Home Front: Post-Deployment Mental Health and Divorces

Abstract

Introduction

In recent years, as part of U.S. wartime operations since the attacks on America of September 11, 2001 (“9/11”), more than 2.3 million troops from all military services have deployed to Iraq and Afghanistan. These deployments have been longer and much more frequent than before 9/11 (Hosek and Martorell 2009) and, coupled with prolonged exposure to combat, may have been traumatic both for soldiers and their families. In fact, many soldiers who deployed to these countries suffer from war-related psychological problems (Cesur et al. 2013; Hoge et al. 2006; Milliken et al. 2007; Tanielian and Jaycox 2008). Ample anecdotal evidence emphasizes the burden such mental conditions place on military families: many postwar individual accounts paint the picture of acute marital strife, which often leads to divorce (see, e.g., Dao 2012).

The purpose of this article is threefold. First, we provide a systematic analysis of the effect of post-deployment mental health symptoms, and PTSD symptoms in particular, on the stability of military families in recent years. Second, we provide estimates of the extent to which mental health problems explain the increase in divorces documented by Negrusa et al. (2014). Third, we contribute to the larger debate on the overall costs imposed by recent wars on society. We use an individual-level longitudinal data set from U.S. Army administrative sources that we combine with post-deployment health screening data, available in the Post Deployment Health Assessment (PDHA) forms. The resulting data set includes individual deployment and marital history, as well as individual-level information on post-deployment mental and physical health for a large sample of Army active-duty soldiers over the period between March 2003 and June 2010.

To date, a limited number of studies have provided evidence on the relationship between post-deployment PTSD and marital dynamics. Moreover, these studies are often based on small cross-sections and limited survey samples, use potentially outdated data and a small number of covariates, and do not focus on divorce as their main outcome. They analyze the relationship between post-deployment PTSD and measures such as levels of anger (Chemtob et al. 1994), rates of domestic violence (Carroll et al. 1985), violent behavior within the home (Jordan et al. 1992), or a marital adversity index (Gimbel and Booth 1994). In this study, we focus exclusively on divorces by employing an extensive panel data set with a large number of recently serving soldiers. Also, our data set includes a rich set of individual sociodemographic characteristics that were unobservable to past researchers, such as aptitude scores, military rank, occupation, education, and time in service. In addition to providing new estimates of post-deployment PTSD symptoms on divorces in recent populations, our findings have two more advantages over those from previous studies: they do not suffer from the “reverse causality” issue that may be the result of divorce leading to PTSD, and they are obtained by controlling for a set of pre- and post-deployment individual characteristics.

Using longitudinal data spanning the period between 1999 and 2008, Negrusa et al. (2014) found that a longer time in deployment substantially increases the probability of divorce, with the effect being stronger for females and for members who deployed to Iraq and Afghanistan. They interpreted their findings in light of Becker’s standard family economics theory (Becker 1973) by viewing the deployment experience as a marital shock that destabilizes the marriage. Although Negrusa et al. (2014) advanced a number of hypotheses of why deployments increase divorce risk, they did not disentangle empirically the mechanisms through which deployments affect marriages. The current article fills this gap by isolating one such potential mechanism: namely, post-deployment mental health symptoms, and PTSD symptoms in particular. In this sense, we assess whether soldiers with post-deployment PTSD symptoms are more likely to divorce than soldiers without such symptoms, and then we estimate the number of divorces that are due to PTSD symptoms in the total number of total divorces caused by deployments. We also expand the work in Negrusa et al. (2014) by analyzing the effect of deployments and post-deployment mental health on the divorce risk of officers.

The impact of post-deployment mental health on divorces may be viewed as part of the overall costs of wars, along with other social costs quantified in terms of increases in health care costs resulting from combat-induced PTSD (Cesur et al. 2013), long-term health effects (Costa and Kahn 2010), lower employment rates in the civilian labor market (Autor et al. 2011), impact on household and family life (Angrist and Johnson 2000; Conlin and Heerwig 2011), or increase in violent behavior among veterans (Lindo and Stoecker 2012; Rohlfs 2010). Married soldiers account for more than one-half of the U.S. military, so an increase in the number of divorces resulting from mental health problems may further magnify the long-term consequences of wars in the form of spillovers on spouses and children, given that divorces exacerbate inequality and poverty (Ananat and Michaels 2008) and are associated with negative impacts on mental health and well-being (e.g., Kessler et al. 1998), children’s mental health even after transition to adulthood (Cherlin et al. 1998), and children’s educational attainment (e.g., Keith and Finlay 1988). Also, because large cohorts of individuals in their 20s and 30s leave the military every year to become civilians, these costs will not be confined only to the currently serving military personnel and their families but rather will have lasting impacts on large parts of the population.

Estimating a discrete-time survival model with fixed and time-varying military and demographic characteristics, we find that PTSD symptoms increase the divorce probability among Army enlisted personnel by 10 % to 20 % relative to soldiers who return from deployment without such symptoms, and by 50 % to 75 % for officers. We interpret these estimates as consistent with the hypothesis that PTSD symptoms represent marital shocks that could not have been anticipated by couples at the time of marriage formation. These shocks decrease the couple’s expected gains from marriage and thus increase the risk of divorce. We also find that even soldiers without post-deployment mental health symptoms are more likely to divorce relative to soldiers who do not deploy, which means that PTSD symptoms are not solely responsible for the increase in deployment-related divorces. Also, the relationship between divorce and PTSD symptoms does not vary by gender or by length of deployment, and the divorce hazard increases more steeply in the first months after deployment. Finally, in the case of families experiencing multiple deployments, most of the damage is done after the first deployment. In an exploratory instrumental variables (IV) analysis, we address the potential concern that those who develop PTSD symptoms may be inherently more likely to divorce. The direction and magnitude of the effects remain the same as in the main models, suggesting that endogeneity between post-deployment PTSD symptoms and divorce is unlikely to be present in our data.

Conceptual Framework

From a theoretical perspective, marriage formation is motivated by the nonnegative gains that spouses expect to achieve, and marital dissolution occurs when there are revisions in the determinants of match quality within the couple or when unanticipated shocks reduce the spouses’ ex-ante expected gains from marriage (Becker 1973, 1974; Becker et al. 1977). The family’s response to shocks depends on the degree to which those shocks are anticipated at the time of marriage. If the ex-post realization of an event experienced by the family is not worse than the couple’s ex-ante anticipations, the gains from marriage remain unchanged, and thus the marital stability of the couple is unaffected. If instead the family experiences a negative unanticipated event, the probability of divorce increases as the gains from marriage decline.

At the time of marriage, military families take into account that deployments may be an integral part of their married life. However, certain aspects of deployments may not be accurately anticipated, such as the timing, length, and frequency of deployments—and, in particular, whether the military member is exposed to combat or other traumatic events during deployment. As previously shown, these traumatic events may lead to mental health problems, but what is important for our purposes is that the actual occurrence of post-deployment mental health problems cannot be accurately anticipated by families at the time of marriage. Although the whole deployment experience can be considered an unanticipated marital shock, as in Negrusa et al. (2014), we view post-deployment mental health problems as one component of the overall deployment shock. In general, mental health problems are negatively correlated with marital satisfaction and stability in civilian populations (e.g., Bartel and Taubman 1986; Davila et al. 2003; Kessler et al. 1998; Whisman 2001). When such changes in marital satisfaction are unanticipated, they ultimately increase the probability of divorce.¹ Our article contributes to the empirical literature that analyzes the effect of unanticipated shocks on marital turnover (Becker et al. 1977; Charles and Stephens 2004; Doiron and Mendolia 2012; Weiss and Willis 1997).

Previous literature shows that deployments have positive aspects as well, such as applying skills acquired in training and a fulfilled sense of duty (Hosek et al. 2006), which may contribute to an increase in the marital gains of military families experiencing deployments. Karney and Crown (2007) found that deployments do not increase the risk of divorce, concluding that perhaps families are resilient and that the family programs may be providing sufficient support to keep families together. Other studies appear to suggest that positive aspects may sometimes be enough to compensate for the negative aspects of deployments. For example, Angrist and Johnson (2000), found that deployments increase the divorce rates of female soldiers but have virtually no effect on the divorce rates of males; and Conlin and Heerwig (2011), although being unable to identify a deployment effect, found that veteran status is associated with a slight decrease in the divorce probability for white veterans and with a potential increase for black veterans. However, the data in Karney and Crown (2007) covered only those individuals who entered service during the 2002–2005 period, a timeframe that may be insufficient for estimating a deployment effect. Comparatively, Angrist and Johnson (2000) and Conlin and Heerwig (2011) used cross-sectional survey data and focused on conflicts from past periods (i.e., the First Gulf War and the Vietnam War, respectively). Also, Pavalko and Elder (1990) found that World War II veterans were in fact more likely to divorce than nonveterans. Using data from recent years, Hosek et al. (2006) provided evidence that the positive aspects of deployment tend to dissipate quickly as deployment time increases, and Castro (2008) documented a significant decline in self-reported satisfaction of service members returning from Iraqi deployments. Finally, Negrusa et al. (2014), who estimated a strong deployment effect on divorces in recent years, argued that the positive aspects of deployment—including deployment-related pays—are unlikely to play a role in reducing the risk of divorce.

Empirical Framework

Using our rich longitudinal data set, we estimate the effect of post-deployment mental health symptoms on divorce in a discrete hazard framework with time-varying variables by comparing the divorce probability of deployed soldiers who report post-deployment mental health symptoms with that of deployed soldiers who do not report such symptoms after deployments.

A key identifying assumption is that mental health is uncorrelated with predispositional variables that in turn increase the risk of divorce even in the absence of deployments. This endogeneity between mental health and divorce is well known from the previous literature. For instance, Kessler et al. (1998) found that psychiatric disorders are associated with a substantially higher risk of divorce but also acknowledged that unobservable variables, such as childhood adversity or stressful living conditions, could lead both to psychiatric disorders and to subsequent adverse marital outcomes. To isolate the effect of interest from the confounding effects of predispositional variables, we focus on PTSD rather than other mental health problems that may be correlated with predispositional characteristics. Jordan et al. (1992) found that in a population of Vietnam War veterans, exposure to combat trauma is the factor that explains most of the variation in PTSD, and that variables such as childhood behavior problems, parental violence, or criminal problems of family members have only minor contributions to explaining the incidence of PTSD. Similarly, Smith et al. (2008) found that about 76 % of PTSD cases in recent years may be attributed to combat exposure during deployment, and Tanielian and Jaycox (2008) found that combat exposure is associated with odds ratios of having PTSD between 3 and 4. Based on preliminary evidence, Omalu et al. (2011) argued that PTSD may be the result of permanent brain damage, which in the case of recent deployments to Iraq and Afghanistan is caused by repeated blasts that result in shock waves that damage the brain. Not least, Cesur et al. (2013) documented a strong causal relationship between combat exposure and PTSD.

Another identifying assumption is that combat exposure, the most important predictor of PTSD symptoms, cannot be not known a priori by the soldier. Infantry soldiers may appear to be the most likely to experience combat during deployments, but Hosek and Martorell (2009) showed that exposure to combat after 9/11 was almost equally spread over all large categories of military occupations. This allocation of combat exposure among deployed soldiers ensures that the occurrence of PTSD symptoms is a random shock and could not have been anticipated by the family at the time of marriage. Therefore, by focusing on post-deployment PTSD, we have the advantage of studying a mental health condition that is unlikely to be endogenous with divorce, represents a valid marital shock, and has an incidence in the population that is unrelated to individual characteristics.

In our main regression model, the divorce hazard for individual i in period t, D_i,t, is a function of (1) being symptomatic of PTSD after deployment, PTSD_i,t; (2) an indicator variable for whether the soldier experiences any deployments up to time t, Depl_i,t; (3) cumulative number of months spent in deployment up to time t, CumDepl_i,t; (4) marriage duration, M_i,t; (5) calendar quarter dummy variables, Q_t; and (6) other fixed and time-invariant individual demographic and military characteristics, X_i:

(1)

An advantage of using the discrete hazard approach in Eq. (1) is that it assesses the effect of fixed and time-varying covariates on divorce in each period rather than at an arbitrary point in time after deployment. Military members go through multiple spells of deployment, and the model in Eq. (1) evaluates the impact of deployments in each marriage quarter t, given the deployment time the individual accumulates up to time t. Marriage duration enters the model as a time-varying variable to account for the possibility that the risk of divorce varies with time in marriage (e.g., Bergstrom 1997). Another advantage of Eq. (1) is its ability to deal naturally with right-censoring, given that many marriages are intact at the end of the observation period (Singer and Willett 1993). Also, Eq. (1) avoids reverse causality between mental health and divorce. Not only can mental health symptoms lead to divorce but divorce itself may represent a trauma that subsequently deteriorates mental health (Bartel and Taubman 1986; Kessler et al. 1998). With our data, we are able to avoid this issue because we construct time sequences of marriage, onset of mental health symptoms, and subsequent divorce.

The coefficient of interest in Eq. (1) is ⁠, which estimates the differential impact on the divorce hazard D_i,t, of screening positive for PTSD post-deployment relative to soldiers who return from deployments without screening positive for PTSD. One potential issue is that those who report post-deployment PTSD symptoms may have had these symptoms even before deployment, in which case would not reflect a pure post-deployment effect of PTSD symptoms. However, this is unlikely because based on Smith et al. (2008), the fraction of nondeployers with PTSD symptoms is very low (i.e., 2.3 %) and because only one-quarter of those few with PTSD before any deployment were eventually deployed. Not least, because most PTSD cases in recent years may be attributed to combat exposure, it is unlikely that is affected by pre-deployment PTSD symptoms.

As detailed in the next section, we build a longitudinal measure of post-deployment PTSD symptoms, in which we combine information on PTSD symptoms from the PDHA form and the Post Deployment Health Re-Assessment (PDHRA) form. The merit of this longitudinal measure is that it provides an update on the soldier’s PTSD symptoms, allowing us to account for the possibility that soldiers develop, or recover from, PTSD symptoms after the completion of their initial PDHA form. In Eq. (1), the PTSD symptoms variable “resets” after each deployment accompanied by a PDHA form.

In addition, Eq. (1) enables us to estimate the effect of deployment and time in deployment on the divorce probability. Although soldiers may engage in deployment avoidance or instead may volunteer for deployment, over our period of analysis, individual deployments in active duty were strongly tied to unit deployments; unit deployments, in turn, were unrelated to individual unit members’ preferences for deployment but rather depended on higher-level decisions regarding which unit to deploy and when (Cesur et al. 2013; Engel et al. 2010; Lyle 2006). It follows that the individual deployment can be viewed as a “treatment” applied independently of individual preferences. Coupled with the heterogeneity in the deployment experience and the notion that couples could not have formed accurate deployment expectations (Negrusa et al. 2014), the coefficient measures the effect of deployments on the divorce hazard. Identification of this effect is ensured by keeping in the regression sample soldiers who have not yet deployed as of time t and soldiers who never deployed.² If prevalence of PTSD symptoms is nonzero among nondeployers, provides a conservative estimate of the true deployment effect—independent of PTSD symptoms—because nondeploying PTSD symptomatics are potentially more likely to divorce than nondeployers without PTSD symptoms. Remember that is estimated by comparing the divorce probability of deployers without PTSD symptoms with that of nondeployers without PTSD symptoms. However, this is unlikely to be a serious issue because, as discussed earlier, the fraction of nondeployers with PTSD symptoms tends to be very low.

Finally, the coefficient on the cumulative time in deployment, γ₃, provides a measure for whether more time deployed around the mean duration has an additional impact on the hazard of divorce. In vector X_i, we include age at first marriage, race, gender, number of children, education at the time of enlistment in the Army, linear and quadratic terms for time since last deployment, a quadratic term for cumulative time in deployment as of the current period, a quadratic term for marriage duration, Armed Forces Qualification Test (AFQT) scores (indicative of mental aptitude), 10 indicators for military occupation specialties (MOS), and military rank in the current period. A detailed list with all the variables used in the analysis is provided in Fig. S1 in the Online Resource 1.

Data Description and Key Measures

We use multiple sources of data to link deployment and marital histories with individual-level data on post-deployment mental health. Information on military occupation, education, pay grade, gender, race, and AFQT come from Proxy Perstempo, an individual-level longitudinal file on active-duty soldiers that is administered and updated quarterly by the Defense Manpower Data Centers (DMDCs). Deployment information is inferred from administrative files: namely, pay records. From the Defense Enrollment Eligibility Reporting System (DEERS), which is a computerized panel database of military soldiers and their families, we obtain information on the marital status and number of children of soldiers. We obtain information on mental health symptoms from the PDHA and PDHRA forms. All returning soldiers are required to fill out the PDHA form around the date when their deployment ends. They are also required to fill out the PDHRA form within three to six months from their deployment end date. The PDHA and PDHRA forms include a series of detailed questions about the soldier’s general health, physical symptoms, mental health concerns, and concerns about exposure to harmful chemicals. All soldiers who complete the two forms are interviewed immediately by a primary care provider, who determines whether referral for further evaluation is necessary.

The diagram in Fig. 1 shows the temporal sequence of enlistment, marriage, deployment, completion of PDHA form, completion of PDHRA form, subsequent deployment, and potential divorce. Figure 1 also helps illustrate our conceptual framework of marriage formation, predeployment marital expectations, and how the marital surplus may be affected by deployment-related marital shocks. The expected value of the marital surplus at the time of marriage, E(MS), is a function of personal traits and, in the case of military families, of expectations regarding deployments and their consequences. As discussed earlier, the probability of divorce is smaller the greater E(MS) and the smaller the variance of unanticipated shocks affecting the gains from marriage. If the couple has a marital surplus after their first deployment of MS₀ that is the same as E(MS), their marital stability should not be affected. If instead the marital surplus after the second deployment, MS₁, is smaller than E(MS) because of post-deployment mental health symptoms, the couple’s risk of divorce increases.

We construct our PTSD symptoms measures using the four-item screen for PTSD available in the PDHA and PDHRA forms. The screen for PTSD, developed by the National Center for PTSD for primary care settings (PC-PTSD), includes questions covering four symptoms of PTSD: reexperiencing trauma, numbing, avoidance, and hyperarousal.³ Following Hoge et al. (2006), we consider endorsement of any two of the four questions to be symptomatic of PTSD. Although we do not observe a PTSD diagnosis, we are interested in evaluating the way in which post-deployment symptoms and manifestations of PTSD affect the married life of soldiers. Even if soldiers are not diagnosed with PTSD, their PTSD symptoms may represent a shock for the family, potentially causing interpersonal difficulties and, ultimately, divorce. Figure 2, which compares the annual divorce rates and the incidence of self-reported PTSD symptoms, suggests a potential correlation between divorces and PTSD symptoms.⁴

For soldiers with multiple deployments, we use the information from the most recent PDHA form as of quarter t. We build a second measure of PTSD symptoms by combining PTSD information from two instances: (1) right after soldiers return from deployment (i.e., PDHA form) and (2) three to six months after a given deployment (i.e., PDHRA form). We refer to this variable as our longitudinal PTSD measure because it uses information from two instances following a given deployment. The information on post-deployment PTSD symptoms derived from the PDHA form is updated only for soldiers who fill in a subsequent PDHRA form. For soldiers who do not fill out the PDHRA form, the longitudinal measure includes information from the PDHA form only. The advantage of our longitudinal measure stems from the fact that, when available, subsequent information confers a valuable update on whether symptoms disappear over time, as well as on the timing of symptoms onset. Previous studies have argued that soldiers may willingly underreport symptoms in order to avoid delays in being reunited with their families or for fear of stigma (e.g., Hoge et al. 2004). In addition, even if individuals correctly report their PTSD symptoms or other mental health symptoms on the PDHA form, these conditions could possibly subside as soon as soldiers are reunited with their families. Conversely, perhaps the PTSD symptoms appear only months after the end of deployment. However, some degree of measurement error may be induced by the longitudinal PTSD measure because we do not know when the symptoms start or stop between PDHA and PDHRA completion.

The PDHA form also includes a two-question screen for depression and inquires about a range of physical and other mental health symptoms. We use this additional information to construct an indicator for depression symptoms, as well as a general measure, named “any mental health concern,” that was also designed and used by Hoge et al. (2006).⁵, ⁶

In the analytic sample, which includes active-duty Army soldiers who entered service between March 1999 and June 2010, we keep only individuals who married after enlistment to ensure that both spouses are aware at marriage that they are becoming a military family and thus form appropriate expectations regarding their gains from marriage. Ultimately, our sample includes: (1) soldiers who deployed while in their first marriage and filled out at least one PDHA form, and (2) nondeploying soldiers. This leaves us with 69,557 soldiers and 360,012 soldier-quarter observations. Applying the same restrictions, we obtain a sample of 8,181 officers and 48,086 individual-quarter observations. More details on the construction of the analytic sample are provided in Online Resource 1.

As reported in Table 1, 86.0 % of enlisted personnel filled out at least one PDHA form, and 70.4 % have at least one PDHRA form, bringing the follow-up rate to 82 % (0.704/0.860). Similarly, 90.0 % of officers have at least one PDHA form, and 81.6 % have at least one PDHRA form. The fraction of soldiers with two or more PDHA forms is 16.6 % for enlisted members and 15.6 % for officers. Given that we include only PDHA completers in our sample, the fraction of deployers is the same as the fraction of individuals with at least one PDHA form (i.e., 86.0 % of enlisted members and 90.0 % of officers).

Because time in marriage is a time-varying variable, we assess the average divorce hazard at 2.9 years in marriage for enlisted personnel and at 3.4 years in marriage for officers. At the time of divorce or exit from sample, the average time in marriage is 3.2 years for enlisted personnel and 3.9 years for officers. The cumulative time spent in deployments is, on average, 14.9 months for enlisted personnel and 14.8 months for officers. At divorce or exit from the sample, total time spent in deployments is an average of 15.8 months for enlisted personnel and 16.2 months for officers. The average age at marriage for officers is 25.4, almost three years higher than that for enlisted members.

As shown in Table 1, about 12 % of enlisted members and 7.6 % of officers screen positive for post-deployment PTSD. The longitudinal PTSD measure has an average that is slightly higher: 13.5 % among enlisted personnel and 7.7 % among officers. These rates are in line with Milliken et al. (2007), who found that about 12 % of Army active-duty soldiers returning from Iraq screened positive for PTSD at deployment end, and 17 % screened positive for PTSD in their follow-up PDHRA form.

The fraction of soldiers with post-deployment depression symptoms is higher: approximately 32 % for enlisted personnel and 19 % for officers. About 27 % of enlisted personnel and 14 % of officers screen positive for “any mental concern” after deployments. However, only 10.2 % of enlisted personnel and 3.4 % of officers state that they have sought help for a mental concern, probably because of the stigma generally associated with seeking mental health treatment. Also, very few soldiers are referred for family counseling after PDHA completion.⁷

The PDHA form asks respondents whether their health improved, stayed the same, or worsened during deployment. About 18 % of enlisted personnel and about 15 % of officers declare that their overall health worsened during deployment. Following Hoge et al. (2006), we construct an indicator variable for combat exposure, taking the value of 1 if the individual responds in the affirmative to at least one of the questions of whether they saw dead bodies, used a weapon, or saw anyone being killed. Finally, the PDHA form asks soldiers whether they visited the sick bay during a deployment. About two-thirds of soldiers made at least one visit, and the average number of visits was 3.0 for enlisted members and 2.6 for officers.

Results

In Table 2, we present the regression coefficients for enlisted personnel obtained by estimating discrete hazard models based on Eq. (1).⁸ The estimates in column 1 of Table 2 show that soldiers who are deployed have a higher divorce probability than that of married soldiers who are not deployed. The estimated difference in the log-odds ratio of divorce between these two groups is 1.93. The PTSD coefficient in column 1 shows that post-deployment PTSD symptoms differentially impact the divorce risk of enlisted soldiers, relative to deployed soldiers who are not PTSD symptomatic. The odds of divorce of PTSD symptomatics are estimated to be 1.13 (= e^0.124) times higher than that of non-PTSD symptomatics.

The second column of Table 2 shows a model that allows us to separately measure a baseline effect of ever being deployed, an additional effect of deployment duration, and a differential impact of screening positive for PTSD on the divorce hazard. The baseline effect of deployments remains of the same order of magnitude as in column 1, and the PTSD differential impact decreases slightly from 0.124 to 0.112, without losing its statistical significance. Also, consistent with Negrusa et al. (2014), we find that a higher cumulative time in deployment has a strong positive effect on the divorce risk of enlisted families. Next, in the third column of Table 2, we include an additional interaction term to test whether the contribution of post-deployment PTSD symptoms to divorces varies with time deployed and find no additional PTSD effect coming through increased deployment durations.⁹

The PTSD variable used in Models 1–3 in Table 2 is based on information available only in the PDHA form. Using the longitudinal measure of PTSD, which allows us to observe the soldier’s PTSD symptoms at two points in time after return from a given deployment, we find that the new estimate of PTSD symptoms in column 5—our preferred model—is larger and more precisely estimated than in column 2, corresponding to an increase in the log odds of divorce of 0.182. This difference is consistent with the scenario of later onset of PTSD symptoms as well as with that of stigma-related underreporting of symptoms on the PDHA form. Given that soldiers who are medically evacuated because of severe injuries do not complete the PDHA form (Heaton et al. 2012), our effect of post-deployment PTSD symptoms may be underestimated if these soldiers are both more likely to develop PTSD symptoms and more likely to divorce.

To facilitate the interpretation of our regression coefficients, we construct predictions of the divorce hazard and the cumulative divorce hazard, using the coefficients from column 5 of Tables 2 and 3. In Fig. 3, we present the predicted quarterly divorce hazard of enlisted personnel at different points in time after returning from a first deployment of 12 months. Soldiers who report PTSD symptoms have divorce hazards between 1.1 % 6 months post-deployment and 1.6 % 48 months post-deployment. These predicted values are 21.4 % larger than those among soldiers without PTSD symptoms, whose predicted divorce hazard ranges between 0.9 and 1.3 %. Furthermore, in Fig. 4, we plot the predicted cumulative hazard for soldiers who experience a first deployment of 12 months in their first years of marriage. We estimate that individuals who deploy for 12 months in their first two years of marriage have a cumulative hazard of divorce of 4.6 % if they do not report PTSD symptoms and a cumulative hazard of 5.5 % (20 % higher) if they report PTSD symptoms. Comparing these predictions with those of soldiers who do not deploy, the predicted cumulative divorce hazards of nondeploying enlisted personnel calculated at two, three, four, and five years in marriage are 1.4 %, 2.8 %, 4.1 %, and 5.3 %, respectively.

Table 3 reports the same models as in Table 2 estimated on the sample of officers. The divorce hazard of deploying officers is higher than that of nondeploying officers, but the difference is not as large as in the case of enlisted personnel. However, the divorce hazard of officers with PTSD symptoms is much higher than that of officers without such symptoms. Figure 5 shows that the estimated impact of PTSD symptoms is larger in the case of officers at all marriage durations considered. For instance, officers who have PTSD symptoms after a 12-month deployment have a cumulative divorce hazard of 4.9 %, and officers without such symptoms have a much smaller cumulative divorce hazard of 2.8 %. To test the statistical significance of the differential effects on enlisted personnel and officers, we estimated models in which we pooled enlisted personnel and officers (not shown for space considerations). In these models, we interacted the main variables of interest with an indicator for officer and estimated statistically different effects (at levels lower than 1 %) of deployment and PTSD symptoms for enlisted relative to officers. The effects for enlisted and officers, respectively, are similar in magnitude to the estimates presented in Figs. 4 and 5.

This higher effect of PTSD symptoms suggests that for officers, even more so than for enlisted personnel, incidence of PTSD symptoms is an important negative shock that increases the risk of divorce. Given the typical functions assumed by officers, which translate into less exposure to combat or other traumatic events during deployments, having post-deployment PTSD symptoms is probably much less expected by officers’ families than by the families of enlisted personnel, and therefore more disruptive for officer marriages. Consistent with our conceptual framework, a larger negative marital shock has a stronger impact on the probability of divorce. Also, the smaller effect of deployments on the divorce risk of officers is in line with our theory. Deployments may be less of a shock for officers if they are capable of forming more accurate ex-ante deployment expectations than enlisted personnel families.

The estimated effects of other demographic characteristics on the risk of divorce are similar with findings from previous literature. Like Weiss and Willis (1997), we find that having children decreases the risk of divorce, as does having more education. Hispanics are relatively more insulated from divorce relative to whites, and getting married at a younger age increases the divorce risk. In addition, consistent with Angrist and Johnson (2000) and Negrusa et al. (2014), in a model where we interact the deployment variable with the female dummy variable (not shown), we find that female soldiers are much more likely to divorce as a result of deployment than male soldiers. Also, although the PTSD variable captures determinants of mental health, the variable “health worse” accounts for the impact of physical health, which appears to reduce the divorce hazard in our models. This is consistent with Charles and Stephens (2004), who also found that negative shocks in physical health may actually strengthen the marriage, in opposition with other negative shocks, such as loss of income.

Additional Results

Next we assess the robustness of our results by focusing on gender differences, the period of PDHA completion, and a subsample that includes data only from the first PDHA form.

We explore the possibility that PTSD symptoms affect female and male soldiers differently. We find that for both PTSD measures, the relationship we estimate between PTSD symptoms and divorce is the same for male and female enlisted members. The coefficient on the interaction term between PTSD and an indicator for female is statistically insignificant for both enlisted personnel and officers, as reported in columns 1 and 2 of Table 4 and columns 1 and 2 of Table 5, respectively. Therefore, although women’s marriages seem to be more negatively affected by deployments relative to male soldiers, the effect is not likely to be driven by post-deployment PTSD symptoms.

Because of increased interest from policymakers regarding soldiers’ post-deployment mental health, both the PDHA and PDHRA forms were slightly changed, starting in March 2008, to ensure more comprehensive reporting of soldiers’ psychological problems. Including an interaction term between PTSD and the post-2008 period, we conclude that although the effect of deployments on divorce increases after 2008, the same is not true for any of the PTSD estimates (columns 3 and 4 of Table 4).

To this point in our estimation, we have included all observations for any given soldier regardless of the number of deployments and subsequent PDHA/PDHRA forms. To understand whether our estimated relationship between PTSD symptoms and divorce is the result of an accumulation over multiple deployments, we now compare our estimates with those obtained in a subsample restricted to the soldiers’ first PDHA form. As shown in the rightmost columns in Tables 4 and 5, both in the case of enlisted members and officers, the PTSD coefficient estimates are close to those generated by the main sample, suggesting that most of the effect on divorce is driven by the first deployment (or rather, the first deployment associated with a PDHA completion). Also, the effects estimated using the two PTSD measures remain about the same as in the main case, with the PTSD longitudinal estimates being larger and more precisely estimated than the estimates based on the PTSD cross-sectional measure.

Importantly, because most of the effect of PTSD symptoms on divorce occurs early in the career, the main estimates presented in Tables 2 and 3 are not the result of soldiers’ selection into the analytic sample at the time of reenlistment. After completing a typical term of four years, most soldiers have the possibility to reenlist. If, however, the soldiers that decide to reenlist are more likely to divorce than the initial entrants, our estimate of the impact of PTSD symptoms on divorce would be overstated. However, given that the estimates using only data from the first PDHA form are virtually the same as those from Tables 2 and 3, we conclude that our estimates are unaffected by this type of selection.

In Table S2 in Online Resource 1, we discuss instrumental variable models dealing with the potential endogeneity between PTSD symptoms and divorces, as well as estimates from models with other mental health measures (Table S3). In the models in Table S2, we instrument for post-deployment PTSD symptoms using information on whether the individual visited the sick bay during deployment, the number of visits in the sick call, and whether the individual experienced combat. Estimating two-stage least squares (2SLS) models at different points in time after the deployment end date (6, 12, and 18 months), we obtain IV estimates of PTSD symptoms on divorce that are positive and, in most cases, statistically significant. These models should be viewed with caution because their validity depends on the choice of instrument; however, they nonetheless provide evidence that endogeneity between post-deployment PTSD and divorces is not an issue in our case.

Finally, using the regression coefficients from columns 5 in Tables 2 and 3, we construct predictions of the proportion of divorces that are explained by post-deployment PTSD symptoms. At the means of the data, we calculate that the fraction of deployment-related divorces that are due to post-deployment PTSD symptoms is 17 % for enlisted personnel and 40 % for officers.

Conclusions

In this study, we combine longitudinal administrative data on deployments and marital histories with data measuring post-deployment PTSD symptoms to provide novel estimates of the relationship between post-deployment PTSD symptoms and the divorce risk among active-duty U.S. Army soldiers serving in recent wars. Using these data, we are able to control for a large number of fixed and time-varying sociodemographic and military characteristics, such as pre- and post-deployment marital histories, cumulative time deployed, and change in PTSD symptoms over time. Our estimates indicate that soldiers who experience deployments are more likely to divorce than their nondeployed counterparts and that soldiers who screen positive for PTSD after deployment are more likely to divorce than deploying soldiers without such symptoms. In addition, we estimate a stronger relationship between post-deployment PTSD symptoms and the divorce risk for officers relative to enlisted members. An exploratory instrumental variable analysis of whether post-deployment PTSD symptoms and divorces are endogenous provides credible evidence that this issue is unlikely to be present in our data.

In light of increased concern among policymakers about maintaining and improving the well-being of military families (White House Report 2011), our study highlights several issues that may be relevant in designing new policies for military families. Our most important result is that post-deployment mental health symptoms, such as PTSD, have consequences that are not limited to the individual soldier but spill over to the relationship with the soldier’s spouse and affect the stability of their marriage. Divorce itself may be a relevant outcome, given its documented negative effects discussed earlier. In addition, from a policy perspective, it may be important to keep military marriages together because soldiers from stable marriages may be more likely to seek treatment for mental health problems. Milliken et al. (2007) found that military spouses are more willing to seek medical care for themselves as well as for their families. Moreover, designing new programs for married soldiers suffering from PTSD symptoms that would co-opt their spouses may be part of a broader set of intervention and treatment policies aimed at enhancing the psychological well-being of soldiers and their families. Also, our study suggests that after a soldier is observed to be PTSD-symptomatic, the effect on divorce is already apparent. The direct implication is that timely intervention after the first positive screen for PTSD is critical.

Prior research has found that the wives of deployed soldiers have poorer mental health than wives of nondeployed soldiers (Mansfield et al. 2010). Further research may be needed to understand how the combined post-deployment mental health problems of both spouses affect marital stability. Emerging literature, including Mansfield et al. (2010), shows evidence of spousal PTSD resulting from exposure to the service member’s PTSD. Most likely, spousal PTSD exacerbates the interpersonal difficulties within the family, thus further increasing the risk of divorce. If so, our coefficients on post-deployment PTSD symptoms represent a lower-bound estimate of the actual effect of post-deployment PTSD symptoms on the risk of divorce. Also, although post-deployment PTSD is unlikely to be caused by family stressors, our estimates of the average effect of post-deployment PTSD symptoms on divorces may be limited in the absence of data on those stressors. This would be important to consider in future research. Finally, PTSD symptoms may also affect the marriage prospects of single soldiers (Teitler and Reichman 2008). To the extent that young soldiers with mental health symptoms are more likely to remain unmarried or enter more fragile marriages, understanding the relationship between PTSD symptoms and the soldiers’ marital prospects may be an important venue for future research as well as a focus for new policies and interventions.

Acknowledgments

This research was conducted while the authors were affiliated with the RAND Corporation. The authors gratefully acknowledge the funding and support from the Office of the Undersecretary of Defense. We especially thank Beth Asch, James Hosek, Juergen Maurer, Amalia R. Miller, Sonia Oreffice, Climent Quintana-Domeque, Rajeev Ramchand, Terri Tanielian, and John T. Warner for their valuable suggestions and comments. We thank Adria Jewell and Arthur Bullock, whose contributions were critical in developing the data set that was used in the empirical analysis. We also thank Audrey L. Luken, Program Manager in the U.S. Army Medical Command, for her assistance in obtaining the PDHA and PDHRA data for this project. This study does not reflect the official position of the Department of Defense. All opinions, as well as any remaining errors, belong solely to the authors.

Notes

References