The effect of military deployment on mental health

Economics and Human Biology 23 (2016) 193–208

Contents lists available at ScienceDirect

Economics and Human Biology journal homepage: www.elsevier.com/locate/ehb

The effect of military deployment on mental health Stéphanie Vincent Lyk-Jensena , Cecilie Dohlmann Weatherallb,c,* , Peter Winning Jepsend a

The Danish National Centre for Social Research (SFI), Herluf Trolles Gade 11, DK-1052 Copenhagen K, Denmark Kraks Fond Institute for Urban Economic Research, Frederiksholms Kanal 30, DK-1220 Copenhagen K, Denmark1 c The Danish National Centre for Social Research (SFI), Herluf Trolles Gade 11, DK-1052 Copenhagen K, Denmark2 d Psykiatrisk Center København (Rigshospitalet), Distriktspsykiatrisk Center Indre By-Østerbro, Strandboulevarden 96, DK-2100 Kbh. Ø, Denmark b

A R T I C L E I N F O

Article history: Received 27 December 2015 Received in revised form 20 September 2016 Accepted 20 September 2016 Available online 30 September 2016 JEL classification: I1 H56 Keywords: Mental health Soldiers Deployment Administrative records

A B S T R A C T

Public concern about soldiers’ mental health has increased over the last decade. Yet the large literature on the mental health problems of returning soldiers relies primarily on self-reported measures that may suffer from non-response bias, usually refers to older conflicts, and focuses mainly on specific diagnoses such as PTSD. Another challenge is that the differences between soldiers and non-soldiers are not necessarily causal, instead possibly reflecting an underlying propensity towards active military service. Using the objective measures of hospitalizations and the purchase of mental health medication, this paper is the first to investigate the effect of recent military deployments on a broader measure of mental health, for a full population of Danish soldiers and a comparison group of eligible men. We exploit a panel of Danish health administrative records and use propensity score matching process, using unique observables such as ability tests. Matching is helpful in both reducing the selection bias from observables and determining pre- and post-deployment period for the comparison group. Then, we estimate the effect of deployment in a difference-in-differences setting, controlling for time trends and other omitted variables affecting both groups. Overall, we find a significant and long-lasting adverse effect of military deployment on soldiers’ mental health, regardless of the comparison groups and underlying assumptions. ã 2016 Elsevier B.V. All rights reserved.

Over the last 20 years, with involvement in the Gulf War, the Balkans, and the Global War on Terrorism, the number of veterans has dramatically increased in many countries, all of which are now questioning the costs and benefits of these wars. On the one hand, military deployment may offer young people the opportunity of job training, education, and the building of new skills, e.g., learning discipline and teamwork (Berger and Hirsch, 1983; Mangum and Ball, 1989). On the other, deployment experiences – in addition to

the risk of injury or death – may also affect soldiers’ later mental health by producing psychological and behavioral dysfunction. This paper investigates the effect of military deployment1 on the use of mental health services in the form of hospitalization and the purchase of mental health medication. To account for the expected selection bias, we exploit a rich dataset for a full population of soldiers with unique observables and contribute to a better understanding of the link between mental health problems and participation in deployment for more recent wars, such as those in Iraq and Afghanistan. With a substantial increase in the

* Corresponding author at: Kraks Fond Institute for Urban Economic Research, Frederiksholms Kanal 30, DK-1220 Copenhagen K, Denmark. E-mail addresses: svj@sfi.dk (S.V. Lyk-Jensen), [email protected] (C.D. Weatherall), [email protected] (P.W. Jepsen). 1 www.kraksfondbyforskning 2 www.sfi.dk

1 The focus in this paper is on deployment, not basic military service, because – while basic military service in Denmark is not expected to be challenging – situations during deployment may affect soldier’s mental health. Moreover, using random assignment in the Danish draft lottery, Bingley et al. (2014) find a negative average effect of peacetime military service on earnings but no effect on mental health. Bingley et al. (2014) also measure mental health with administrative records for psychiatric diagnosis and purchase of medicine.

1. Introduction

http://dx.doi.org/10.1016/j.ehb.2016.09.005 1570-677X/ã 2016 Elsevier B.V. All rights reserved.

194

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

number of soldiers deployed overseas since the late 1990s, Denmark provides a sufficiently large sample for investigating the impact of deployment on soldiers’ mental health. With a population of only 5.5 million, Denmark has sent 27,000 soldiers on more than 50 international military missions, with over 50,000 individual deployments over the last 20 years. To measure the mental health consequences of military deployment, we exploit the availability of health administrative records that in Denmark can be linked to each individual’s personal registration number. Investigating the relationship between deployment and mental health is complicated by positive selection into the military, whereby selected individuals have to satisfy both physical and mental criteria. Furthermore, an increase in manpower or economic changes may also influence the selection criteria or the probability of volunteering. Indeed, important differences may exist among individuals who are deployed, those who are not, and those who did not join the military. Differences between soldiers and non-soldiers may reflect the underlying propensity (i.e., selection differences), not necessarily the causal influence. Therefore, we expect selection to play an important role in understanding the influence of deployment on mental health. To minimize this selection bias when identifying a comparison group for the deployed soldiers, we use data for eligible men2 from the Danish draft board. Eligible men constitute an excellent comparison group, as both soldiers and eligible men have undergone the same selection system (i.e., eligibility screens at the Armed Forces Day, the AFD). Because of both the tests and requirements, the sample of eligible men is positively selected, meaning that our inferences are conditional on their being eligible. To reduce the difference in observables between the deployed and the non-deployed, we use propensity score matching process to identify a matched sample of men from different sub-samples of eligible men, applying a rich set of characteristics. As our set of characteristics includes the Armed Forces Qualification Test (AFQT) scores and the mental health measures before the AFD, both of which are important predictors for mental health (Neal and Johnson, 1996; Macklin et al., 1998), it contributes to identifying the effect of deployment on mental health. Moreover, the matching process allows us to impute pre- and post-periods for the comparison group. We can therefore exploit the panel structure of the data in a difference-in-differences (DID) setting – controlling for time trends and other omitted variables that affect both the soldiers and the different comparison groups of eligible men – as a way of further handling the problem of selection bias resulting from the voluntary nature of deployment. We find a significant adverse effect of deployment on soldiers’ mental health, an effect occurring a few years after the soldiers’ last deployment. The effect is about a one-percentage-point increase in mental health problems: an increase of 24% compared to the sample mean. This result holds for different sample of soldiers and different comparison groups of eligible men and health measures. This paper is structured as follows. First, we present the previous literature. Second, we describe the institutional backgrounds and the data. Third, we present the empirical strategy and the results. Fourth, we show the robustness of the results and conclude.

“post-traumatic growth” (Fontana and Rosenheck, 1998; Dohrenwend et al., 2004; Maguen et al., 2006; Forstmeier et al., 2009)3 and reduce the likelihood of developing depressive symptoms (Whyman et al., 20114 ; Wells et al., 2010). Furthermore, Larson et al. (2008) report that, in the U.S. Marines, psychiatric disorders are diagnosed most frequently during the initial months of training. This process causes a disproportionate loss of psychologically unfit personnel early in training, creating what is known as the “healthy warrior effect.” That men entering the military are not a random subset of the male population complicates any determination of the relationship between military deployment and mental health. Early studies examining the health effects of active military service compared health outcomes of those in combat with those of civilians (Jordan et al., 1991; Price et al., 2004; Card, 1987). The problem with these studies is that the average individual and family background characteristics of active duty servicemen differ greatly from those of civilians (Dobkin and Shabani, 2009). Furthermore, active military service is not random in relation to the men’s education or socio-economic status – variables known to be linked to mental health. While high socio-economic status is negatively related to the probability of joining the armed forces (Segal et al., 1998; Bachman et al., 2000; Kleykamp, 2006), it is positively related to mental health (Miech et al., 1999), and lower educational achievements and low IQ scores are considered risk factors for PTSD (Macklin et al., 1998). Studies comparing deployed and non-deployed soldiers – e.g. Smith et al. (2008), Shen et al. (2010) and Wells et al. (2010) – may suffer from a lack of information about why soldiers are not deployed or why they are deployed to different areas of conflict. To handle this selection bias problem, some researchers have tried to find instrumental variables that could influence an individual's decision to join the military without influencing the outcomes (Angrist, 1990, 1998; Hearst et al., 1986; Angrist and Krueger, 1994; Angrist et al., 2011, 2010; Conley and Heerwig, 2009). However, despite the use of the same instrument, results vary with the choice of the cohorts or the sub-sample. Others have used exogenous variation in the location of overseas deployment assignment (Cesur et al., 2013). However, as a selection always occurs for sending troops on different missions, comparing units of soldiers staying at home or being sent on other missions does not necessarily provide a counterfactual for the deployed units if these units are not identical or perfectly comparable. Moreover, measuring mental health is difficult. A number of studies (Whyman et al., 2011; Cesur et al., 2013; Gade and Wenger, 2011) rely on survey data that may suffer from non-response bias.5 Studies may underestimate mental health problems if soldiers do not self-report symptoms because they fear for their career (Hoge et al., 2004). Using self-reported survey data, Cesur et al. (2013) find a large effect for PTSD for combat zone deployment. Larson et al. (2008) also find that all psychiatric conditions except PTSD occur at a lower rate in combat-deployed personnel than in those not similarly deployed. Thus the PTSD diagnosis is likely to be lower for soldiers not deployed to combat zones (i.e., soldiers not exposed to such situations of extreme danger). Although Autor et al. (2011) use administrative records, their indirect measures of health among veterans make

2. Previous literature A large body of research finds that deployment to war zones and combat in particular may have negative effects on the mental health of soldiers (Keane et al., 1989; Dohrenwend et al., 2006; Schnurr et al., 2002). By contrast, deployment can also provide

2

We use men because only men have to attend the Armed Forces Day meeting.

3 See Park et al. (1996), Tedeschi and Calhoun (1996) and Linley and Joseph (2004) for reviews of this concept. 4 Whyman et al. (2011) suggest several mechanisms, such as the development of “buddy” relationships between enlistees and a structured environment with little crime and drug use, through which active duty military service may reduce the likelihood of a soldier’s developing depressive symptoms. 5 Indeed, combining survey data with administrative records Lyk-Jensen et al. (2012) show that soldiers who do not answer surveys are usually overrepresented among those soldiers recorded in health registers as having a mental disease.

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

disentangling causes difficult.6 Furthermore, because of the time lag in the reporting of adjustment problems (Gade and Wenger, 2011), using longitudinal data is important. Many studies focus on specific diagnoses like PTSD (Schnurr et al., 2002; Cesur et al., 2013), a mental disorder that can develop after a person is exposed to a traumatic event – such as warfare, traffic accidents, or other threats to the person’s life. As psychiatric disorders are usually rare and under-reported, with high diagnostic uncertainty, using information on any subsequent psychiatric illness rather than limiting the analysis to fewer diagnoses or only PTSD would give a broader picture of mental illness among previously deployed soldiers. Evidence of the effect of military deployment in studies using broader definitions of mental health is more scarce and less significant. For example, Cesur et al. (2013) do not find significant effects on depression or suicidal thoughts, likely due to either their small sample or further selection issues. Using a broader definition of mental health, Angrist et al. (2010) find that the Vietnam-era conscription induced a small overall increase in self-reported disability rates among Whites. However, the estimated effects are small and only marginally significant. Likewise, Autor et al. (2011), studying the enrollment of Vietnamera Army veterans in the Disability Compensation program, find it difficult to disentangle the effect adverse health consequences from policies that have expanded benefit eligibility. This literature review emphasizes the challenges involved in measuring the impact of military deployment on soldiers’ mental health. Our contributions to the literature are as follows: First, we extend previous studies by investigating the effect of deployment on mental health outcomes for a broader definition of mental health, making use of very high quality data. Second, our data also allows better estimation of the effect of deployment on mental health, as we can both control for many relevant observables such as AFQT scores and use different control groups in a DID-approach to both minimize the selection problem and assess the external validity of our results. Third, in contrast to many previous studies, our results come from military deployments to more recent wars, such as those in the Balkans, Iraq, and Afghanistan. 3. The health and military systems 3.1. The Danish military system Unlike the U.S. and other European countries, Denmark does not have a professional military. Instead, attendance at the Armed Forces Day (AFD) meeting is mandatory for all men when they turn 18.7 – and, since 2004, women have also been invited to participate. Before the mandatory AFD meeting, all prospective draftees fill out a health questionnaire that forms the basis for a health assessment. The military physician can, if necessary, seek additional medical information from public health records prior to the AFD. On the AFD, prospective draftees undergo a medical examination and an

6 For example, Autor et al. (2011) find evidence of a recent increase in disability uptake in the U.S. but cannot distinguish whether a long-term adverse health effect of combat or a recent liberalization of the benefit rules drives this effect. 7 As postponing the AFD examination is possible, not all men are necessarily age 18 when they take it. We therefore control for age in the propensity matching.

195

IQ test (Armed Forces Qualification Test, AFQT)8 On average, 60– 70% of a birth cohort is declared eligible.9 Only men declared eligible must participate in the draft lottery, which inducts men into military service. After military service, eligible men who volunteer and obtain a contract with the Danish Armed Forces (DAF) are deployed.10 As DAF capacity is limited, not all men who want to join necessarily obtain a contract. Although men volunteer to sign the contract, they cannot volunteer for a deployment to a specific country. On average, 7000 men perform their military service each year. During 1992–2009, the DAF annually sent an average of 2000– 4000 soldiers on international military deployments.11 These soldiers sign contracts of varying lengths.12 From 1994 to 2005, the Danish International Brigade (DIB) was a peace-keeping force offering a three-year contract for international missions. Most of the soldiers deployed during 1994–2009 were recruited through this DIB contract. In Denmark soldiers can serve short-term contracts (e.g., three years) and then return to civilian life after only one deployment. 3.2. The Danish health care system Although the DAF has its own medical system, “Military Health,” it is not an alternative parallel to the civilian health system (National Health Service, NHS). The role of “Military Health” is mainly to ensure that soldiers are fit for a deployment and to provide medical support during the deployment, whereas access to hospital treatment is possible only through the NHS. Soldiers are treated in the same hospitals as civilians and have access to the same general practitioners (GPs) and specialists.13 In the period under analysis, Danish soldiers did not have special health insurance, nor was the NHS focused on or particularly aware of

8 The test has been used for over 50 years (Teasdale, 2009; Teasdale et al., 2011). Teasdale et al. (2011) assess the test-retest reliability and the possible negative effects of either lack of motivation or under-performance among the men taking the test. 9 About 10% of a cohort is declared ineligible for military service before the AFD and do not have to appear. Typical reasons come through documentation of serious somatic and psychiatric disorders. Of the 90% who show up, 30 percent are declared ineligible because of low AFQT scores (10%), high (10%) or low (5%) body mass index, or certain medical conditions (e.g., ADHD, musculoskeletal disorders, asthma, 5%). Bingley et al. (2014), studying the opportunity costs of mandatory military service in Denmark, show a strong association between academic achievement and passing the AFQT, thereby supporting the validity of AFQT-taking. In addition, they show a gradient in AFD attendance by grade point average on 9th grade national tests, suggesting that learning difficulties – not "outsmarting" the draft board – are the main cognitive ground for some men not being required to attend the AFD. 10 Likewise, throughout the Vietnam era, most soldiers were also volunteers (Angrist et al., 2011). 11 Until January 1, 2006, the length of military service averaged eight months. Since then, military service has been called “Hærens Basisudddannelse” (HBU, or basic military education). It constitutes an average of four months of military education that can be followed by a voluntary eight months of further military training, the “Hærens Reaktionsstyrkeuddanelse” (HRU, or Military Reaction Forces education). The HRU prepares conscripts for deployment on international military missions as privates. Officers tend to bypass HBU, going straight to a four-year military training college. 12 The contract length mainly reflects soldiers’ age and ability to fulfill different type of tasks in the military. As we can control for both age and AFQT, the type of contract does not constitute an important source of variation. 13 In contrast to soldiers in many other NATO countries, most Danish soldiers leave the military after one or two deployments. Military Health offers free psychological help from military psychologists or private psychologists with DAF contracts, even after the soldier has left the military system. However, the program was not in great use during our analysis period. Moreover, military psychologists cannot prescribe medicine, nor can they refer soldiers to psychiatric hospitals.

196

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

soldiers’ mental health impairment.14 Soldiers can be formally recorded as having a psychiatric diagnosis either as acute emergency cases or through contacts with hospitals (as either admissions or day patients) via prescriptions from GPs, specialists, or military doctors. Prescription drugs are common and heavily subsidized in Denmark. Moreover, all people living in Denmark have access to a GP, with expenses covered by the NHS. 4. Data and descriptives The dataset comprises administrative records from the DAF. The military register dataset contains longitudinal information on the population of eligible men: 6221 deployed eligible and 104,741 non-deployed eligible men (hereafter “deployed” and “nondeployed” men).15 This dataset includes the AFQT for men born between 1975 and 1982.16 For the deployed men, we also have the number, date, location, and type of mission. The date of the deployment allows us to determine the pre- and post-deployment periods.17 Given the unique Danish civil registration number for each individual, the administrative Danish military records are linked to a huge variety of socio-demographic characteristics (e.g., education, social affiliation, health and medical care, death and cause of death) from Statistics Denmark. From the data, we can also distinguish the one-time deployed from the many-time deployed – data providing information on the number of deployments (Table A1 in the AppendixA).18 The deployed born between 1975 and 1982 were deployed mainly between 1996 and 2009 (Table A2 in the AppendixA), when the main missions from 1996 to 2009 were Balkans missions.19 Fewer were deployed to Afghanistan and Iraq, where operations began in 2002 and 2003, respectively (Fig. A2 in the AppendixA). 4.1. Mental health measures To proxy mental health, we combine the objective measures of psychiatric diagnoses from the Danish Psychiatric Central register,20 and the purchase of mental health-related medication

14 The Danish involvement in international military missions has changed so rapidly that neither the DAF nor the NHS had been aware of the consequences of deployment on soldiers’ mental health. The Danish veterans’ policy was first implemented in October 2010. 15 We include all the non-deployed eligible men regardless of their military service status. Indeed, simple descriptive statistics show that these two groups (served/not served) differ only in their AFQT scores, a variable we use in the matching process. Moreover, Bingley et al. (2014) show that military service has no impact on mental health. They also use psychiatric diagnosis and purchase of medicine as a proxy for mental health. However, as the two groups may still differ on non-observables, we therefore perform sensitivity analyses for these two groups. 16 However, the data are not fully representative for the 1975 cohort. 17 Longitudinal data are important when analyzing changes in health behavior due to major crisis or shocks (Ásgeirsdóttir et al., 2014) 18 As we have no information on combat units, we focus on the effect of deployment. Nonetheless, to investigate heterogeneous effects, we use deployment mission country. 19 The one-time deployed have not necessarily left the DAF; they are simply recorded as having only one deployment during 1996–2009. 20 The Danish Psychiatric Central Register (Munk-Jørgensen and Mortensen, 1997; Hageman et al., 2008) contains no psychiatric diagnoses from GPs or psychiatric specialists who practice outside hospital settings. As we can observe medical prescriptions from GPs and specialists, this data restriction constitutes a minor issue.

(“N-drugs”),21 from the Danish Register of Medicinal Product Statistics (Lægemiddelsdatabase),22 both available from 1995 to 2010.23 For each year that we consider in the analysis, we look at whether the person has received a psychiatric diagnosis24 or has purchased N-drugs.25 We use the purchase of N-drugs as an indirect measure of psychiatric morbidity. This indirect measure is useful because our records on diagnoses capture only patients who have been treated in the hospital-based NHS. The purchase of N-drugs represents a lower bound, as it captures only those purchasing N-drugs who have consulted a doctor. However, as N-drug prescriptions by GPs are common – and largely financially supported in Denmark – we have identified almost 100% of those both having a mental health disease and consulting their GPs. 4.2. Descriptive statistics and full sample Because our sample encompasses cohorts of eligible men (i.e., with similar initial health), we can easily rule out the survivor bias described in Seltzer and Jabon (1974) – that soldiers often live longer than the rest of the population because of their initial better health. The non-deployed eligible men therefore constitute a powerful comparison group. Moreover, our using characteristics prior to the AFD year, together with the AFQT, can further improve the comparability of the two groups. AFQT scores constitute a basic skills measure that predicts job performance and is a good proxy for labor market prospects and predictor for mental health (Neal and Johnson, 1996; Macklin et al., 1998). Many studies show that physical and mental health are correlated. Thus physical and ability tests are important covariates for describing soldiers and their counterparts. Cuesta and Budría (2015) show that impacts on mental health could be different, depending on the subject’s personality traits, and Macklin et al. (1998) show that lower precombat intelligence is a risk factor for post-traumatic stress disorder (PTSD). Nonetheless, to take the heterogeneity into account, we use the physical and ability tests. War-related deaths are not expected to constitute an important source of bias. Since 1992, about 60 Danish soldiers have died, and 300 have been physically injured during military deployment.26 To ensure that possible differences between the two groups do not

21 NS drugs denote drugs acting on the nervous system as listed in the second column of Table A4 in the AppendixA. We pool these pharmaceutical drugs into one group – which we denote “NS drugs” – as there may be substitution and complementarity between the different types of medication for mental disease. 22 The register contains information on sales of drugs prescribed by GPs, specialists, or military physicians. 23 Individuals recorded as having undergone substance abuse treatment are already recorded with either a psychiatric diagnosis or the purchase of mental health medication. 25 The purchase of medication per person refers to the number of prescriptions written by the GP (the primary sector), because hospitals (the secondary sector) record only the diagnosis on the patient's personal identification number (CPR number). Soldiers can also obtain medicine from the Military Health Service. As long as the medication is prescribed, it will appear in the registers. 24 In the Appendix A and B we also examine the type of diagnoses before and after deployment. We use the World Health Organization International Classification of Disease, version 10 (WHO ICD-10 classification system) in the analysis (World Health Organization, 2004). The categories of psychiatric diagnoses that we use appear in Table A4. These diagnoses were selected by a group of psychiatrists specialized in veterans’ care. We include ICD F00-09 because the group of eligible men that we consider is basically "healthier" than the average population. In such a group, registering any subsequent psychiatric illness is important, because psychiatric disorders are usually rare and under-reported, with high diagnostic uncertainty. As we compare the deployed and the non-deployed, there is no risk of over-estimating the effect. Moreover, mild traumatic brain injury, a new military syndrome introduced by the war in Iraq – while labeled “somatic” – is identical with the psychiatric diagnosis of postconcussional syndrome (ICD-10: F07.2). 26 See Fig. A4 in the AppendixA for the number of dead, wounded, and returnees, by mission, in 1996–2009.

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

197

Table 1 Socio-demographic characteristics for the deployed (both one-time and many-time) and the non-deployed born 1975–1982. Percent. Deployed A–B

Born Born Born Born Born Born Born Born

in in in in in in in in

1975 1976 1977 1978 1979 1980 1981 1982

Immigrant Placement in out-of-home care as a child Raised in single-parent family Average no. of AFQT tasks solved out of 78

***

*** ***

***

Number of observations

Non-deployed A

B

C

One

Many

Total

C-D

D

A-D

11 13.6 15.5 14.2 13.6 10.2 9.6 12.3

10.4 13.1 13 13 12.6 14.1 12.7 11.1

10.6 13.3 14 13.5 13 12.5 11.4 11.6

***

8.1 13.2 14 14.1 13.7 13.2 12 11.8

***

1.6 4.2 20.1 45.1

1.6 5.1 20.9 44.4

1.6 4.7 20.6 44.7

*** *** ***

3.4 3.9 17.4 44.5

***

2664

3557

6221

**

*** ***

*** **

104,741

Chi2 test and t-test ***p < 0.01; **p < 0.05; *p < 0.10. A–B-compares one-time (A) and many-time (B) deployed. C–D compares deployed (C) and non-deployed (D). A–D compares one-time deployed (A) and non-deployed (D) Note: We only include men with information on AFQT and AFD year.

result from changes in the sample during the period, we exclude deaths (including war-related deaths and post-service mortality) and persons who immigrated or emigrated during the 16-year analysis period (1995–2010). Given that some soldiers were deployed more than once during 1996–2009, and that they were deployed at different times, we chose the first deployment year for determining the periods before and after deployment. Table 1 compares deployed and nondeployed men.27 It shows that the deployed are on average older, mostly ethnic Danes, more often raised in a single-parent family, and more often experienced placement in out-of-home care as a child (i.e., foster care and residential care homes). When we compare the pre-AFD characteristics of the one-time and the many-time deployed, we find that the one-time deployed have a higher score on the AFQT and belong to older cohorts. The comparison of the one-time deployed with the eligible men shows that the one-time deployed are on average significantly different because they are mostly ethnic Danes, more often raised in a single-parent family, and have a higher AFQT score. These comparisons show the importance of controlling for individual characteristics in the estimation or even use matching to select a better comparison group. We begin our analysis with the one-time deployed soldiers, as their experience (one deployment) is less heterogeneous.28 However, we include the many-time deployed in our sensitivity analysis. 4.3. Matching soldiers and eligible men: estimation sample The data description shows that the deployed and the nondeployed men differ in terms of pre-AFD covariates. To both reduce

27 Because the AFQT as an observable variable – usually not available in other studies – increases the quality of the matching, we exclude 251 one-time and 306 many-time deployed without an AFQT, out of the original sample of 6778 deployed. The 557 excluded individuals usually have basic education, were deployed in the 1990s in the former Yugoslavia, and usually have more psychiatric diagnoses than the soldiers for whom we had an AFQT score. Thus removing them from the sample likely weakens the effect of deployment on mental health. For some eligible men, the AFD year was also missing. We therefore excluded those 49 individuals from the original sample of 104,790. 28 For the different mission experience for both the one-time and the many-time deployed, see Fig. A2 in the AppendixA.

differences in observable pre-characteristics and provide a useful comparison group, we use the propensity score matching process, conditional on eligibility. As the unconfoundedness assumption in matching is based on the hypothesis that all the relevant covariates are controlled for, the choice of the covariates is crucial. A metaanalysis of PTSD-related studies (Brewin et al., 2000) shows that, depending on the population being studied, factors such as gender, race, age, education, previous trauma, previous abuse, and psychiatric history were highly predictive of PTSD symptoms. Furthermore, Kessler et al. (2005) show that the starting age for some psychiatric diagnoses is usually early adulthood. As characteristics one year before the deployment could be considered endogenous, i.e., reflecting the results of entering the military system, we use covariates measured on the AFD – because these are always measured before induction to military service and before the first deployment. We use covariates such as the AFQT score, ethnicity, mental health service use up to the AFD year, variables indicating how the person lived until age 17 (e.g., out-ofhome placement, raised in a single-parent family), and year of birth.29 We thus avoid systematic differences between the deployed and non-deployed men up to the AFD year. In particular, our having the AFQT scores and pre-AFD mental health measures is very important for identifying the effect of deployment on mental health. Furthermore, the deployment varies between 1996 and 2009 for the deployed men, thereby giving us different pre- and post-deployment periods for measuring mental health outcome. Before exploiting the panel structure of the data in DID setting to control for time trends and other omitted variables affecting both the soldiers and the comparison group of eligible men, we impute pre- and post-periods for the comparison group.30 Although matching does not fundamentally differ from a fully saturated OLS model (Angrist, 1998), we also use it to impute the counterfactual year of deployment of these matched non-deployed men by using the deployment year of the comparable deployed. We use the first deployment as the cut-off point for the pre- and post- periods. The pre-deployment period is defined as the years before the first deployment, while the post-deployment period includes the first deployment year. Thus we can infer what would

29 As eligibility status also relies on physical conditions, further matching on these characteristics is unnecessary. 30 Becker and Hvide (2013) use a similar approach.

198

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

Table 2 The matched sample for the one-time deployed and the non-deployed eligible men.

Grew up in single-parent family Placed in out-of-home care as a child Immigrant AFD year Year of birth Average no. of AFQT tasks solved out of 78 Combined measure of psychiatric diagnoses and purchase of NS drugs up to AFD year No. Observations

Deployed

Matched non-deployed

Total

(1)

(2)

(3)

(4)

(5)

(6)

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev.

0.201 0.042 0.016 1998 1978 45.05 0.035 2664

0.401 0.201 0.126 2.575 2.222 7.903 0.184

0.203 0.038 0.016 1998 1978 45.20 0.033 2664

0.402 0.191 0.125 2.484 2.205 7.867 0.18

0.202 0.04 0.016 1978 1978 45.13 0.034 5328

0.401 0.196 0.125 2.53 2.213 7.884 0.182

Differences (7)

n.s. n.s. n.s. n.s. n.s. n.s. n.s.

Column 7 shows whether there are significant differences between deployed and non-deployed. Chi2 test and t-test ***p < 0.01; **p < 0.05; *p < 0.10.

have happened to the mental health of the deployed had they not been deployed. The propensity score used in the matching is the probability of military deployment conditional on pre-AFD characteristics, i.e., it selects eligible men, whose ex ante probability of being chosen for a deployment is closest to that of the 2664 one-time deployed soldiers. To reduce the problem associated with matching on a multidimensional covariate vector (Rosenbaum and Rubin, 1983), we use a probit specification to estimate the propensity score. Given the estimated propensity score matching, we use nearestneighbor matching (without replacement) to combine deployed men and eligible men.31 We run a probit model of being deployed on pre-deployment characteristics (i.e., characteristics measured in the AFD year) and obtain propensity scores for all 2664 soldiers and for the 104,741 eligible men (see Table A4 in the AppendixA). Ex ante, the deployed men represent 2.5% of the sample. To ensure that we compare pairs of deployed and non-deployed eligible men who are the same age in the same calendar year, we impose exact matching on the birth year. Compared to Table 1, Table 2 shows that matching manages to largely remove the pre-AFD differences between the two groups on observable characteristics. Column 7 shows no significant differences in the pre-deployment characteristics between the matched one-time deployed and the non-deployed men. We re-run the propensity score specification on the matched sample: the 2664 one-time deployed and their 2664 matched controls. The median absolute standardized bias (Rosenbaum and Rubin, 1985) drops from 5.6 before matching to 1.1 after matching. Thus the two groups are very similar. 5. Empirical strategy: difference-in-differences (DID) To estimate the effect of deployment on mental health, we use DID models. The key assumption is that the average difference in the outcome is presumed the same for both the eligible men and, counterfactually, the deployed men, had they not participated. In other words, we assume that unmeasured factors, changes in economic conditions, or other policy initiatives similarly affect both the deployed and the eligible men. In the analysis we have multiple observations per individual, as we observe them for several years before and after the deployment (5328 men observed during 16 years, from 1995 through 2010). We run a DID regression explaining our mental health combined measures, i.e., we estimate Eq. (1), where the dependent variable psy denotes a combination of the two objective measures

31 To perform propensity-score matching and covariate balance testing, we use a version of Edwin Leuven and Barbara Sianesi’s Stata module psmatch 2 (210, version 4.0.4) (http://ideas.repec.org/c/boc/bocode/s432001.html).

for mental health (psychiatric diagnosis and purchase of NS drugs) and is equal to 1 if person i is recorded with either a psychiatric diagnosis or medication purchase in year t.32 We interpret the coefficient of interest b2 as the difference between the two groups. p syit ¼ b0 þ b1 deployedi þ b2 af terit  deployedi þ b3 af terit þ g X i þ dt þ e

ð1Þ

deployed is a dummy variable, equal to 1, if i has been deployed, or else 0 after is a dummy variable equal to 1, if the observed period occurs after the deployment year or the imputed deployment year for the non-deployed eligible men. X i is the matrix of variables that entered the original matching procedure, i.e., the characteristics of the men in their AFD year (Table 3). To increase efficiency and adjust for any small residual bias, we control for the original matching variables. We also control for the AFD year (dt ) and, to address the possibility of correlated error terms within an individual over time, we cluster the standard errors at the individual level. As we estimate (1) with a linear probability model (LPM) using ordinary least squares (OLS), we assume linearity for mental health. As the dependent variable is bivariate, we also estimate Eq. (1) using a probit model, and we find that the marginal effects at the mean were not much different. To study heterogeneous effects, we estimate different interactions effects, and, as Breen et al., 2013 (p. 80) point out, the interpretation and comparison of interaction terms in non-linear probability models can be problematic. Therefore, we use OLS models for all our empirical results. We focus on differences after the first deployment year for the deployed and the imputed year for the non-deployed men. Our parameter of interest is b2 (the coefficient of the interaction term of being deployed and after deployment, i.e., deployed*after), which measures the difference between deployed and the nondeployed men. 6. Results We estimate Eq. (1) for different sub-samples. Table 3 summarizes the results. Column 1 shows the overall results, while columns 2–5 investigate heterogeneous effects for some selected covariates, and column 7 decomposes the temporal effect by year. The effect of deployment on mental health (b2 ) is significant, showing an adverse effect on soldiers’ mental health after deployment. The mean effect of deployment on mental health (i.e., the combined mental health measures) is 1 percentage point (column 1). Thus, given the sample mean for mental health (4.2%),

32 We show the results, where psy denotes psychiatric diagnosis only in our sensitivity analysis.

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

199

Table 3 DID estimation results of the effect of deployment on combined mental health measures from 1995 to 2010. Cohorts 1975–1982. Sample: One-time deployed and matched non-deployed eligible men with matched dates “before” and “after.” LMP estimations. Combined mental health measures

Main

Interaction of individual characteristics

Interaction of country

Interaction of time

(1)

(2)

(6)

(7)

(3)

(4)

(5)

Deployed

0.007*** 0.008*** 0.006*** 0.008*** 0.005** 0.007*** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002)

After

0.040*** (0.002)

0.038*** (0.002)

0.039*** (0.002)

0.040*** (0.002)

0.043*** (0.002)

Deployed*after

0.010*** (0.003)

0.011*** (0.003)

0.009*** (0.003)

0.010*** (0.003)

0.008*** (0.003)

0.005** (0.003)

0.042*** (0.002)

(1, 2, 3) years before deployment

0.003 (0.003)

Deployed*(1, 2 3) years before deployment (1, 2, 3) years after deployment

0.001 (0.004) 0.001 (0.003)

Deployed*(1, 2, 3) years after deployment (4, 5, 6) years after deployment

0.003 (0.004) 0.001 (0.003)

Deployed*(4, 5, 6) years after deployment (7, 8, 9) years after deployment

0.008*** (0.004) 0.004* (0.003)

Deployed*(7, 8, 9) years after deployment

0.008*** (0.004) 0.003 (0.007)

Deployed*after * (Grew up in single-parent family) Deployed*after* (In out-of-home care as child) Deployed*after * (AFQT score) Deployed*after* (health measures up to AFD year)

0.023 (0.014) 0.006 (0.015) 0.032** (0.015)

after*deployed in Afghanistan

0.017** (0.007)

after*deployed in Balkans

0.012*** (0.003)

after*deployed in Iraq

0.004 (0.005)

after*deployed in other countries

0.013** (0.006)

Adj. R2 No. of individuals (clusters) No. of observations Mean of the dependent variable

0.02 5328 85,248 0.042

0.02 5328 85,248 0.042

0.02 5328 85,248 0.042

0.02 5328 85,248 0.042

0.03 5328 85,248 0.042

0.03 5328 85248 0.042

0.015 2812 35980 0.047

Standard errors clustered at the individual level in parentheses: * p < 10, ** p < 0.005, *** p < 0.01. Note: The matched sample includes the 2664 deployed and the 2664 matched non-deployed men. Matched date means that within a matched pair the deployment year for non-deployed is imputed from the corresponding matched deployed. The controls include a matrix of variables that entered the original matching procedure. To avoid having a lagged dependent variable on the right-hand side of the regression, we exclude psychiatric diagnosis and mental health-related medication up to the AFD year. results remain unchanged when we include them. We use the following specification: psyit ¼ b0 þ b1 deployedi þ b2 afterit  However, deployedi þ b3 afterit deployedi zi þ b4 afterit þ g W i þ dt þ e, where Wi is the matrix of the characteristics entering the matching procedure minus zi. AFD: Armed Forces Day.

we obtain a 24% increase. Table 3 shows, for all seven specifications, that deployment has a significant adverse effect on soldiers’ mental health, varying between 19 and 26%. Our results include coefficients on deployment (b1 ) and the period after (b3 ). While the results in Table 2 showed no significant differences between the deployed and the matched non-deployed

on their characteristics as measured at the AFD, the coefficient of overall deployment (b1 ) is significantly negative and small in all the seven specifications (about 1% at the sample mean). We choose to match the two groups on their characteristics at the AFD because the AFD always takes place before the draft and the first

200

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

Fig. 1. DID estimated coefficients. The matched sample of eligible men with matched dates and the sample of one-time deployed men born in 1975–1982 recorded yearly with a psychiatric diagnosis or purchase of N-drugs (combined mental health measures) during the period 1995–2010. Note: The estimation coefficient comes from the matched sample of non-deployed eligible men with matched dates and the sample of the 2664 one-time deployed men. We restricted the period to 3 years before to 9 years after deployment/imputed deployment. The controls include a matrix of variables that entered the original matching. To avoid having a lagged dependent variable on the right-hand side of the regression, we exclude psychiatric diagnoses and mental health-related medication up to the AFD year. However, results remain unchanged when we include them.

deployment, and thus these characteristics do not reflect any further selection from the military. Because the period between the AFD and the first deployment is three years on average, the small significant effect of the parameter deployed – capturing some unobserved characteristics— represents the healthy warrior effect, showing that military training selects the mentally strongest for deployment (Larson et al., 2008). However, to avoid further endogenous selection, as explained earlier, we had to use the characteristics from the AFD. The DID setting allows us to control for this level difference between the two groups. The coefficient of overall time effect (b3 ) is positive, meaning that over time the risk of mental health impairment increases with age. To investigate heterogeneous effects, we interact binary indicators of the individual’s characteristics with the deployment and after*deployment dummies (columns 2–5). Previous studies have shown that growing up in an unstable environment increases the risk of becoming mentally ill (Miech et al., 1999). One might therefore assume that deployed men growing up with only one parent or experiencing out-of-home care placement have a higher risk of becoming mentally ill when they experience the extreme situations that may accompany deployment. Although the parameter estimate of growing up in an unstable environment – out-of-home care (not shown) – is positive and significant (i.e., increases the risk of mental health service use), the results in Table 3, columns 2 and 3, show that combining growing up in an unstable environment with deployment does not have a significant effect on the risk of becoming mentally ill after deployment.33

33 Not controlling for AFQT as a covariate in our models makes the heterogeneous effect on an unstable environment significant, indicating that the AFQT score is a very important control variable.

Macklin et al. (1998) find that lower educational achievements and low IQ scores are risk factors for PTSD. In our model, we also find that a lower AFQT score is a risk factor for mental health. Nevertheless, our results on the heterogeneous effect of the interaction of an individual’s AFQT score and deployment are positive but not significant (column 4). Thus the combination of low AFQT score and deployment does not significantly increase the risk of more mental health service use. Nonetheless, deployed soldiers who had used mental health services before age 18 have a higher risk for post-deployment mental illness (column 5). To investigate the heterogeneous effects of missions, we use information about the mission country. We cannot measure intensity of exposure, as we do not have the appropriate data, but instead investigate whether the effects on mental health may differ for different mission countries. Column 6 in Table 3 shows that soldiers sent to the Balkans have 1.2 percentage points higher risk of having a mental illness than the matched non-deployed eligible men. One possible explanation is that soldiers deployed to the Balkans were not prepared to witness such violence while not being allowed to intervene (Orsillo et al., 1998). Moreover, Balkan soldiers received very little post-deployment support or help. We find the highest effect (1.7 percentage points increase) on mental health service use for soldiers deployed to Afghanistan (Table 3, column 6). War conditions in Afghanistan – with more combat situations in the form of ambush and improvised explosive devices – likely explain this result. Given that the effect on mental health may occur some years after the last deployment, we can expect even higher incidences of mental illness for these soldiers in the future. In contrast, we find no significant effect on mental health for the soldiers sent to Iraq, compared to the non-deployed eligible men. One possible explanation is that the degree of exposure was lower for these missions compared to those in the Balkans and Afghanistan, reflected by a lower number of killed and wounded Danish soldiers in Iraq, compared to those killed and

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

201

Table 4 Sensitivity analyses of the results of the effect of deployment on mental health. Cohorts 1975–1982. Point estimate

Mean of the dependent variable

Point estimate as% of mean

No. of observations

No. of individuals

1,775,392

110,962

85,248

5328

1,718,480

107,405

26.2

880,400

55,025

15.6

872,240

54,515

26.3

10,656

5328

40.0

85,248

5328

Dependent variable: Combined mental health measures (psychiatric diagnosis, purchase of N-drugs) Panel A. Full sample of deployed (one-time and many-time deployed) and full sample of eligible men with random dates for deployment Deployed*after 0.004*** 0.047 8.5 (0.002) Panel B. One-time deployed and matched eligible men with random dates for deployment 0.014*** 0.042 Deployed*after (0.003)

33.3

Panel C. One-time deployed and full sample of eligible men regardless of military service status with random dates for deployment Deployed*after 0.009*** 0.047 19.1 (0.002) Panel D. One-time deployed and sample of eligible men excluding men who serve with random dates for deployment Deployed*after 0.011*** 0.042 (0.002) Panel E. One-time deployed and sample of eligible men excluding men who did not serve with random dates for deployment Deployed*after 0.007*** 0.045 (0.002) Panel F. One-time deployed and matched eligible men with matched dates for deployment and two-periods (before and after) Deployed*after 0.047*** 0.179 (0.014) Dependent variable: psychiatric diagnoses only Panel G. One-time deployed and matched eligible men with random dates for deployment and psychiatric diagnosis as an outcome Deployed*after 0.002** 0.005 (0.001)

Standard errors clustered at the individual level in parentheses: * p < 10, ** p < 0.005, *** p < 0.01. Note: The matched sample includes the 2664 one-time deployed and the 2664 matched non-deployed eligible men, while the full sample of eligible men include 104,741 nondeployed and 2664 one-time deployed and 3557 many-time deployed. Matched dates mean that within a matched pair the deployment year for non-deployed is imputed from the corresponding matched deployed. Random dates mean that we randomly allocate the deployment year for non-deployed eligible men. The controls include a matrix of variables that entered the original matching procedure. To avoid having a lagged dependent variable on the right-hand side of the regression, we exclude psychiatric diagnosis up to AFD year. However, results remain unchanged when we include them.

wounded in missions in Afghanistan and the Balkans (Fig. A4 in the AppendixA). To illustrate how the results vary by year, we decompose the effect by group of years pre- and post-deployment. However, the number of observations becomes very small if we consider long pre- and post-deployment periods (Fig. A3 in the AppendixA). We therefore restricted the sample to observations within the time window of 3 years before and 9 years after the deployment (reducing the sample from 85,248 to 35,980 observations). In Table 3, column 7, the interaction term after deployment – now split into 3-year intervals – illustrates in which periods the mental health problems are most likely to occur. As expected, the results in Table 3, column 7, show no significant pre-deployment time trend (the coefficient for 3 years before deployment is not significant) in the matched sample, confirming that the two groups are comparable. As previously mentioned, we have to match on pre-characteristics before entering the military, i.e., at the AFD, which occurs on average 3 years before the first deployment. Therefore, the small but significant coefficient for the deployed reflects the healthy warrior effect for the deployed (Larson et al., 2008). The effect of deployment on soldiers’ mental health is significant after 4 years, compared to the comparison groups. Furthermore, the delayed effect on mental health can also explain why Balkans soldiers deployed at the beginning of the analysis

period have a higher probability of being recorded with a mental health problem than soldiers more recently deployed. Thus deployment clearly increases mental health problems over time. Fig. 1 illustrates the yearly effect of deployment on mental health for the matched sample of eligible men and the sample of the 2664 one-time deployed men, and Table 3, column 7, shows the results by 3-year intervals. As the numbers of observations per year and for the 3-year intervals are different, both significance levels and parameter size in Table 3 and Fig. 1 likewise differ. Nonetheless, Fig. 1 also shows that deployment has a long-lasting effect on mental health, thereby confirming the results in Table 3. 7. Sensitivity analysis We now examine the sensitivity of our results with respect to certain choices we made about measurements and sample inclusion and show that our results are robust to these changes. Our full sample both includes one- and many-times deployed and a large group of non-deployed men. We investigate whether the effects we find for the one-time deployed and their matched eligible men can be generalized to the full sample. While the group of one-time deployed could be a selected group if the DAF prevents soldiers with a mental illness from being deployed more than once, then being deployed only once could also result from an individual choice. Indeed, many Danish young

202

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

men choose to be deployed only once, simply to have a once-in-alifetime experience (Lyk-Jensen and Glad, 2016). As we use the first deployment year to divide the pre-and post-deployment periods, and as most of the one-time deployed were deployed at the beginning of the analysis period, this choice means that we observe the many-time deployed for fewer years than the one-time deployed (Fig. A3 in the AppendixA). Although the many time-deployed could also have more accumulated stress because of their number of deployments, we expect that they do not use mental health services while deployed to avoid being recorded as having a mental illness during their deployment (i.e., what is known as the “incapacitation effect”). Excluding the many-time deployed causes a bias only if this group is ex ante very different from the one-time deployed. However, Table 2 shows that the two groups differ only (a) in their AFQTscores (slightly higher for the one-time deployed) and (b) in age (the one-time deployed are slightly older). Given the empirical findings about the positive correlation of high IQ and mental health (Macklin et al., 1998), excluding the deployed with lower AFQT scores should not increase the effect of deployment on mental health. Moreover, because of the incapacitation effect, we expect the effect to be smaller for the many-time deployed than for the onetime deployed. Although in principle the many-time deployed can seek help during deployment, the main consequence would be repatriation, thus making the soldiers less likely to seek help during and after deployment – especially if they want a military career. Consequently, we expect such soldiers to be more reluctant to seek help, as doing so may negatively affect that career (Hoge et al., 2004).34 Moreover, as we observe the many-time deployed for fewer years after their last deployment, this shorter time window could also explain a lower likelihood of the many-time deployed to be recorded in our registers. In addition, re-deployment as self-therapy cannot be excluded. The results in Table 4 (panel A) confirm this hypothesis, showing a significant effect of 0.4 percentage point. Thus, given that the sample mean on mental illness rate is about the same, this effect is smaller than among the one-time deployed but still significant and positive (8.5%). To check that the results are not driven by the choice of the imputed deployment year for the eligible men in the comparison group, we also randomly impute these deployment years. In Table 4, Panel B shows the result for the matched sample with random imputed deployment years. The 1.4 percentage point deployment effect in Panel B is not significantly different from the 1.0 percentage point in Table 3, column 1. Furthermore, we also estimate the model on a non-matched sample and only address the possible selection by controlling for individual characteristics in the DID setting, including the onetime deployed soldiers and the full sample of non-deployed eligible men (Table 4, panel C). In this sample we randomly impute the deployment year for the comparison group of all the eligible men, and find a 1 percentage point change (same result as in Table 3, column 1). Consequently, while the results hold with other imputation methods, the matching method for imputing dates is more relevant because it relies on appropriate covariates that influence mental health. As there could be further selection issues linked to the military service status of the non-deployed, we divided our sample of nondeployed eligible men according to their military service status, i.e.,

whether they had basic military training (military service) or not. The two groups differ only by AFQT mean score. Panels D and E in Table 4 show the estimate when we split the comparison group by service status. Again, the effect of deployment on mental health is a significant increase in the use of mental health services. The estimates are 16 and 26% increase relative to the mean, depending on the composition of the comparison group. These estimates are very similar to the 19% increase relative to the mean obtained with random dates and the full sample of non-deployed men in Panel C, Table 4. Consequently, our choice of keeping the non-deployed regardless of their military service status does not much affect the results. The sensitivity estimates from C to E, ranging from 16 to 26% increase relative to the mean (about 1 percentage point), are much less than Cesur et al.’s (2013) large effect (615.9% increase in PTSD diagnoses, i.e., 12.4 percentage points). This difference is clearly explained by our broader definition of mental health and by our studying deployment in general, not only PTSD diagnoses, among active duty soldiers in combat zones compared to their active duty counterparts in non-combat zones. Evidence of the effect of military deployment in studies using broader definitions of mental health is more scarce. Angrist et al. (2010) find that the Vietnam-era conscription induced a small overall increase in self-reported disability rates among Whites. However, the estimated effects are small and only marginally significant. Our effect is larger than that of Angrist et al. (2010) because we use a direct measure for mental health and look both at diagnoses and the purchase of N-drugs. In contrast to Autor et al. (2011), we have no problem with disentangling adverse health effects and benefit eligibility policies, because Denmark first implemented a veterans’ policy after 2010, with no changes in the benefit rules occurring during the analysis period and therefore no special rules benefitting soldiers. Indeed, to the contrary, we expect the soldiers to be more reluctant to seek help, as seeking help may have a negative impact on their military career (Hoge et al., 2004). Moreover, as we excluded some soldiers deployed in the Balkans in the 1990s – because we did not have their AFQT scores – our results likely provide the lower bound of the problem. As Bertrand et al. (2004) argue, a fairly long time series can cause serial correlation and result in inconsistent standard errors. To deal with this problem in our panel of 16 years, we collapse our data into one pre-period and one post-period. Table 4, Panel F, shows that the deployment effect is a significant 4.7 percentage points, equal to a 26% increase compared to the sample mean. This result is not significantly different from the results for the panel data in Table 3. As psychiatric diagnosis could be a more precise measure of mental health, compared to the purchase of N-drugs that may be used for different pathologies, Table 4, Panel G, shows the results only for psychiatric diagnoses. The results are similar – and even larger – when we use only psychiatric diagnoses as health measures. The combined health measure results are more significant, especially because of the increased number of occurrences for mental health in the sample. Our findings of the relationship between mental health and deployment are remarkably similar across the different subsamples. Overall, our sensitivity analysis shows that the results in Table 3 are reliable under different assumptions about the comparison group of non-deployed eligible men and the measure of mental health. This sensitivity analysis thus contributes to the external validity of our results.

34 Hoge et al. (2004) show that among the respondents who met the screening criteria for mental disorders, the main barriers for seeking mental health services were “I would be seen as weak (65%)”; “My unit leadership might treat me differently (63%)”; “Members of my unit might have less confidence in me (59%)”; “There would be difficulty getting time off work for treatment (55%)”; “My leaders would blame me for the problem” (51%); and “It would harm my career (50%).”

8. Summary and conclusion In the substantial literature on soldiers over the past 20 years, no previous studies have investigated pre- and post-deployment mental health – using a broader definition of mental health – for a

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

full population of soldiers as fully as this study does. We measure the effect of deployment in general, using administrative records to examine the entire population of deployed, and measure mental health as a combination of psychiatric diagnosis and the purchase of mental health-related medication. In our study, we have access to administrative records not suffering from non-response bias or other under-reported behavior stemming from, for example, soldiers fear for their career. Furthermore, having access to very detailed individual characteristics, all measured before the soldiers entered the military system, we are able to reduce the selection bias by using the matching procedure and the DID setting to control for time trends and other omitted variables affecting both soldiers and their comparison groups. In addition, we follow soldiers for several years both before and after deployment. We therefore show that deployment has a significantly negative impact on soldier’s overall mental health, not only PTSD. Thus, in contrast to previous studies, we establish an effect of deployment for a broader definition of mental health. More specifically, we find that the effect of deployment on soldiers’ mental health is significant after 4 years, compared to the comparison groups of non-deployed men. Thus deployment increases mental health problems by 1 percentage point for the deployed, i.e., a 24% increase compared to the sample mean. This result holds relative for different subsamples and comparison groups of eligible men. For the heterogeneous effects, although we find that low AFQT scores and growing up in an unstable social environment are correlated with more use of mental health services, we could not identify a significant effect for the interaction effects of being deployed and having low AFQT or having grown up in an unstable social environment. Only the interaction of deployment with deployed men who had a psychiatric diagnosis before age 18 was significant, increasing the use of mental health services. Moreover, we find that Danish soldiers sent to Afghanistan and the Balkans have a higher risk of having a mental illness. This finding is consistent with the context and level of exposure of these two missions: extreme war conditions for Afghanistan and lack of post- deployment help for the Balkans. Given that the effect on mental health may occur some years after the last deployment, and that soldiers deployed at the beginning of the analysis period have a higher probability of being recorded with a mental illness than soldiers more recently deployed, we therefore expect even higher incidences of mental illness for soldiers from the former Afghanistan.

203

Overall, our results show an indirect cost of war due to soldiers’ higher post-deployment risk of mental health deterioration. Therefore, when politicians and governments estimate the costs of war, they need to not only take into account the higher risk of mental health deterioration for all deployed soldiers but also consider the implications of this risk for both soldiers and their families – as well as the larger societal impact. Conflict of interest The authors declare that they have no conflict of interest. Acknowledgments We would like to thank the Danish Defense for providing us with the military dataset. This work was supported by Soldaterlegatet with financial support from Tryg, Lundbeck, Novo Nordisk, and Aase and Ejnar Danielsen’s funds. We also thank Sascha Becker and participants at the SFI Advisory Research Board, participants at the SFI health workshop with Lund University, participants at the CAM seminar at University of Copenhagen, and participants at the ESPE Conference 2014. Finally, we thank participants at the Health seminar at PAM, Cornell University, especially Sean Nicholson, who commented on this paper, and Natalie Reid for providing valuable language help. Appendix A. Additional material See Tables A1–A4 and Figs A1–A4 . Table A1 Number of deployments in the period 1996–2009. Number of deployments

1 2 3 4 5 6 7 8 9 >9 Total

Freq.

Percent

2664 1849 866 425 236 111 46 11 10 3 6221

42.82 29.72 13.92 6.83 3.79 1.78 0.74 0.18 0.16 0.05 100

Table A2 The one-time deployed. AFD year

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 Total

Age at the first mission

First deployment year Freq.

Percent

156 388 403 359 432 180 216 280 157 51 13 7 15 5 2 2664

5.86 14.56 15.13 13.48 16.22 6.76 8.11 10.51 5.89 1.91 0.49 0.26 0.56 0.19 0.08 100

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Total

Freq.

Percent

72 144 250 285 481 174 154 291 259 172 144 75 97 66 2664

2.70 5.41 9.38 10.70 18.06 6.53 5.78 10.92 9.72 6.46 5.41 2.82 3.64 2.48 100

19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 Total

Freq.

Percent

15 186 509 536 450 323 168 140 101 73 62 45 26 11 12 7 2664

0.56 6.98 19.11 20.12 16.89 12.12 6.31 5.26 3.79 2.74 2.33 1.69 0.98 0.41 0.45 0.26 100

Note: The table shows summary statistics for the one-time deployed. Furthermore, they are 19.51 years old on average in their AFD year (median: 19 years) and 23 years old on average at their first mission (median: 23). We have excluded soldiers for whom the AFD year and/or AFQT score was missing. On average there are three years between the AFD year and the first deployment, corresponding to about one year before starting military service and almost two years of preparing for a deployment.

204

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

Table A3 Psychiatric diagnoses (WHO-ICD-10) and drug groups defined as mental health-related medication for the analysis. Psychiatric diagnoses according to World Health Organization International Classification of Disease, Version 10

Drug groups defined by mental health-related medication classified according to the international ATC classification system (Anatomical Therapeutic Chemical)

Diagnosis Code Organic, including symptomatic, F00-09 a mental disorders F10-19 Mental and behavioral disorders due to psychoactive substance use F20 Schizophrenia Schizotypal and delusional disorders F21-29 b Mood [affective] disorders F30-39 c Neurotic, stress-related, and somatoform disorders F40-48 Behavioral syndromes associated with physiological disturbances and physical factors F50-59 Disorders of adult personality and behaviord F60-69 Other F-diagnosis Notes

Drug preparation: Nervous system medicine (N-drugs) Antipsychotics

ATC codes N05A

Anxiolytics and hypnotics

N05B

Sedatives Antidepressants Psychostimulants and nootropicse Drugs used in alcohol dependence

N05C N06A N06B N07BB

a The war in Iraq introduced a new military syndrome: Mild Traumatic Brain Injury (MTBI). Although the diagnosis is labeled “somatic,” it is identical with the psychiatric diagnosis of postconcussional syndrome (ICD-10: F07.2). b In this category we are particularly interested in depressive episodes (F32) and recurrent depressive disorder (F33). c This category includes reaction to severe stress and adjustment disorders (F43), including PTSD (F43.1). d In this category we are particularly interested in persisting personality change after catastrophic experience (F62.0). e In the analysis we have very few soldiers taking such drugs.

Table A4 Probit model of being deployed on pre-deployment characteristics. Parameter estimates Grew up in single-parent family (ref. not grew up in single-parent family)

0.079*** (0.021)

Placed in out-of-home care as a child (ref. not placed in out-of-home care as a child)

0.029 (0.042)

Immigrant (ref. Dane)

0.296*** (0.061)

AFD year

0.012 (0.007)

Average no. of AFQT tasks solved out of 78

0.003*** (0.001)

Single (ref. not single)

0.016 (0.158)

Combined measure of psychiatric diagnoses and purchase of NS drugs up to AFD year (ref. no mental health service use before AFD)

0.076 (0.103)

Born 1976 (ref. born 1975)

0.121*** (0.034)

Born 1977 (ref. born 1975)

0.103*** (0.035)

Born 1978 (ref. born 1975)

0.154*** (0.038)

Born 1979 (ref. born 1975)

0.172*** (0.041)

Born 1980 (ref. born 1975)

0.286*** (0.047)

Born 1981 (ref. born 1975)

0.282*** (0.053)

Born 1982 (ref. born 1975)

0.186*** (0.057)

Observations Log-likelihood

107,405 12420.704

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

Fig. A1. Deployed soldiers by first deployment year and number of deployments (one or many). Deployment period 1996–2009. Percent. The figure shows the percentage of the one-time and the many-time deployed measured in the period 1992–2009 and the total by mission year.

Fig. A2. Mission experience for the one-time and many-time deployed, born 1975–1982.

205

206

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

Fig. A3. Number of treated observations before and after the first deployment.

Fig. A4. Number of Danish soldiers: dead, wounded, or returnees (1996–2009). Note: Fig. A4 shows the number of dead, wounded, and returnees before the end of the mission (year of the mission in parentheses). The data for returnees is available only from 2005. SFOR: Stabilization Force (December 1996–2004) in Bosnia-Hercegovina; IFOR: UN-Mandated Implementation Force (December 1995, December 1996) in BosniaHercegovina. UNPROFOR: United Nations Protection Forces (1992–1995) in Bosnia-Hercegovina, Croatia, Serbia, Montenegro and Macedonia; ISAF: International Security Assistance Force (2002–2014) in Afghanistan. UNAMI: United Assistance Mission for Iraq (2003-). IRAK: Coalition in Iraq (2003–2009).

Appendix B. Before and after changes in mental health 35. Before and after changes in mental health To analyze whether a deployment has an effect on the mental health of individuals, we compare whether individuals who are experiencing a deployment are more subject to experiencing mental health deterioration. Our primary focus is on differences

after exposure between two similar groups of men. However, as previous mental health is a good predictor of future mental health, we are also interested in mental health differences in the two groups before the exposure. Table B1 shows that the deployed have significantly more mental health problems – measured with the combined measures – during the post-deployment period (i.e., 80% of all mental health problems for the deployed men) than the matched non-deployed eligible men (72% of all mental health

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

207

Table B1 Descriptives for pre- and post-mental health measurements for the deployed and non-deployed. Number and percent. Matched sample with imputed deployment date. Psychiatric diagnosis

Combined health measures

35

Only before Before and after Only after None Total Recorded after deployment (percent) Recorded both before and after (percent)

Deployed

Non-deployed

Deployed

Non-deployed

27 7 123 2507 2664 4.6 5.9

23 13 98 2530 2664 3.7 5.0

85 78 646 1855 2664 24.2 30.4

104 109 541 1910 2664 20.3 28.3

Note: The matched sample includes the 2664 deployed and the 2664 matched eligible men. A matched date means that within a matched pair the deployment year for nondeployed is imputed from the corresponding matched deployed. The combined health measures include psychiatric diagnosis and purchase of NS-drugs. Variables used in the matching procedure are single in AFD year, raised in single-parent family until age 17, placed in out-of-home care as a child up to 18 years, immigrant, AFD year, year of birth, no. of tasks solved on the AFQT on average, psychiatric diagnosis, and purchase of NS-drugs up to AFD year.

Table B2 Descriptives for type of diagnosis before and after deployment. Per thousand. Deployed

F00-29: Organic, including symptomatic, mental disorders F20: Schizophrenia, F21-29: Schizotypal and delusional disorders, Mood disorder F30-39: Mood disorder F40-49: Stress-related F50-59: Behavioral syndromes F60-69: Disorder of adult personality F10-F19: Substance use Other diagnoses Total

Non-deployed

Before

After

Before

After

Per.

Per.

Per.

Per.

1000

1000

1000

1000

1

3

2

6

2 4 0 2 1 4 13

13 20 2 4 3 3 49

2 3 1 1 1 4 14

8 12 1 5 6 5 42

Note: The table shows, per thousand, men recorded with some selected psychiatric diagnoses, i.e., the most severe recorded diagnoses in the period according to the diagnosis hierarchy. Diagnoses F00–F29 are grouped because of low frequencies.

problems). Results relative to psychiatric diagnoses show that 4.6% of the deployed men and 3.7% of the eligible men in our matched sample are recorded with a psychiatric diagnosis in the postdeployment period. However, when we look at the combined measures of psychiatric diagnosis and purchase of N-drugs, the percentages for post-deployment mental health problems are 24.2% for deployed men and 20.3% for eligible men in the matched sample (Table B2). References Ásgeirsdóttir, T.L., Corman, H., Noonan, K., Ólafsdóttir, Þ., Reichman, N.E., 2014. Was the economic crisis of 2008 good for Icelanders? Impact on health behaviors. Econ. Hum. Biol. 13, 1–19. Angrist, J., Krueger, A.B., 1994. Why do world war II Veterans earn more than nonveterans? J. Labor Econ. 12, 74–97. Angrist, J.D., Chen, S.H., Frandsen, B.R., 2010. Did Vietnam veterans get sicker in the 1990? The complicated effects of military service on self-reported health. J. Public Econ. 94, 824–837. Angrist, J.D., Chen, S.H., Song, J., 2011. Long-term consequences of Vietnam-era conscription: new estimates using social security data. Am. Econ. Rev. 101, 334– 338. Angrist, J.D., 1990. Lifetime earnings and the Vietnam era draft lottery: evidence from social security administrative records. Am. Econ. Rev. 80, 313–336. Angrist, J.D., 1998. Estimating the labor market impact of voluntary military service using social security data on military applicants. Econometrica 66, 249–288. Autor, D.H., Duggan, M.G., Lyle, D.S., 2011. Battle scars? The puzzling decline in employment and rise in disability receipt among Vietnam era veterans. Am. Econ. Rev. 101, 339–344.

35 The diagnosis can be recorded between birth and the AFD examination. These diagnoses are not necessarily known by the examining physician on the AFD. Moreover, a person recorded with, for example, an ADHD diagnosis but who is not taking medication for it can be declared eligible.

Bachman, J.G., Segal, D.R., Freedman-Doan, P., O’Malley, P.M., 2000. Who chooses military service? Correlates of propensity and enlistment in the U.S. armed forces. Mil. Psychol. 12, 1–30. Becker, S., Hvide, H., 2013. Do entrepreneurs matter? CEPR Discussion Paper 9295. Berger, M.C., Hirsch, B.T., 1983. The civilian earnings experience of Vietnam-era veterans. J. Hum. Resour. 455–479. Bertrand, M., Duflo, E., Mullainathan, S., 2004. How much should we trust differences-in-differences estimates? Q. J. Econ. 119, 249–275. Bingley, P., Lundborg, P., Vincent Lyk-Jensen, S., 2014. Opportunity Cost and the Incidence of a Draft Lottery. No.8057, IZA Discussion Paper, No. 8057. Breen, R., Karlson, K.B., Holm, A., 2013. Total, direct, and indirect effects in logit and probit models. Sociol. Methods Res. 42 (2), 164–191. doi:http://dx.doi.org/ 10.1177/0049124113494572. Brewin, C.R., Andrews, B., Valentine, J.D., 2000. Meta-analysis of risk factors for posttraumatic stress disorder in trauma-exposed adults. J. Consult. Clin. Psychol. 68, 748–766. Card, J.J., 1987. Epidemiology of PTSD in a national cohort of Vietnam veterans. J. Clin. Psychol. 43, 6–17. Cesur, R., Sabia, J.J., Tekin, E., 2013. The psychological costs of war: military combat and mental health. J. Health Econ. 32, 51–65. Conley, D., Heerwig, J.A., 2009. The long-term effects of military conscription on mortality: estimates from the Vietnam-era draft lottery. Natl. Bur. Econ. Res. Work. Paper Ser. 15105. Cuesta, M.B., Budría, S., 2015. Income deprivation and mental well-being: the role of non-cognitive skills. Econ. Hum. Biol. 17, 16–28. Dobkin, C., Shabani, R., 2009. The health effects of military service: evidence from the Vietnam draft. Econ. Inq. 47, 69–80. Dohrenwend, B.P., Neria, Y., Turner, B., Turse, N., Marshall, R., Lewis-Fernandez, R., Koenen, K.C., 2004. Positive tertiary appraisals and posttraumatic stress disorder in U.S. male veterans of the war in Vietnam: the roles of positive affirmation, positive reformulation and defensive denial. J. Consult. Clin. Psychol. 72, 417–433. Dohrenwend, B.P., Turner, J.B., Turse, N.A., Adams, B.G., Koenen, K.C., Marshall, R., 2006. The psychological risks of vietnam for U.S veterans: a revisit with new data and methods. Science 313, 979–982. Fontana, A., Rosenheck, R., 1998. Psychological benefits and liabilities of traumatic exposure in the war zone. J. Trauma Stress 11, 485–503. Forstmeier, S., Kuwert, P., Spitzer, C., Freyberger, H.J., Maercker, A., 2009. Posttraumatic growth, social acknowledgment as survivors, and sense of

208

S.V. Lyk-Jensen et al. / Economics and Human Biology 23 (2016) 193–208

coherence in former German child soldiers of world war II. Am. J. Geriatr. Psychiatry 17, 1030–1039. Gade, D.M., Wenger, J.B., 2011. Combat exposure and mental health: the long-term effects among US Vietnam and gulf war veterans. Health Econ. 20, 401–416. Hageman, I., Pinborg, A., Andersen, H.S., 2008. Complaints of stress in young soldiers strongly predispose to psychiatric morbidity and mortality: Danish national cohort study with 10-year follow-up. Acta Psychiatr. Scand. 117, 148–155. Hearst, N., Newman, T.B., Hulley, S.B., 1986. Delayed effects of the military draft on mortality. N. Engl. J. Med. 314, 620–624. Hoge, C.W., Castro, C.A., Messer, S.C., McGurk, D., Cotting, D.I., Koffman, R.L., 2004. Combat duty in Iraq and Afghanistan, mental health problems, and barriers to care. N. Engl. J. Med. 351, 13–22. Jordan, B.K., Schlenger, W.E., Hough, R.L., Kulka, R.A., 1991. Lifetime and current prevalence of specific psychiatric disorders among Vietnam veterans and controls. Arch. Gen. Psychiatry 48, 207–215. Keane, T.M., Fairbank, J.A., Caddell, J.M., Zimering, R.T., Taylor, K.L., Mora, C.A., 1989. Clinical evaluation of a measure to assess combat exposure. Psychol. Assess. J. Consult. Clin. Psychol. 1, 53–55. Kessler, R.C., Berglund, P., Demler, O., Jin, R.W.E.E., 2005. Lifetime prevalence and age-of-onset distributions of DSM-IV disorders in the national comorbidity survey replication. Arch. Gen. Psychiatry 62, 593–602. Kleykamp, M.A., 2006. College, jobs, or the military? Enlistment during a time of war. Soc. Sci. Q. 87, 272–290 (Blackwell Publishing Limited). Larson, G.E., Highfill-McRoy, R.M., Booth-Kweley, S., 2008. Psychiatric diagnoses in historic and contemporary military cohorts: combat deployment and the healthy warrior effect. Am. J. Epidemiol. 167, 1269. Linley, P.A., Joseph, S., 2004. Positive change following trauma and adversity: a review. J. Trauma. Stress 17, 11–21. Lyk-Jensen, S.V., Glad, A., 2016. Why do they serve? Changes and differences in motives of Danish soldiers deployed to peace-keeping and peace-enforcing missions. Def. Peace Econ. doi:http://dx.doi.org/10.1080/ 10242694.2016.1200220. Lyk-Jensen, S.V., Heidemann, J., Glad, A., Weatherall, C.D., 2012. Danske Hjemvendte Soldater: Soldaternes Psykiske Sundhedsprofil før Og Efter Udsendelse. SFI – Det Nationale Forskningscenter for Velfærd, København. Macklin, M.L., Metzger, L.J., Litz, B.T., McNally, R.J., Lasko, N.B., Orr, S.P., Pitman, R.K., 1998. Lower precombat intelligence is a risk factor for posttraumatic stress disorder. J. Consult. Clin. Psychol. 66, 323–326. Maguen, S., Vogt, D.S., King, L.A., King, D.W., Litz, B.T., 2006. Posttraumatic growth among gulf war I veterans: the predictive role of deployment-related experiences and background characteristics. J. Loss Trauma 11, 373–388. Mangum, S.L., Ball, D.E., 1989. The transferability of military-provided occupational training in the post-draft era. Ind. Labor Relat. Rev. 42, 230–245. Miech, R.A., Caspi, A., Moffitt, T.E., Wright, B.R.E., Silva, P.A., 1999. Low socioeconomic status and mental disorders: a longitudinal study of selection and causation during young adulthood. Am. J. Sociol. 104, 1096–1131. Munk-Jørgensen, P., Mortensen, P.B., 1997. The danish psychiatric central register. Dan. Med. Bull. 44, 82–84.

Neal, D.A., Johnson, W.R., 1996. The role of premarket factors in black-white wage differences. J. Polit. Econ. 104, 869–895. Orsillo, S.M., Roemer, L., Litz, B.T., Ehlich, P., Friedman, M.J., 1998. Psychiatric symptomatology associated with contemporary peacekeeping: an examination of post-mission functioning among peacekeepers in Somalia. J. Trauma Stress 11, 611–625. Park, C.L., Cohen, L.H., Murch, R.L., 1996. Assessment and prediction of stress-related growth. J. Pers. 64, 71–105. Price, R.K., Risk, N.K., Haden, A.H., Lewis, C.E., Spitznagel, E.L., 2004. Post-traumatic stress disorder, drug dependence, and suicidality among male Vietnam veterans with a history of heavy drug use. Drug Alcohol Depend. 76, S31–S43. Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41–55. Rosenbaum, P.R., Rubin, D.B., 1985. Constructing a control group using multivariate matched sampling models that incorporate the propensity score. Am. Stat. 39, 33–38. Schnurr, P., Friedman, M., Bernardy, N., 2002. Research on post-traumatic stress disorder: epidemiology, pathophysiology, and assessment. Psychoter. Pract. 58, 877–889. Segal, M.W., Segal, D.R., Bachman, J.G., Freedman-Doan, P., O’Malley, P.M., 1998. Gender and the propensity to enlist in the U.S. military. Gender Issues 16, 65–87. Seltzer, C.C., Jabon, S., 1974. Effects of selection on mortality. Am. J. Epidemiol. 100, 367–372. Shen, Y.-C., Arkes, J., Boon, W.K., Lai, Y.T., Williams, T.V., 2010. Effects of Iraq/ Afghanistan deployments on PTSD diagnoses for still active personnel in all four services. Mil. Med. 175, 763–769. Smith, T.C., Ryan, M.A.K., Wingard, D.L., Slymen, D.J., Sallis, J.F., Kritz-Silverstein, D., 2008. New onset and persistent symptoms of posttraumatic stress disorder self reported after deployment and combat exposures: prospective population based US military cohort study. BMJ: Br. Med. J. 336, 366–371. Teasdale, T.W., Hartmann, P.V.W., Pedersen, C.H., Bertelsen, M., 2011. The reliability and validity of the danish draft board cognitive ability test: Børge Prien’s Prøve. Scand. J. Psychol. 52, 126–130. Teasdale, T.W., 2009. The danish draft boards intelligence test, Børge Priens Prøve: psychometric properties and research applications through 50 years. Scand. J. Psychol. 50, 633–638. Tedeschi, R.G., Calhoun, L.G., 1996. The posttraumatic growth inventory: measuring the positive legacy of trauma. J. Trauma. Stress 9, 455–472. Wells, T.S., LeardMann, C.A., Fortuna, S.O., Smith, B., Smith, T.C., Ryan, M.A.K., Boyko, E.J., Blazer, D., 2010. A prospective study of depression following combat deployment in support of the wars in Iraq and Afghanistan. Am. J. Public Health 100, 90–99. Whyman, M., Lemmon, M., Teachman, J., 2011. Non-combat military service in the United States and its effects on depressive symptoms among men. Soc. Sci. Res. 40, 695–703. World Health Organization, 2004. ICD-10 : International Statistical Classification of Diseases and Related Health Problems / World Health Organization. World Health Organization, Geneva.