with a positive rating	356	931,480	629,282	329,148
	All surviving veterans	2779	8,224,585	5,149,093	2,661,782
	Average age	63.1	69.5	78.0	84.4
	Sample size	2,779	4,016	4,565	32,692
Korea Average (1950-1953)	VA disability, all survivors	4.7	1.9	3.4	3.6
	Average VA disability, positive rating	42.9	29.3	34.8	34.9
	Share with a positive rating	11.0	6.5	9.7	10.3
	Veterans with a positive rating	12	283,663	354,050	270,209
	All surviving veterans	109	4,350,228	3,641,419	2,612,820
	Average age	56.2	59.6	69.0	75.9
	Sample size	109	2,111	3,085	33,342
Vietnam (1964-1972)	Average VA disability, all survivors		2.7	5.3	6.3
	Average VA disability, positive rating		29.4	35.7	42.0
	Share with a positive rating		9.2	15.0	15.0
	Veterans with a positive rating		778,129	1,252,556	1,100,773
	All surviving veterans		8,477,848	8,361,037	7,358,856
	Average age		44.0	53.5	60.4
	Sample size		3,393	7,063	87,203
First Gulf War (1990-1991)	Average VA disability, all survivors			4.8	6.3
	Average VA disability, positive rating			29.1	35.4
	Share with a positive rating			16.6	17.7
	Veterans with a positive rating			494,997	626,495
	All surviving veterans			2,989,579	3,534,460
	Average age			32.8	38.6
	Sample size			2,129	33,718
Iraq and Afghanistan (OEF/OIF) (2001-)	Average VA disability, all survivors				2.9
	Average VA disability, positive rating				33.1
	Share with a positive rating				8.6
	Veterans with a positive rating				148,559
	All surviving veterans				1,725,203
	Average age				27.4
	Sample size				14,588

Sources: Author’s calculations from data in the 1983 Survey of Aging Veterans (SAV), the 1992 and 2001 National Surveys of Veterans (NSV), and the 2008 American Community Survey (ACS). Veterans of multiple wars or periods are counted as belonging to the earliest war. The 1983 SAV did not have sample weights but was designed to be representative; italics denote unweighted raw counts of veterans in the survey.

Across all cohorts, there is a clear upward trajectory in the average VA disability rating over the life cycle, which appears to decelerate after age 60 but continues to increase. As revealed by the head counts of disabled veterans, which appear below the prevalence indicators, the aggregate level of need ultimately declines as the cohort dies off over time. Aggregate need drives the real dollars spent on veterans shown in Figures 2–4, which explains the hump-shaped trajectory. But per capita need appears to increase continuously throughout age for these cohorts, driven by increases along both extensive and intensive margins. This is the opposite of what one would expect to happen naturally through attrition if mortality rates were increasing in the level of VA disability. That both the prevalence and severity of disability appear to increase within a cohort over time suggests either that the course of aging is worsening service-connected conditions, or revealing latent conditions, or that the generosity of the VA disability ratings system is increasing (Angrist et al., 2010).

There also is interesting variation across conflicts or cohorts, especially along the extensive margin, or the share of veterans with a positive VA disability rating. Successively younger cohorts of veterans seem to have higher VA disability ratings on average than older cohorts did, both overall and at comparable ages. This appears to be driven more by a greater prevalence of positive ratings within the cohort rather than a higher level or intensity of VA disability among those with positive ratings. Interestingly, the patterns in VA disability in Table 4 do not match the trends in percent wounded across conflicts that were presented in Table 1. In World War I (1917–1918) and World War II (1941–1945), the latter was 4.2 percent; in Korea (1950–1953) and later, it dropped to 1.8 percent and remained there. By comparison, the average VA disability rating has risen, reaching 6.3 percent among Vietnam (1964–1972) and First Gulf War (1991) veterans, the latter at a far younger age. If the VA disability rating is the gold standard of measurement, it would appear that the incidence of war wounds in the defense department statistics is a poor measure of the average health impact of warfare. Given how the Pentagon’s definition of nonfatal casualties has shifted over time, as discussed by Goldberg (2010), this is perhaps unsurprising.¹⁷

A similar picture emerges when I examine VA disability among veterans who have served only during peacetime. These statistics are reported in Table 5 for cohorts who served only during one of three recent interwar periods. Average VA disability ratings for these peacetime cohorts are not zero, with a minimum average rating of 0.5 percent, but these ratings also average only 1 or 2 percent rather than the 6 percent seen among war cohorts in Table 4. This is driven by considerably lower prevalence of any VA disability than we see among war veterans, here never above 8.2 percent. Accidents or combat could result in service-related disabilities; these patterns imply that wartime service effectively raises the risk of disability, presumably through combat exposure. Both the share disabled, and to a lesser extent the intensity of disability among the disabled, rise within these peacetime cohorts as they age, just as within wartime cohorts. And another similarity emerges: successive cohorts have higher average ratings, which also appear to be driven by increased prevalence.

Table 5.

VA disability ratings among peacetime-only veterans

		Dataset and year
Conflict	Measure	NSV 1992	NSV 2001	ACS 2008
Between WWII and Korea (1947-1950)	Average VA disability, all survivors	0.5	0.7	1.6
	Average VA disability, positive rating	21.8	27.8	29.0
	Share with a positive rating	2.5	2.4	5.6
	Veterans with a positive rating	1,291	1,031	9,438
	All surviving veterans	52,033	42,792	167,376
	Average age	61.8	71.5	78.9
	Sample size	25	30	2,009
Between Korea and Vietnam (1955-1964)	Average VA disability, all survivors	0.6	1.2	1.4
	Average VA disability, positive rating	25.0	34.8	28.7
	Share with a positive rating	2.4	3.4	4.8
	Veterans with a positive rating	53,798	74,280	115,482
	All surviving veterans	2,275,898	2,188,335	2,418,879
	Average age	53.4	62.8	69.4
	Sample size	744	1,473	30,838
Between Vietnam and First Gulf War (1975-1990)	Average VA disability, all survivors	1.1	2.8	2.8
	Average VA disability, positive rating	21.4	34.8	34.4
	Share with a positive rating	5.3	8.2	8.2
	Veterans with a positive rating	190,966	221,610	251,184
	All surviving veterans	3,631,572	2,706,919	3,080,714
	Average age	30.0	40.3	47.3
	Sample size	1,149	1,624	264,692

Sources: Author’s calculations from data in the 1992 and 2001 National Surveys of Veterans (NSV), and the 2008 American Community Survey (ACS). Veterans of multiple wars or periods are counted as belonging to the earliest war. For example, all veterans identified as belonging to the period between the Korean and Vietnam wars served then but could not have served during any conflict nor during any earlier peacetime period.

The evolution of VA disability over the life course motivates extrapolation to forecast average VA disability trends. Assuming a floor on the average VA disability rating of 0.5 percent, which is the smallest measure within samples shown in Tables 4 and 5, I extrapolate average VA disability ratings by age for the five wartime cohorts starting with World War II assuming piecewise linear trajectories through age. (Data are available upon request.) Strong age effects in the average VA disability rating raise the question of what rate to use in determining the total lifetime impacts of war on the health of a cohort. If there were known bottlenecks in assigning VA disability ratings that drove the age-related trajectory, one could argue that the maximum observed rating is the appropriate measure of lifetime harm. But the delays of several decades apparent in the data seem more likely to have emerged from the latency of health impacts. If latency is important, the lifetime impact on health is arguably best captured by an average of VA disability ratings over all ages, ideally weighted by survivorship. I assume VA disability rates are flat after age 90, and I apply the survivorship probabilities shown in Figure 1 to recover the following lifetime average VA disability ratings for wartime cohorts, which are also shown in Table 6.

Table 6.

Costs of death and injury resulting from major U.S. wars

			Value of Statistical Life (VSL) from Costa and Kahn (2004)			Value of Statistical Life (VSL) from Viscusi and Aldy (2003)
Conflict	GDP per capita in 2008 dollars	Lifetime average VA dis- ability rating	VSL in 1,000’s of 2008 dollars	Costs of death billions of 2008 dollars	Costs of disability billions of 2008 dollars	VSL in 1,000’s of 2008 dollars	Costs of death billions of 2008 dollars	Costs of disability billions of 2008 dollars
Revolutionary Wars (1775-1783)	1,516		93	0.4		1,612	7.1
War of 1812 (1812-1815)	1,902		131	0.3		1,805	4.1
Mexican War (1846-1848)	2,519		199	2.6		2,078	27.6
Civil War (1861-65)	3,147		278	173.1		2,322	1,445.5
Confederate	3,147		278	71.7		2,322	599.1
Union	3,147		278	101.3		2,322	846.4
Spanish American War (1898)	5,317		610	1.5		3,018	7.4
World War I (1917-1918)	7,628		1,049	122.2		3,615	421.2
World War II (1941-1945)	16,542	1.8	1,885	764.2	532.9	5,324	2,158.3	1,505.2
Korea (1950-1953)	16,910	1.7	2,353	86.1	227.3	5,383	196.9	520.1
Vietnam (1964-1972)	25,486	5.2	6,499	378.2	2,935.3	6,608	384.6	2,984.6
First Gulf War (1990-1991)	36,215	9.8	9,483	3.6	2,067.4	7,877	3.0	1,717.3
Iraq and Afghanistan (OEF/OIF) 2001-	49,021	9.4	14,935	80.3	2,940.6	9,165	49.3	1,804.5

Sources: See earlier tables. Income (GDP) and the value of a statistical life (VSL) are measured at the final year of the conflict. I specify two alternative time series of the VSL by extrapolating from the results of Costa and Kahn (2004) or Viscusi and Aldy (2003). Costa and Kahn (2004) measure the VSL directly for the U.S. between 1940 and 1980; I geometrically interpolate the VSL for intervening years, and I forecast from 1980 and backcast from 1940 using the GDP series and their preferred estimate of the income elasticity of the VSL, η_{V SL} = 1.5. Viscusi and Aldy (2003) place the median estimate of the VSL at $7 million in current dollars in the year 2000 and recover η_{V SL} = 0.5 based on their meta-analysis. I forecast and backcast their VSL from 2000 using the GDP series and η_{V SL} = 0.5. Lifetime average VA disability ratings apply to the entire cohort of surviving veterans (not just those injured) and are based on data and extrapolations shown in Table 4 and Figures 1 and as described in the text. The costs of death are the product of killed and the VSL. The costs of disability are calculated as the product of surviving veterans, the VSL, and the lifetime average VA disability rating.

4.2. Historical trends in the value of a statistical life

Stiglitz and Bilmes (2008) assume the value of a statistical life equals $7.2 million in 2007 dollars, which is the central estimate of $6.2 million in 2002 dollars used by the Environmental Protection Agency (Dockins et al., 2004) adjusted for inflation. The earlier estimate was itself an inflation-adjusted update of EPA’s earlier assumption of $4.8 million in 1990 dollars. All these figures fit within the relatively wide range of $4 to $9 million implied by U.S. labor market data on wage differentials associated with mortality risks (Viscusi and Aldy, 2003).

As I discuss in Appendix A.3, economic theory suggests the VSL ought to depend on the marginal utility of living relative to that of consumption or income. Two prior studies systematically assess how the VSL varies over space and time. Viscusi and Aldy (2003) conduct a meta-review of the literature on the VSL in U.S. labor market studies since the 1960s and in a cross section of countries. They estimate the income elasticity of the VSL, η_{V SL}, in the range of 0.5 to 0.6. Costa and Kahn (2004), who examine U.S. data between 1940 and 1980, report η_{V SL} to be around 1.5 to 1.7. In an interesting study of the decision to deploy costly military resources in order to reduce casualties, Rohlfs (2006) estimates a VSL in World War II, roughly $1 million in 2003 dollars, that is consistent with the findings of Costa and Kahn (2004). Although it seems likely that service members might have different attitudes toward risk than civilians, this result suggests that applying the same values of statistical life or injury to both groups is reasonable.

Uncertainty about the magnitude of η_{V SL} is a clear hindrance for studies of the value of health improvements or harms over long historical periods.¹⁸ Because there is uncertainty about η_{V SL}, I produce two sets of time series of the VSL, one based on the findings of Costa and Kahn (2004) where η_{V SL} = 1.5, and the other on Viscusi and Aldy (2003) with η_{V SL} = 0.5.

4.3. Estimates of economic costs of injury and death in 2008 dollars

Table 6 presents estimates of the economic costs associated with injury and death among service members for as many major U.S. wars as I have microdata. The economic costs of death are the product of the number of war fatalities and the estimated VSL at the end of the war the number of war deaths, which equals the aggregate willingness to pay to avoid them. Similarly, the economic costs of disability is the product of the number of surviving veterans, the lifetime average VA disability rating estimated for that cohort, and the VSL. The implicit assumption is that being x percent disabled is equivalent to losing x percent of a life.

These costs are very large regardless of which VSL baseline I use. This is because the levels of the VSL always tend to be high regardless of their income elasticity. The cost of World War II (1941–1945) deaths ranges from $764 billion to $2.2 trillion, or between roughly 20 and 50 percent of the direct military cost. In the case of the Civil War (1861–1865), the costs associated with deaths are several orders of magnitude greater than the direct military costs. Costs of deaths in Iraq (2003–2011) and Afghanistan (2001–present) total between $50 and $80 billion depending on on the VSL series used. The costs of disability are very large also because the VSL is high, and also because lifetime average VA disability rates among veteran cohorts are high, especially among the Vietnam (1964–1972) and later cohorts. Disability among the World War II cohort was less costly than deaths, presumably because the mortality rate was sufficiently high relative to VA disability rates among survivors. In Korea (1950–1953), however, survival had improved enough to push the costs of disability to around two or three times the cost of deaths. This trend has continued. For each of the last three conflicts starting with Vietnam, I estimate the costs of disability to be $2 trillion or more, a figure that dwarfs all associated budgetary costs and is large relative to other recent estimates. By comparison, Stiglitz and Bilmes (2008) estimate the total costs of disability to be between $250 and $367 billion,¹⁹ while Wallsten and Kosec (2005) project $100 billion. If the lifetime average VA disability rate among OEF/OIF soldiers were 1 percent rather than my extrapolation of 9.4 percent, then the costs of statistical injury would total between $200 and $300 billion, or about a ninth of the estimates shown in Table 6. If VA disability rates for this cohort were instead to remain at the level of 2.9 percent shown in Table 4, total costs of disability would range between $550 and $900 billion.

5. Conclusion

Since the Civil War, at least a third of the federal budgetary costs associated with warfare have been decidedly long-lived. Direct military spending still accounts for the majority of war-related spending, but veterans’ benefits represent a significant minority that follows a very different life cycle, namely the relatively long remaining life spans of veterans and their survivors. Even for short, seemingly cheap engagements like the Spanish American War or the more recent First Gulf War, the unfunded liability of future veterans’ benefits looms as a significant fiscal burden.

The half-life of the present value of veterans’ benefits tends to be 30 years or more. The front-loading of benefits through educational subsidies like the G.I. Bill tends to reduce the half-life of costs, but the net effect of such an expansion is to raise the fiscal burden, not reduce it. The committing of troops to a war zone has lasting implications for fiscal policy in addition to short-term impacts on the economy and tax revenues via direct military spending. Depending on the scope of the conflict, the unfunded obligation to pay future veterans’ benefits starting from the end of the conflict can range from 5 to 50 percent of GDP. This is a very large commitment.

It is sometimes argued that the increasing mechanization of war is motivated in part by the desire to save the lives and limbs of service members. Another perspective not at all unrelated to this is that it may be far cheaper in the long run to conduct even a seemingly expensive mechanized war if it conserves future veterans’ benefits. There are tactical considerations as well, but this insight may be relevant for understanding military budgeting strategies.

Less clear are what the net effects of transfers to veterans and their dependents may be. As opposed to military purchases, transfers do not affect income unless the propensity to spend them is higher than it is among taxpayers. At worst, higher tax rates needed to fund transfers are disincentives to work and save that may have a depressing effect on GDP. At its root, veterans’ benefits are designed either to compensate for service-related disability or for time served, or to treat injuries. One of the open questions is to what extent veterans are compensated for their injuries. Estimates of the private cost of war-related injuries and deaths seem to imply there are many uncompensaated wounds of war. Part of this is due to high estimates of the value of a statistical life, and part is due to high VA disability rates among recent cohorts of veterans.

This study has omitted any costs of war associated with the loss of civilian life or the destruction of capital. Nor has it assessed the costs borne by opposing sides in conflicts nor the costs of military inaction. As a result, I have little to say about the calculus of military conflict that might lead governments to war or peace, which is a clear gap in our knowledge and a subject of much interesting research (Blattman and Miguel, 2010; Garfinkel and Skaperdas, 2010). My goals have been to illustrate the life cycle of veterans’ benefits and assess their fiscal implications, which are large, and to explore the economic costs associated with war wounds.

Highlights.

• I examine historical trends in U.S. veterans’ benefits paid to war cohorts

• The budgetary costs of veterans’ benefits are small in any single year

• But veterans’ benefits are very long-lived, with typically only half paid out nearly 30 years after the end of the conflict

• Veterans’ benefits eventually account for between one third and one half of total war spending

Appendix A. Data and assumptions

Appendix A.1. Historical Statistics of the U.S.

My primary data source for information on wars and veterans are the Historical Statistics of the United States (Carter et al., 2006), Tables Ed1–399. Of these, the preponderance of statistics are derived from Tables Ed1–5, which list casualties for major U.S. wars; Tables Ed168–170, which show the estimated direct military costs of U.S. wars; and Tables Ed324–336, which present the expenditures of the Veterans Administration and its predecessor agencies by veterans’ period of service.²⁰ I augment the casualty lists for recent wars using data from the Statistical Information Analysis Division (SIAD) of the Defense Manpower Data Center, which are current through March 6, 2010.²¹ Military spending for the First Gulf War and for Operations Enduring and Iraqi Freedom (OEF/OIF) are taken from are taken Nordhaus (2002) and Stiglitz and Bilmes (2008) respectively, the latter the “realistic-moderate” forecast as of 2008.

Costs of veteran cohorts prior to 1866, which are primarily those associated with the Revolutionary War cohort, are given in a single lump sum, which I distribute across years according to other sources.²² Costs by conflict end in 1970, so I extrapolate costs for World War I and later wars to 1997 using Tables Ed311–323, which lists VA costs by function through 1998; and Ed351–371, which presents compensation and pension (VBA) costs by period of service through 1997.²³ For veterans of the Spanish-American War and earlier conflicts, who are effectively extinct by 1970, I model nominal costs as declining 15 percent annually, which is roughly the average rate of decline observed for extinct cohorts in the data. I also institute this rule for the World War I cohort after 1997.

For the World War II, Korean, Vietnam, and First Gulf War cohorts after 1997, I model costs per surviving veteran plus costs per survivor. A more precise method of forecasting would model additional parameters, such as the proportion of veterans who utilize health care and compensation programs and the intensity of utilization, where the latter is based on health status. But data limitations hamper that level of analysis, and evidence both from aggregate time series and cross sections suggest that focusing on usage per surviving veteran is a reasonable alternative, at least for cohorts past a certain age. I explore the cross-sectional trends that support this perspective in section 4.1 below.

Veteran populations are given for historical periods in Tables Ed245–261, and for future periods by the VetPop 2007 projections obtained from the VA website.²⁴ I measure total nominal costs net of survivor benefits per surviving veteran in 1997 as $1,531 for the World War II cohort, $890 for the Korean cohort, and $1,489 for the Vietnam cohort, and I assume a per capita cost of $1,500 for First Gulf War veterans. This assumption seems reasonable in light of similarities in disability rates between Vietnam and First Gulf War veterans that emerge in the cross-sectional surveys I examine in section 4.1.²⁵ Growth in real per capita costs after 1997 is 0 percent for the World War II cohort, 1.5 percent for the Korean cohort, and 3.5 percent for the Vietnam cohort, all extrapolated from trends in the data after 1980. For First Gulf War costs, which the Historical Statistics do not measure well, I assume the same rate of annual increase as found for the Vietnam cohort, and I backcast costs to 1991 by combining that assumption with past headcounts and the $1,500 per capita cost in 1997. I assume a future rate of CPI price inflation of 3 percent.

Numbers of surviving spouses by period of service are given by Tables Ed388 and Ed295 up to 1995 for the World War II, Korea, Vietnam, and Gulf cohorts. I decrement each series according to female death probabilities derived from an age-appropriate female cohort life table published by Social Security (Bell and Miller, 2005).²⁶ Levels of benefits are given by Tables Ed362 and Ed367, and I forecast 6 percent nominal growth beginning from their 1995 levels, roughly the average annual rate of historical increase.

Estimates of real GDP and population from 1790 are available in Tables Ca9–19. For years prior to 1790, I assume constant growth in GDP and population based on the period 1790 to 1810. After 2002, I supplement this series with recent BEA data, and I inflate the series to 2008 chained dollars. Data on CPI inflation with base period 1982–1984 are available in Table Cc1, which I append using recent BLS data. To obtain nominal discount rates, I use a composite of series in Tables Cj1192–1194, which are averages of yields on longer-term issues of either the U.S. government or municipalities. I splice gaps in the first series assuming fixed risk premia between the three. Between 1997 and 2009, I use the market yield on 10-year Treasury securities reported in the Fed’s H15 release, spliced with the same assumption. For forecast periods after 2009, I specify a 4.5 percent nominal discount rate and a 3 percent rate of CPI inflation, mirroring the assumptions of Stiglitz and Bilmes (2008). Data on federal government debt are provided in Tables Ea587 and Ea728.

In order to assess the total costs of relatively recent wars, one must project veterans’ benefits over a fairly long horizon. According to my forecast, nominal flows of costs associated with the First Gulf War are unlikely to cease before 2090. For OEF/OIF, I use the forecasts of spending on veterans provided by Stiglitz and Bilmes (2008).

Appendix A.2. Cross-sectional surveys of veterans

In addition to dollars of benefits, another key variable is the health status of surviving veterans. Past studies of historical conflicts have made do with rules of thumb based on official casualty statistics: the “Wounds Not Mortal” category (Goldin and Lewis, 1975; Glick and Taylor, 2010). These studies assume each nonfatal casualty represents a case of 50 percent disability.

My strategy improves on that method by examining VA disability ratings, which are traditionally designed to reflect work disability and can be interpreted as revealing the percentage reduction in quality of life (Stiglitz and Bilmes, 2008). With sufficient data, I could also calculate quality adjusted life years or some other measure, but the added difficulty would be specifying the change in QALYs or health status that is attributable to military service. The VA disability rating is designed to do exactly that.

In principle, all veterans examined by the VA during modern periods should have medical records that indicate their health status in some fashion. We also know from the CPE Union Army Dataset that medical records exist for Civil War era pensions. But a challenge is obtaining statistics that are representative of the average veteran’s condition. The Union Army data are neither universal nor weighted, for example.

As of this writing, I am unaware of any official statistics on disability ratings or health conditions for entire veteran cohorts. One of the issues is that under U.S. law, veterans have had to voluntary approach the VA and predecessor agencies in order to obtain VA disability ratings. Time trends in disability headcounts and payments suggest that veterans do not obtain disability ratings all at once, and even if they did, the average service-related disability rating for the cohort is in considerable flux during the first decades after the conflict (Institute of Medicine, 2010). While military service records would probably have recorded physical and possibly mental injuries, it seems unlikely that these records have ever been systematically assessed for entire cohorts of veterans.

What I currently have are several cross-sectional surveys of veterans in which respondents are asked their VA disability ratings and their period of service. These datasets include the 1983 Survey of Aging Veterans (SAV), the 1992 National Survey of Veterans (NSV), the 2001 National Survey of Veterans, and the 2008 American Community Survey (ACS). The first three surveys were commissioned by the VA, they exclusively cover veterans, and they range in size from about 3,000 in the SAV to 11,600 and 20,000 in the two NSVs. The 2008 ACS surveyed some 250,000 veterans out of a total population of around 23.4 million. Earlier waves of the ACS did not ask about VA disability ratings, and neither did the decennial Census. Coverage of VA disability status in various Current Population Surveys exists but is spotty and does not begin before 1985.

Appendix A.3. The value of a statistical life

The question is how these may vary over time. The former is equal to the lifetime sum of discounted streams of flow utility from consumption, u(c(t)), while the latter is u’(c(t)). Then the VSL for a particular cohort is given by

$$\mathit{VSL}=\frac{{\int }_0^T{e}^{-\rho t}u\left(c\right(t\left)\right)\mathit{dt}}{u^{\prime }\left(c\right(t\left)\right)},$$ (A.1)

where ρ is the time discount rate and T are years of remaining life. A standard treatment in the literature on the value of life in the aggregate (Nordhaus, 2002; Becker et al., 2005; Hall and Jones, 2007) is to assume flow utility is isoelastic plus a constant:

$$u\left(c\right(t\left)\right)=\frac{c{\left(t\right)}^{1-\gamma }}{1-\gamma }+b.$$ (A.2)

If one further assumes a constant interest rate equal to ρ, then optimal consumption is flat over the lifetime at some c̄, which is determined by the level of lifetime wealth, and equation (A.1) simplifies to

$$\mathit{VSL}=\frac{1-e^{-\rho T}}{\rho }\cdot \frac{u\left(\stackrel{‒}{c}\right)}{u^{\prime }\left(\stackrel{‒}{c}\right)}=\frac{1-e^{-\rho T}}{\rho }\cdot \left(\frac{\stackrel{‒}{c}}{1-\gamma }+b{\stackrel{‒}{c}}^{\gamma }\right).$$ (A.3)

For determining change in the VSL over time, the two additive functions of c̄ in equation (A.3) are important; the other elements are either constant or relatively inconsequential. As discussed by Hall and Jones (2007), the critical issue is the level of γ, which the inverse of the elasticity of intertemporal substitution in consumption, and it is the rate at which the marginal utility of consumption declines. When that rate is high, the marginal utility of consumption falls quickly as consumption rises over time. If γ is sufficiently high that the marginal utility of consumption falls faster than the marginal utility of life, the VSL rises. When γ is low, the reverse is true. And to a first approximation, the consumption elasticity of the VSL, is roughly equal to γ:

$$\eta \mathit{VSL}\equiv \frac{\partial \log \mathit{VSL}}{\partial \log \stackrel{‒}{c}}\approx \gamma .$$

Murphy and Topel (2006) provide a more extensive theoretical treatment and arrive at a similar result.

A problem is that γ is an unknown parameter. Empirical studies, which are typically based on financial data, do not agree on its magnitude, and its value in calibration exercises ranges between 0.5, 1, 2, and can be larger. As a result, it is better to rely on direct estimations of η_{V SL}.