The Longevity Legacy of World War II: The Intersection of GI Status and Mortality

Abstract

Purpose of the Study:

We examine hypotheses involving the potential health advantages of selection into military service and the potential health disadvantages associated with the experience of military service by comparing later-life mortality rates for veterans and nonveterans as well as among veterans based on their cohort of reentry into civilian life.

Design and Methods:

We use data on 3,453 men, including 1,496 veterans from the older men cohort of the National Longitudinal Surveys to estimate Cox proportional hazard mortality models. We distinguish between veterans and nonveterans and further classify veterans by age at exit while incorporating measures associated with military selection, health behaviors, and socioeconomic status.

Results:

Veterans who were discharged from the military at older ages have a mortality advantage relative to veterans discharged at younger ages. For the 1914–1921 birth cohorts, the mortality advantage for veterans who exited around age 30 is apparent for deaths before age 65, but rates equalize across all groups when deaths at older ages are included. These findings are robust to the inclusion of background characteristics, education, occupation, body mass index, smoking, marital status, and proxies for service deferments.

Implications:

Rather than focusing on a general health effect of military service, per se, future research should distinguish among individual traits; the nature, timing, and duration of exposures relative to life course stage; and the sociohistorical context of military service to expand our understanding of the differential health consequences of military service.

Even as it fades from the cache of lived memory, World War II (WWII) remains the defining event of the 20th century. Having reshaped the lives of those born in the first part of the century, the war arguably had its strongest effect on those called up to military service. In today’s volunteer armed forces, military service has become a career choice for young men and women, drawing disproportionately from groups with lower socioeconomic status (SES). However, for most of the 20th century, the military relied heavily on conscripts to provide the necessary numbers. The manpower needs for WWII were so high, the military drew from all social strata and across a relatively wide age range to build a military force that would exceed 16 million.

For men born in the first third of the 20th century, military service was a pivotal experience, one seldom included in research that argues the importance of earlier life experiences for later-life outcomes. Being in the military often uprooted people from their families and jobs, distanced them from their homes and communities, and asked them to respond to challenging circumstances in ways they never imagined. War mobilization provided a shared experience for the U.S. population, but military service fundamentally reshaped the life course of the subset of men who volunteered or were drafted. These men comprised from one-quarter to three-quarters of the men whose birth cohorts placed them in the specified age range and who were deemed eligible for service. Almost one-in-four men in the current U.S. populations are veterans, and for those aged 65 and older, the proportion exceeds one-half ( http://www.gallup.com/poll/158729/men-women-veterans.aspx ).

In this article, we take up the issue of mortality differentials among WWII veterans and between veterans and their age counterparts who were not in the military. Using data from the original cohort of older men in the National Longitudinal Surveys (NLS) and recently matched data from the National Death Index and the Social Security Death Index, we estimate Cox proportional hazards models to determine whether and how the later-life mortality rates for WWII veterans may have been affected by the cohort and life course features of their military service. We further limit our sample to white men, and distinguish among WWII veterans by their age at exit from the military.

The current research literature lacks consensus on whether or how military service shapes health and mortality. Arguments as to the negative health consequences of military service, particularly for those serving in or after Vietnam, are counterbalanced by claims of a salutary effect of the training, comradery, and confidence gained through the experience. A health advantage for veterans is also consistent with the epidemiological argument that the military selection criteria initially sort healthier men into the service, but over time, this health advantage will lessen and eventually the groups will converge ( Seltzer & Jablon, 1974 ).

We test hypotheses linked to arguments for both advantages and disadvantages between veterans and nonveterans and among veterans that persist to later life. If the epidemiological argument that positive health selection creates an initial mortality advantage is correct, then we should see a mortality advantage for veterans in earlier deaths, but not necessarily for all deaths. In contrast, if the argument of the higher stress and physiological consequences of war holds, we should see a higher rate of mortality for veterans at older ages. Further, if men who served at older ages were at a particular disadvantage, we should see a mortality advantage for those whose service ended in their early 20s rather than 30s among veterans. The selection argument is particularly relevant if we find a positive effect of military service, so if the effect of veteran’s status is displaced by controls for military selection, then the advantage is consistent with the initial sorting process rather than the experience. Finally, we will assess whether smoking, body mass index (BMI), alcohol consumption, and SES mediate any relationship between veteran status and mortality risk.

Conceptual Framework

Studying mortality from a life course perspective directs our attention to people’s exposure to specific experiences and events within the context of cohort membership ( Elder, 1974 ; Ryder, 1965 ; Settersten, 2003 ). The social meaning of these experiences can be shaped by broader historical and cultural contexts, and the consequences of these exposures can differ by one’s structural circumstances within the cohort ( Mannheim, 1928 ). Although cohorts are defined most often relative to year of birth, more generally cohorts organize people relative to when exposures begin or end.

Within the context of WWII, birth cohorts organize men according to whether and when they were called up for military service. Those called up at the same time also comprise cohorts of entrants, and those who were discharged from military service at the same time comprise cohorts of reentrants into civilian life. Although cohort structures relative to these three timing dimensions overlap, they also place military service within different temporal contexts. And while all age-eligible men registered within a common time frame, once quotas were filled the enlistment process, which proceeded unevenly across regions and started with January birthdays, was suspended until the next round; therefore, the exact age of entry into service depended on birth month, day, and year, quotas, and region.

If we distinguish age at entry from age at exit, we focus on different life course stages and different critical transitions ( George, 1993 ). Men who were older at induction were more likely to have finished their education, but were also more likely to have started their families and careers. For these men, service came after they transitioned to adulthood and were in the process of establishing themselves. Men whose service started at younger ages were more likely to be in the process of transitioning to adulthood. On the other hand, men who were older at enlistment were more likely to be older at discharge; they were returning to families and to employers who welcomed them back, whereas those who reentered civilian life at younger ages were likely returning to less structured circumstances. One positive feature of that timing was that they were in a better position to gain additional education, taking advantage of the opportunities for a college education. In addressing the longer term consequences of WWII military service, we conjecture that the timing of service matters, although we will be unable to distinguish whether the reason timing matters is because of the level of maturity and adult experience when the service was initiated, the duration and nature of exposure, the family and employment context into which they reentered, or any number of other possible mechanisms.

Answering these questions raises a number of additional difficulties. First, the study of veterans is necessarily the study of survivors, those who stayed alive until they were discharged as well as those whose discharge was not motivated by physical or mental condition. Second, any health effects of military service are likely to depend on the nature of the military experience. Those experiences differ according to whether they served in peace or during war, where they were stationed ( Elder, Clipp, Brown, Martin, & Friedman, 2009 ), the war they were fighting ( Angrist & Krueger, 1994 ; Teachman, 2004 ; Teachman & Call, 1996 ; Wilmoth, London, & Parker, 2010 ), the ages when they served ( MacLean & Elder, 2007 ), and the conditions to which they returned. These differences underscore the importance of making as many distinctions as possible among veterans. Third, studies of WWII veterans have limited data resources on which to draw. Longitudinal data projects were rare, and the few being conducted involved special populations defined by things like age, gender, IQ, state of residence, or other features that limit generalizability. When longitudinal studies became more prominent, samples were necessarily drawn from survivors, and information from the past had to be collected retrospectively. Fourth, selection into the military has not been a random process. Until recently, the military has consisted of both volunteers and conscripts, and both groups had to pass physical and mental exams to determine eligibility. In addition, some men deemed fit to serve were nevertheless granted exemptions on other criteria. These difficulties are reflected in the set of existing findings. Subsequently, we review the relevance of these issues for this study.

Military Service and Selection

The Selective Service Act of 1940 initiated the first prewar draft with fewer than 20,000 inductees that same year; however, the next year the number of inductees increased more than 50-fold, to near 1 million. During 1942 and 1943 (when voluntary enlistment was suspended and the minimum age of entry dropped to 18) more than 6 million inductees were added to the rosters. By 1947 (when the draft ended), 50 million aged 18–45 were registered; 36 million were classified, and more than 10 million men were inducted ( Selective Service Administration, 2015 ).

The first group of conscripts drew from the 1914–1919 birth cohorts, who were aged 21–26 in 1940. What was initially a 1-year service commitment was extended by a year in August 1941; after Pearl Harbor in December of 1941, a goal of 200,000 men per month was set, and all inductees were in for the duration of the war plus 6 months ( Selective Service Administration, 2015 ). More than 16 million (out of a total population of 131 million) served in the military; the number of U.S. deaths exceeded 400,000, and more than 670,000 were wounded ( Selective Service Administration, 2015 ).

Military Service and Health

The Effect of Health on Military Service

Attempts to determine whether veteran status is associated with a health advantage or a health disadvantage are complicated by the processes that select men into the military. The World War I draft ended in 1918; until 1940, the military relied on volunteers to maintain a standing force of 330,000. Thereafter, the military relied on two volunteers for every three draftees ( Selective Service Administration, 2015 ). All inductees had to be classified as 1-A, and because of the urgent need, exemptions were limited. Even so, exemptions were granted to conscientious objectors, farmers and those in war industries, and for unusual hardship; those in college could defer only until the end of the academic year. The primary sorting mechanism was the examination for physical and mental fitness. Among white men, about one-third of all those examined were rejected, with about twice as many rejected due to physical issues as for other reasons. Based on figures for 1943, 22% were rejected because of poor eyesight, ear, nose, and throat problems; 28% for cardiovascular and musculoskeletal problems; and 15% due to hernia, feet, or gastrointestinal problems (Statistical Review: WWII, United States Army Service Forces, 1946 ). Mental fitness was evaluated with the Army General Classification Test. Those below a threshold score also were exempted from service.

The Effect of Military Service on Health

Although relatively healthy men were being selected into the military, the experience of military service was challenging. From 1941–1945, men in the armed forces served for an average of 33 months, with almost three-quarters going overseas for an average of 16 months. Noncombat jobs in administration, support, or manual labor were assigned to nearly 40% of enlisted personnel. Of every 1,000 men in combat positions, 8.6 were killed in action, 3 died from other causes, and 17.7 received nonfatal combat wounds ( Selective Service Administration, 2015 ).

Most injuries were inflicted by artillery and bombs. The deployment of field medics made quick treatment more likely and reduced the rate of infection; evacuation of the wounded to field hospitals where they could be moved into surgery within hours increased overall survival rates to 50% ( Covey, 2002 ). About 70% of the war wounds were musculoskeletal, with traumatic amputations occurring in the most serious cases ( Covey, 2002 ). More soldiers died from disease than injury. Men stationed in the South Pacific were particularly vulnerable to malaria and yellow fever ( Joy, 1999 ).

Arguments that favor negative health consequences draw on medical research showing that combat (and the persistent stress and hardship associated with combat) can lead to “vital exhaustion” ( Appels & Mulder, 1988 ), which can lead to a higher risk of cardiovascular disease; a weakening of the immune system ( Adams, 1994 ; Lipton & Shaffer, 1986 ) that can lower disease resistance; the development of arterial lesions, which may increase the risk of cancer ( Adams, 1994 ); and war neuroses and posttraumatic stress disorder, which have also been linked to cardiovascular disease ( Lynch & Smith, 2005 ). Delayed effects of war injuries or illnesses could also be deleterious. Finally, military service has been linked to unhealthy habits such as smoking and drinking, both of which can have long term consequences. During the war, tobacco companies distributed free cigarettes to members of the U.S. military and encouraged people to send cigarettes to show their support. After successful lobbying efforts, cigarettes were routinely included in soldiers’ rations ( Brandt, 2007 ; Goodman, 2005 ; Smith & Malone, 2009 ).

Counter arguments claim that military selection on physical and mental health should persist after discharge and translate into veterans who are healthier than their counterparts ( Seltzer & Jablon, 1974 ). The development of health-promoting behaviors such as fitness, hygiene, and exercise ( LaVerda, Vessey, & Waters, 2006 ); the promotion of self-discipline; the development of skills in leadership and team work ( Gade et al., 1991 ); and increasing personal health responsibility (Hibbard et al., 2007) can all have a positive influence on later life health. Finally, to the extent that military experience enhances human capital, allowing veterans to step into a career path with higher rewards, the positive impact of more education and better jobs could also produce better health. Due to the GI (more formally known as the Servicemen’s Readjustment Act of 1944) Bill, which provided tuition and stipend support, the possibility of changing career paths and gaining additional education was available to virtually all WWII veterans. Although college attendance was on the rise before the war, WWII military service and the GI Bill fueled substantial gains in college attainment for veterans ( Bound & Turner, 2002 ).

Research Studies

Relatively recent interest in potential longer term health vulnerabilities has increased the number of studies investigating the consequences of military service. Investigations of the relationship between military service and subsequent health outcomes contrasts veterans with nonveterans; in a few cases, distinctions among veterans with regard to the timing of their service and the characteristics of their military experiences are also possible. Different data sources, units of analysis, research designs, control variables, birth cohorts, and eras of service, however, make comparisons of results across studies more difficult.

Most studies report higher rates of illness and mortality and steeper health declines for veterans compared to nonveterans, regardless of the era of service. However, most of these studies do not rely on national samples to focus on WWII veterans with individual data. Instead they draw from the experiences of different subsets of veterans to demonstrate these effects. In part because of data availability, veterans from more recent conflicts have been studied more than those from earlier wars. For example, veterans of the all-volunteer armed forces report worse health in mid-life than their counterparts ( Teachman, 2007 , 2011 ). Vietnam veterans have higher than expected rates of deaths from accidents and suicides ( Dobkin & Shabani, 2009 ; Hearst, Newman, & Hully, 1986 ). Veterans from Korea and the later cohorts (1920–1929) who served in WWII seem to have higher rates of lung cancer than nonveterans ( Bedard & Deschenes, 2006 ). Analysis based on Health and Retirement Study (HRS) data indicates that veterans from WWII, Korea, and Vietnam (birth cohorts 1890–1953) are as healthy as their nonveteran counterparts at age 65, but experience steeper declines as they move into older age. Comparisons among veterans of different wars also suggest that WWII veterans have better age-adjusted health outcomes than veterans from Korea or Vietnam ( Wilmoth et al., 2010 ).

Empirical studies that focus on WWII veterans, in particular, are few and report conflicting results. An epidemiological study that used mortality data collected for a large sample of army veterans compared the number of observed deaths for veterans to expected deaths based on age-specific national death rates ( Seltzer & Jablon, 1974 ). Mortality rates for returning veterans were substantially lower in the first 5 years after discharge. By the end of the observation period (1947–1969), however, the mortality gap between veterans and nonveterans had narrowed from 57% to less than 14% ( Seltzer & Jablon, 1974 ).

Studies based on the Stanford–Tremin data provide the most nuanced view of differences among veterans, but at the cost of a relatively small and selective sample of 856 white men born 1900–1920, who were college educated with high childhood IQs. Using interviews from 1945 to 1950, veterans could be sorted into those who were sent overseas, those who saw combat, and those deployed to the Pacific Theater ( Elder et al., 2009 ). Although the age at induction made no difference in mortality rates among veterans, those in combat—especially those in the South Pacific—had higher rates of early mortality ( Elder et al., 2009 ).

Design and Method

Data

The older men cohort of the NLS was interviewed from 1966 to 1990. This national sample of 5,020 older men aged 45–59 in 1966 represents survivors from 1906 to 1921 birth cohorts. The initial wave collects information about their early life and current circumstances, focusing on schooling, work, income, family, and wealth. Subsequent surveys updated information in these areas and included additional modules of questions as these cohorts aged. The mortality information on these respondents was collected in three waves. When possible, interviewers recorded life status and age at death for men who dropped out of the sample. After the 1990 interview, Census obtained certificates for deaths occurring in 1966–1990 from state vital records departments or collected reports from widows or next of kin. As of 1990, 2,693 deaths were recorded, and the mortality observed coincided closely with estimates of mortality from the U.S. Vital Statistics and the Social Security Administration ( Hayward & Gorman, 2004 ).

In 2009, we initiated a second round of linkages through the Demographic Survey Division (DSD) of the United States Census Bureau (USCB) and the National Center on Health Statistics (NCHS). We followed the established protocol for the National Longitudinal Mortality Study that has a long history of mortality matching with the National Death Index (NDI). NDI uses an algorithm to match respondent information with death certificate data using five criteria to evaluate matches. Results of the matching algorithm are classified into 5 classes: (1) exact match on SSN, first name, middle initial, last name, sex, state of birth, birth month, and birth year; (2) SSN matches on at least seven digits (one or more of the items from Class 1 may not match); (3) SSN unknown; eight or more of first name, middle initial, last name, birth day, birth month, birth year, sex, race, marital status, or state of birth match; (4) same as Class 3 but less than eight items match; (5) no NDI match.). Of the 5,020 men, 4,814 provided valid Social Security numbers; remaining cases had to be matched using the other criteria. Through this process, we added 2,109 deaths, for a total of 4,802 deaths through 2008. We then searched for remaining cases in the Social Security Death Index and assigned 346 additional deaths. By 2012, less than 5% of the initial sample had survived, and average age of survivors was 94.5.

Birth Cohorts

Members of our sample reached age 25 during 1932–1946 time span, and all had passed their 18th birthday when the United States entered the war in 1941. Although all were age eligible for service during the war, the initial draft targeted the 1914–1919 cohorts, and those born in 1920–1921 were eligible once the minimum age was reduced to 18. For this reason, we estimate our models for the full sample, but also narrow our scope to those born in or after 1914 to address the cohorts with the highest rates of military participation in our sample.

The statistics in Table 1 allow us to connect enlistment percentages for white men to our sample. By 1940, our cohorts spanned the age range of early adulthood, from 19 to 34. Some were recent high school graduates, while others were establishing their careers and beginning their families. By 1942, when the draft cohorts tripled in size, all were in the target age range. Therefore, our sample includes the early waves of inductees, but not the 1922–1929 cohorts. The percentage of veterans in our sample closely matches those reported from the Census (1984).

Variables

Military Service

In 1967, respondents were asked whether they had ever served in the U.S. Armed Forces (1,962 served; 2,741 did not). Those who served also were asked when they served (1,822 WWII; 33 WWII and Korea; 27 Korea only; 90 during peacetime). Restricting our sample to white men who served in WWII yields a sample of 3,453 men (1,496 veterans and 1,760 nonveterans) (During WWII, many African Americans also served in the military; however, they receiveddifferent treatment during and after their service. They also served in segregated units, since integration of the military did not occur until 1948.). This additional information allowed us to calculate the age their service began, age at discharge, and duration of service. After conducting sensitivity analyses for different functional specifications, we classified age at discharge into four categories: 19–23, 24–28, 29–31, 32, and older. We also include a dummy variable to denote veteran status and code nonveterans 0 on these variables. Information on age at entry, duration of service, and age at exit is linearly dependent. As the terms of enlistment were routinely until the end of the war plus 6 months, our primary choice was between age of entry and age of exit. Bayesian fit statistics favored age at exit.

Controls for Exemptions

We have no direct information on whether respondents were drafted or volunteered, nor do we know which respondents may have been given exemptions. Draft boards called men up on the basis of birthdays, with men born in January called first and men born in December called last if the quota had not been met ( Selective Service Administration, 2015 ). As exemptions were based on certain types of work, physical health, and mental acuity, we include three variables linked to these deferment criteria. First, we include a dummy variable for whether respondent’s father was in farming, because working on a family farm might have exempted respondents from service. To proxy those who may not have been deemed mentally fit, we include two dummy variables for less than a high school education: 0–7 and 8–11 years of schooling. Last, in 1966 respondents were asked whether they had a health limitation that prevented work or limited the kind or amount of work they could perform. Those who responded in the affirmative were asked when the limitation began. Based on this information, we include a dummy variable to indicate men who reportedly having a work-limiting health condition that began before 1940. While this variable likely identifies those with more serious (and activity limiting) health conditions, it does not capture all those who may have received medical deferments.

Health Behaviors

We control for two important health behaviors that are related to mortality, smoking and BMI. Studies have reported that the provision of cigarettes for those in military service increased rates of smoking and may have had delayed negative health consequences ( Bedard & Deschenes, 2006 ). Therefore, we control for any history of smoking. The variable “smoker” combines responses to the question “Did you ever smoke cigarettes?” asked of surviving respondents in 1990 and “Did he ever smoke cigarettes” asked of surviving relatives in 1990. “Smoker” was coded “1” for respondents who ever smoked cigarettes and “0” otherwise. BMI is calculated from self-reported height and weight in 1973 (respondents are aged 52–67). We then use Center for Disease Control and Prevention guidelines for constructing BMI classifications as “Normal” (BMI < 25), “Overweight” (BMI = 25–29.9), and “Obese” (BMI ≥ 30).

SES and Demographics

We use respondent’s longest job to measure occupation, which serves as one indicator of SES. We classified occupation into three categories, white collar or skilled, farming, and a combination of semiskilled, unskilled, and service professions. We also include a third indicator of education for respondents who attended college. We include it here with the SES measures because it was not a basis of exemption, but as an indicator of higher SES, it has been shown to correlate with mortality (Montez et al., 2010). In addition, because the GI bill increased access to high education for veterans, it could also be a positive outcome linked to veteran status. As a third measure of SES, we include the log of total assets accumulated by age 50. Finally, we include age in 1966 as a continuous measure to capture differences in the mortality hazard by age, and add indicators of marital status at first interview, classifying respondents as married, never married, or previously married but not currently married.

Data on basic demographic and SES measures were missing in less than 1% of the cases. In contrast, because health behaviors were addressed toward the end of the survey, approximately 20% of cases were missing BMI and smoking data. Data specific to veterans were missing most often for the duration of military service (about 15%). We use multivariate normal regression in Stata’s mi command to impute missing data for all independent variables using all covariates in the fully specified model.

Approach

We estimate Cox proportional hazards models (Cox, 1972) using Stata 13, which take the following form:

where h ( t | x_j ) is the hazard rate (or force of mortality) at age t for a man with x_j characteristics. This model assumes that all hazard rates are proportional to a baseline hazard h₀ ( t ), which describes variation by age in the mortality transition rate for an average person in this sample. We estimate four models to test our hypotheses. We begin with a model that includes veteran status, controlling for age at interview. The second model differentiates among veterans by their age at discharge; these variables are analogous to interaction terms and estimate differences in mortality risk by age at military exit among the subset of veterans. The third model adds controls for selection criteria—if respondent’s father was a farmer (agriculture exemption); less than a high school education, because AGCT scores correlate .73 with years of schooling ( Personnel Research Section, AGO, 1945 ), and a pre-1940 work limiting health condition. The fourth and final model adds measures of later life SES, BMI, and smoking behavior. Given the late collection of some of our key variables and other sources of missing data, we use multiple imputation using Stata’s mi commands to utilize all cases ( Rubin, 1987 ; Royston, 2009 ). The estimates presented are based on results from 20 imputed data sets.

Results

We begin with an overview of the full sample, reported on the left side of Table 2 , and the subsample of those born from 1914 to 1921, reported on the right side of Table 2 . Both the full sample and subsample are further divided into those who served in the armed forces and those who did not. Of the 3,453 men, we found no recorded deaths for 171 through March 2012, for an overall mortality rate of 95%; 79% of our sample survived to age 65. Of the deceased, the average age at death was 74.6. The mortality rate for veterans is 4% less than that of civilians, but the age at death is not significantly different for the two groups. These mortality rates are not adjusted for age, and since the age at first interview for veterans is younger, on average, than nonveterans, this difference may reflect the different age distributions.

Of those who served, almost half were discharged when aged 24–28; very few got out at younger ages, but more than one-quarter were not discharged until age 32 or older. Nonveterans were more likely to have grown up on a farm, were more likely to have had a career in farming, and were more likely to have left school before the 8th grade. In contrast, more than one-quarter of veterans had some postsecondary education, and two-thirds worked in white-collar or skilled craftsman positions. Veterans also were significantly more likely to have been smokers, but the two groups had very similar distributions on BMI and marital status, with 9 of 10 respondents spending most of their lives as husbands. The only difference in distributions between the full sample and those born from 1914 to 1921 involves veteran status itself. In the full sample, 43% had been in the military; for the smaller set of cohorts, 60% were veterans. The age at discharge also reflects the exclusion of the older cohorts, with the proportion leaving the military in their mid-30s much smaller.

We organize our discussion of the hazard models into two sections. First we estimate mortality models for any deaths occurring after first interview for our full range of birth cohorts. Then we narrow our focus to the 1914–1921 birth cohorts, whose members were more heavily recruited and more heavily drafted for military service (as noted above). For the 1914–21 birth cohort we estimate models only on deaths that occur before age 65 to see whether we find evidence that any difference between veterans and nonveterans weakens as survival time is extended.

Later-Life Mortality for the 1906–1921 Birth Cohorts

Results for these models are reported in Table 3 . In comparing estimates from models 1 and 2, we see no global difference in mortality rates relative to veteran status. Once we disaggregate veterans relative to their age when discharged, however, it appears that veterans who exited at the youngest ages have higher mortality rates than those discharged at older ages, but not significantly different from nonveterans (evidenced by the nonsignificant coefficient for veteran status). In contrast, veterans who exited at older ages have lower mortality rates, particularly those who age out in their mid-20s and those who age out when aged 32 or older.

In models 3 and 4, we see that these differences remain largely consistent when we introduce correlates of selection into the military, BMI, smoking, and SES. Respondents who were raised on farms have lower mortality rates, while those who reported having health limitations that began before 1940 or left school without a high school diploma have higher mortality rates. Being a smoker increases the mortality rate more than obesity (but does not differ in its effect for veterans vs. nonveterans); working in lower blue collar jobs or being previously (but not currently) married also increased mortality rates ( Table 4 ).

Survival to Age 65 for the 1914–1921 Birth Cohort

When we focus on this subset of cohorts and deaths before age 65, we do see a mortality advantage of almost 30% for veterans (model 1). As before, among veterans mortality rates are highest for those who mustered out in their late-teens or early 20s. In limiting our analysis to these cohorts, we also changed the composition in these age-at-discharge categories. In this subgroup, a smaller proportion was discharged at age 32 or later (12% vs. 28% in the full sample), and almost three-in-five were discharged when aged 29–31. This latter age group has the lowest mortality rate—half the rate of nonveterans—an advantage that persists as additional variables are included in models 3 and 4.

As before, sons of farmers have lower mortality rates as do those with more assets and postsecondary schooling. Smokers, those with health problems earlier in adulthood and less than a high school education have higher mortality rates than their counterparts. In results not reported here, we also looked at all deaths for these birth cohorts, but these findings indicated no difference between veterans and nonveterans ( B = .913, p = .079), no advantage for those with 13 or more years of schooling, and a disadvantage for the obese.

Discussion

No one disputes the horrors of war, nor do we question that those who serve in the military make a considerable sacrifice. Much of the research on those who have been in combat has documented the immediate and longer term consequences of that experience. Although the nature of warfare has changed, the physical and psychological trauma associated with its practice has not. This was our mindset as we began our study of WWII veterans.

In contrast to other studies, we use a national sample of WWII veterans, men who not only survived the war, but were still alive in 1966 when the NLS sample was drawn. When considering the full range of birth cohorts, we found no difference in mortality rates by veteran status, but among veterans, those who were discharged in their 30s had higher survival rates. These findings changed when we focused on the 1914–1921 birth cohorts and looked only at deaths before age 65. Here, we did find a mortality advantage for veterans, and it was located among those who left the military around age 30, which included more than half of the veterans in these cohorts. These results fit the selection hypothesis—that positive health selection would provide a mortality advantage that would weaken over time. However, earlier work comparing mortality rates by veteran status controlled only for age ( Seltzer & Jablon, 1974 ). That the mortality gap persisted when we controlled for a range of other factors suggests that something other than health selection is involved. Other possible explanations should also be noted. For example, to the extent that age at exit is associated with whether or not veterans were in combat situations or whether or not they were officers, age at exit may serve as a proxy for very different types of exposure. Although in general, officers tend to be somewhat older, during wartime that correlation weakens, and promotions can elevate the very young and the more mature soldier.

Study limitations must also be kept in mind. Our sample captures many of the cohorts age eligible for service, but we are missing the 1922–1926 cohorts, who could have been called up before the end of the war in 1945. Therefore, our findings do not reflect the full population of WWII veterans surviving to 1966, but only those born before 1922. Although HRS includes all WWII cohorts, that sample was drawn 25 years after the NLS for an initial interview in 1992; therefore, the HRS cohorts of WWII veterans had been subject to higher mortality selection. Second, we are also unable to distinguish among veterans in terms of their military experiences. We do not know whether they were stationed overseas; whether an overseas assignment placed them in Europe or the South Pacific; whether they were in combat roles; or whether they were wounded. Certainly this last point could be relevant if those who exited at younger ages received medical discharges. Being able to incorporate more detailed information about their experiences in the military would add important layers to the story. We are continuing to investigate how their lives unfolded once they returned home to see if we can better identify and explain how the war changed the lives of those who served.

Finally, we have been careful to frame our questions in terms of military service during WWII. Although certain commonalities of service extend across time and space, other important dimensions of military service, in general, and the war experience, in particular, depend on which assignment, which branch of the service, which war, or which era is under consideration. From that standpoint, an answer to the question—What are the health effects of military service?—may differ by context. What should not differ is the extent to which we provide a strong system of care and support when we bring our men and women home.

Acknowledgements

We are grateful for the helpful comments from Steven Haas, Suzanne Meeks, and the anonymous reviewers.

References

Author notes