Recent findings point to high levels of food insecurity (FI) in veteran households. A 2012 survey of Minnesota-based veterans using the US Department of Veteran Affairs health-care system found high rates (nearly 27 %) of FI among veterans of the wars in Iraq and Afghanistan( Reference Widome, Jensen and Bangerter 1 ) and an analysis of data on veterans from the Veterans Aging Cohort Study who were also engaged with the US Department of Veteran Affairs found that nearly a quarter expressed some concern about having adequate food for themselves or their families( Reference Wang, McGinnis and Goulet 2 ). The Feeding America organization also gained national attention when it estimated that one in four active duty or reserve military households had sought food assistance from its national network of emergency food providers( 3 ). Data from the same report indicated that 15·3 % of emergency food client households had at least one member who had ever served in the US military( Reference Weinfeld, Mills and Borger 4 ).
Although these reports speak to the potential for high levels of food need and FI in US veteran households, they were limited by their reliance on highly selected samples( Reference Widome, Jensen and Bangerter 1 – 3 ), lack of a comparison group of non-veteran households( Reference Widome, Jensen and Bangerter 1 , Reference Wang, McGinnis and Goulet 2 ) or limited measures for FI( Reference Wang, McGinnis and Goulet 2 ). FI, which indicates that household ‘access to adequate food [is] limited by a lack of money or other resources’( Reference Coleman-Jensen, Gregory and Singh 5 ), is an important indicator of health for both children and adults( Reference Cook, Frank and Berkowitz 6 – Reference Hampton 8 ). Thus, documenting FI rates among veteran households is an important public health surveillance goal. Higher rates of veteran FI in national data would provide an important call to action for state and federal government to improve supports for veterans and their families.
Indeed, there is reason to believe that veteran households may be more likely to be food insecure than the average US household. FI is a measure of economic hardship. And while income is strongly and negatively associated with household FI( Reference Coleman-Jensen, Gregory and Singh 5 , Reference Rose 9 ), it is not the sole determinant and other factors relevant to management of household resources are also important. For example, research points to a connection between household FI and adult psychological well-being( Reference Noonan, Corman and Reichman 10 ), including psychosocial problems like post-traumatic stress disorder, depression, binge drinking and substance use disorders that may be associated with military service( Reference Larson, Adams and Mohr 11 – Reference Seal, Cohen and Waldrop 19 ).
The transition from military to civilian life often presents unique barriers to financial security, which may amplify the risk of experiencing FI. New veterans may experience challenges in the labour market due to a mismatch between skills acquired during military service and those required for civilian employment( Reference Kleykamp 20 , Reference Humensky, Jordan and Stroupe 21 ). In addition, the provision of basic needs as a by-product of military life may obviate the need for service members to develop sound money management skills( 22 , Reference Elbogen, Sullivan and Wolfe 23 ), which could translate into an increased risk for FI and other forms of economic hardship in civilian life.
On the other hand, veteran households may have important advantages that promote household food security. For one, the Veterans Health Administration provides 29 % of male veterans and 25 % of female veterans with regular and comprehensive medical care( 24 ), which accounts for higher rates of health insurance coverage among veterans relative to civilians( Reference Kramarow and Pastor 25 ). In addition, the Veterans Benefits Administration administers programmes that offer educational assistance, vocational training, employment services, specialty home loans and income support to certain groups of low-income veterans( 26 ). Veterans facing housing crises may be eligible for a range of specialized programmes( 27 ). Access to this array of benefits may decrease the risk of FI among veterans relative to their non-veteran peers, although evidence suggests that veteran households are less likely to participate in the Supplemental Nutrition Assistance Program (SNAP)( 28 ), indicating that veteran status is not a guarantee of access to benefits.
Veterans’ military experiences and household variables are diverse and may be important to consider. Length of service, combat exposure and role vary among veterans both within and across periods of service, and service era itself may demarcate important differences that could affect the risk of FI. Specifically, those who enlisted after the shift to an all-volunteer force rank less favourably than their non-veteran peers in terms of socio-economic status, educational attainment( Reference Gilroy, Phillips and Blair 29 ) and the presence of behavioural health problems( Reference Gilroy, Phillips and Blair 29 – Reference Rosenheck, Frisman and Chung 32 ), all of which are hypothesized to account for the increased risk of adverse social( Reference Greenberg, Rosenheck and Desai 33 , Reference Fargo, Metraux and Byrne 34 ) and economic( Reference Greenberg and Rosenheck 35 ) outcomes.
Two previous studies of material hardship examined whether household food insufficiency (a more limited measure of food hardship available in the Survey of Income and Program Participation) varied by the presence of veteran and disability status( Reference Heflin, Wilmoth and London 36 , Reference Wilmoth, London and Heflin 37 ). In comparison to households with no veterans and no disabled residents, the first study found that the odds of food insufficiency were significantly higher among disabled veteran households and the second found higher odds of food insufficiency among older adult households with disabled veterans and those with non-disabled veterans( Reference Heflin, Wilmoth and London 36 , Reference Wilmoth, London and Heflin 37 ). However, these two studies were limited by a focus on food insufficiency only, which is much rarer than FI and based only on a single survey item. Indeed, in the second study, no non-disabled veteran households reported food insufficiency, making comparisons with this group impossible( Reference Heflin, Wilmoth and London 36 ).
This review of the literature points to the possibility for both higher and lower rates of FI in veteran homes. Further, it suggests the importance of accounting for differences between veteran and non-veteran households and for distinguishing among different veteran cohorts. Using nationally representative data from 2005–2013 waves of the Current Population Survey – Food Security Supplement (CPS-FSS), the present study was guided by two specific aims:
-
1. to provide reliable and contemporary national estimates of household FI and very low food security (VLFS) by veteran status and most recent period of military service; and
-
2. to identify whether there are significant differences in rates of FI and VLFS after controlling for differences between veteran and non-veteran households.
Methods
Data
We pooled data from the 2005–2013 waves of the CPS-FSS( Reference Coleman-Jensen, Nord and Singh 38 ). The CPS-FSS is administered annually as a supplement to the December Current Population Survey (CPS) and provides national estimates of FI, which are published by the US Department of Agriculture( Reference Coleman-Jensen, Gregory and Singh 5 , Reference Coleman-Jensen, Nord and Singh 38 , Reference Coleman-Jensen, Nord and Andrews 39 ). Respondents to the CPS-FSS first complete the regular CPS which includes information on labour force participation, household demographics and composition along with questions related to current and previous military service. The CPS-FSS asks about FI, participation in food and nutrition programmes, and other household food dynamics. In each year, a portion of CPS households elected not to participate in the CPS-FSS. Thus, all of our analyses rely upon household-level supplement weights, which account for non-response and make the sample representative of the non-institutionalized population of the USA. About 0·9 % of cases were missing data on one or more variables and were dropped from the present analysis. Our final analytic sample consisted of 388 680 households from the nine survey years. Because all analyses were conducted with de-identified secondary data with no means to link information to individual respondents, the present study was considered to be not human subjects research and required no review by an institutional review board.
Measures
Food insecurity
The CPS-FSS contains the eighteen-item Food Security Module developed by the Economic Research Service of the US Department of Agriculture. The Module contains ten adult-referenced questions and eight child-referenced questions asked only of those households with children( Reference Bickel, Nord and Price 40 ) (see online supplementary material, Supplemental Table 1). We used constructed variables in the CPS-FSS( Reference Bickel, Nord and Price 40 , Reference Nord and Bickel 41 ) to classify households as having been food secure or having experienced FI or VLFS over the previous 12 months. VLFS is a particularly severe form of hardship and occurs when households limit food intake or experience disrupted eating patterns because of limited money or other resources( Reference Coleman-Jensen, Gregory and Singh 5 ).
Veteran status and period of service
The monthly CPS asked all respondents aged 17 years and older ‘Did you ever serve on active duty in the US Armed Forces?’ We coded respondents as veterans if they answered yes but were not currently in the Armed Forces and identified veteran households as those with at least one resident veteran. The CPS also asked veterans to identify up to four periods of service, with the following choices: ‘September 2001 or later’, ‘August 1990 to August 2001’, ‘May 1975 to July 1990’, ‘Vietnam War (August 1964 to April 1975)’, ‘February 1955 to July 1964’, ‘Korean War (July 1950 to January 1955)’, ‘January 1947 to June 1950’, ‘World War II (December 1941 to December 1946)’ and ‘November 1941 or earlier’. Because of small sample sizes, we recoded the last four of these groups into a broader category of ‘Korean War or earlier’. We coded each veteran’s most recent period of service and, in households with multiple veterans (0·7 % of the sample and 3·5 % of all veteran households), used the service of the veteran who had most recently been in the Armed Forces as the measure for the entire household.
Control variables
A primary aim of our analyses was to investigate whether differences in FI between veteran and non-veteran households persisted after controlling for other factors that might otherwise account for these differences. As noted above, veteran households might have previous or ongoing access to a number of programmes or benefits that could also be related to their food security status, including education, job training and housing programmes, or regular medical care.
Because we were unable to examine directly which benefits veterans received, our analyses controlled for a number of factors that might index their ultimate effect: housing tenure (housing was ‘owned or being purchased’, ‘rented for cash’ or ‘occupied without payment’) and housing type (‘house/apartment/flat’, ‘hotel or motel’, ‘mobile home’ or ‘other’); highest level of household education (‘less than high school’, ‘high school’, ‘some college’, ‘bachelor’s degree’ or ‘graduate school or higher’); current labour force status (‘employed’, ‘unemployed’, ‘out of the labour force’, ‘disabled’ or ‘active military’ (in non-veteran households only)); and receipt of food and nutrition assistance (‘SNAP/Food Stamp participation’, ‘WIC (Special Supplemental Nutrition Program for Women, Infants, and Children) participation’ or receipt of ‘free or reduced-price meals’ from the National School Lunch Program, School Breakfast Program or a Head Start programme (coded as 0 for households without children)). To account for the potential accumulation of such benefits, we included an additional control for the presence of multiple veterans in the home.
In addition, we controlled for a number of sociodemographic factors, some of which also might act as confounders: respondent sex; marital status (‘married, spouse present’, ‘married, spouse absent’, ‘widowed’, ‘divorced’, ‘separated’ or ‘never married’); race/ethnicity (‘white not Hispanic’, ‘black not Hispanic’, ‘Hispanic any race’, ‘American Indian not Hispanic’, ‘Asian/Pacific Islander not Hispanic’ or ‘multiple race, not Hispanic’); household poverty (‘income above 185 % of the federal poverty line’, ‘income below 185 % of the federal poverty line’ or ‘income not reported’ (only in 2009 or earlier)); the presence of immigrants in the household; the number of children and number of adults in the household; and age and age-squared (to account for the curvilinear relationship between age and household FI). Lastly, to rule out secular effects and potential unobserved differences in state policies or economic climate that could benefit (or detriment) veteran households, we included indicators for state of residence and survey year. We coded person-level variables (e.g. marital status, race/ethnicity, age, sex) based on data from the survey respondent in non-veteran households, from the veteran in households with only one veteran resident and from the veteran with the most recent period of service in multiple-veteran homes.
Analyses
Using the statistical software package Stata version 13, we ran two sets of analyses. First, we specified uncontrolled and controlled logistic regression models comparing rates of FI and VLFS in veteran and non-veteran households. Next, we re-ran these analyses after separating veteran households into most recent period of service. To ease interpretation of our regression results, we generated predicted probabilities of FI and VLFS according to veteran status and most recent period of military service. In supplemental analyses (available upon request), we re-specified all models first using probit regression, next after clustering standard errors at the state level and finally by dropping multiple-veteran households. In addition, because CPS households are surveyed for four months, drop out of the survey for eight months and then are surveyed again for another four months, we ran supplemental models clustering standard errors by household identification number to account for repeat households. Results from all of these models were nearly identical to those shown below.
Results
As shown in Table 1, just over 17 % of the sample was comprised of veteran households. The group with most recent service in the Vietnam War was the largest group of veterans (over 30 % of all veteran households). More than 13 % of all households experienced FI in the past 12 months and 5·1 % experienced VLFS.
Descriptive statistics of the study sample: US households (n 388 680) from 2005–2013 waves of the Current Population Survey – Food Security SupplementFootnote †
SNAP, Supplemental Nutrition Assistance Program; WIC, Special Supplemental Nutrition Program for Women, Infants, and Children.
† All estimates are based on the use of sample weights provided by the Current Population Survey.
‡ Variable applies to the head of household in non-veteran homes or the veteran with the most recent service in veteran homes.
Table 2 presents the results of uncontrolled and controlled logistic regressions. In the unadjusted models, veteran households had significantly lower odds of FI (OR=0·547, P<0·001) and the probability of FI was 0·084 compared with 0·144 in non-veteran homes. Results were similar for VLFS: veteran households had significantly lower odds (OR=0·601, P<0·001) of VLFS and the corresponding probability was substantially lower in veteran households (0·033 compared with 0·054). The bottom part of Table 2 presents estimates from fully controlled models. In these models, veteran status was no longer significantly associated with FI or VLFS.
Unadjusted and adjusted past-year household food insecurity and very low food security by veteran status among US households (n 388 680) from 2005–2013 waves of the Current Population Survey – Food Security SupplementFootnote †
SNAP, Supplemental Nutrition Assistance Program; WIC, Special Supplemental Nutrition Program for Women, Infants, and Children.
*P<0·05, **P<0·01, ***P<0·001.
† All models also control for respondent’s state of residence and the year of data collection. All estimates are based on the use of sample weights provided by the Current Population Survey.
‡ Referent category.
§ Variable applies to the head of household in non-veteran homes or the veteran with the most recent service in veteran homes.
In the top part of Table 3, which presents estimates from unadjusted models, the odds of FI were significantly lower for veterans of every period, especially those households with veterans of the Vietnam War era and earlier. Corresponding probabilities of FI were 50 % or less (between 0·034 and 0·075) of those for non-veteran homes (0·144). The unadjusted results for VLFS showed a similar pattern, although the odds of VLFS were not statistically different for households with veterans from the August 1990 to August 2001 and May 1975 to July 1990 eras.
Unadjusted and adjusted past-year household food insecurity and very low food security by most recent period of military service among US households (n 388 680) from 2005–2013 waves of the Current Population Survey – Food Security Supplement†
SNAP, Supplemental Nutrition Assistance Program; WIC, Special Supplemental Nutrition Program for Women, Infants, and Children.
*P<0·05, **P<0·01, ***P<0·001.
† All models also control for respondent’s state of residence and the year of data collection. All estimates are based on the use of sample weights provided by the Current Population Survey.
‡ Referent category.
§ Variable applies to the head of household in non-veteran homes or the veteran with the most recent service in veteran homes.
In results from adjusted models, which are reported in the lower part of Table 3, the findings were substantially changed: the odds of FI were not significantly different for household with veterans of the post 9/11 period, the period from 1955 to 1964, and from the Korean War or earlier. Odds of FI for households with veterans from the Vietnam War era were still lower but with much smaller predicted differences in the probability of FI: 0·125 compared with 0·133 for non-veteran households. Importantly, there was a change in direction of effect in adjusted models for households with veterans from the August 1990 to August 2001 and May 1975 to July 1990 eras, who had significantly higher odds (OR=1·172, P=0·0005 and OR=1·091, P=0·0174) and probabilities (0·148 and 0·141) of FI. In the fully adjusted results, veteran status was not significantly associated with VLFS.
Discussion
The present study used nationally representative data on American households from 2005 to 2013 to estimate rates of FI in veteran and non-veteran households. In contrast to the limited body of previous research, our analyses found rates of FI (8·4 %) and VLFS (3·3 %) that were significantly and substantially lower in veteran households than in non-veteran households. In analyses that separated veteran households by their most recent period of service, rates of FI and VLFS were again significantly lower for households with veterans of nearly every group. Thus, our uncontrolled estimates based on national household data provide a very different picture from that suggested in previous research( Reference Widome, Jensen and Bangerter 1 , Reference Wang, McGinnis and Goulet 2 ), with most veteran households at significantly lower risk for FI and VLFS than non-veteran homes. We consider the principal value of these findings as contributing to a clearer picture of the basic prevalence of FI and VLFS among households with veterans, particularly given their stark contrast with previous research. However, we are careful not to over-emphasize these results, which do not account for the many differences between veteran and non-veteran households.
Rather, our adjusted models (which controlled for sociodemographic factors along with a number of additional factors that might reflect benefits that accrue to veteran households) are likely more informative for policy makers. In these models, the difference in predicted probability of FI between veteran and non-veteran households was very small and not statistically significant. Thus, differences in the distribution of characteristics among veteran and non-veteran households along with state and year controls explained the apparent average advantage that veteran households had regarding food security. However, controlled models that separated veteran households by period of most recent military service found that results did not generalize across veteran eras. In these models, households with veterans from the Vietnam War still had significantly lower predicted probability of FI (0·125), although the difference between the corresponding probability for non-veteran households (0·133) was much smaller than in uncontrolled estimates. Also, for households with veterans who served from 1990 to 2001 and from 1975 to 1990, FI was significantly higher, 14·8 % and 14·1 % respectively. In fact, results suggest a substantial change in FI between older veterans who served in the Vietnam War or earlier and recent veterans from 1975 onwards. In supplemental controlled analyses (available upon request), we combined veteran households into these two larger groups and found that the odds of FI were significantly higher (OR=1·099, P=0·0006) among recent veterans and significantly lower (OR=0·914, P=0·0025) among older veterans compared with non-veteran homes.
There are several possibilities underlying these results. First, this split between older and more recent veterans may reflect differences in the composition of the US Armed Forces coinciding with the onset of an all-volunteer force. Thus, the differences in FI may reflect these compositional differences such as higher proportions of volunteers from impoverished communities, families with dysfunction( Reference Larson, Wooten and Adams 42 , Reference Kelley, Runnals and Pearson 43 ) and greater numbers of women. Further, the group of oldest veterans have accumulated a lifetime of advantages from the GI Bill and mortgage and health-care programmes. More recent veteran cohorts have accrued these benefits for a shorter period of time. An alternative possibility is that the welfare of veteran groups has been determined in part by macro trends in the US economy. Whereas the oldest group of veterans returned to job opportunities and an expanding middle class, contemporary veterans have returned to stagnant wages and diminished wealth available to the middle class( Reference Faberman and Foster 44 ). Newer veterans may also differ in the length and number of combat rotations, age at first entry into active duty and other factors. While our analyses controlled for a number of factors (household poverty, educational attainment and housing status) which would capture negative or positive selectivity, there may be subtle differences and interactions with the larger economy that we do not account for. Future research should seek to better understand the differences between veteran and non-veteran households and among veteran households, and how these differences contribute to material hardship like FI.
Perhaps most important, our results point to diversity among veteran households; one key finding is that households with veterans from the period May 1975 to August 2001 had slightly higher odds of FI than non-veteran households and post 9/11 veterans had no difference in odds. These results and those of the supplemental analyses described above suggest that recent veterans who served in 1975 or later may be at higher risk for FI and should be the recipients of targeted outreach to improve nutritional outcomes. We cannot project from these findings whether this disadvantage will continue, resolve or grow, but the food security of new-generation veterans should be closely monitored. It is important to note that the higher odds of FI among some recent veterans were independent of participation in SNAP, WIC and the National School Lunch and School Breakfast Programs, which are the three largest nutrition programmes operated by the US Department of Agriculture. It may thus be necessary to explore whether different types of intervention or special outreach to optimize participation in these national programmes can reduce hardship. However, we also note that despite our use of controlled models, our results do not allow us to make causal inferences regarding veteran status and food insecurity, so additional research is necessary before making firm policy recommendations.
The present study was not without limitations. The CPS contained only limited information on the experiences of veterans. The survey lacked information about length of service, deployment, combat exposure, rank and branch, and the nature of separation or discharge from the armed forces, all of which might help to better distinguish among different groups of veterans. The CPS is also cross-sectional, which precluded analyses that might have described the dynamic nature of FI in households. Lastly, because it is a household survey, the CPS does not include information on homeless veterans, who are inevitably at much higher risk of FI than veterans with more stable housing arrangements. Nationally representative data suggest that despite significant decreases in the past five years, there are still nearly 50 000 homeless veterans and veterans are over-represented in the homeless population in the USA( Reference Henry, Cortes and Shivji 45 , 46 ). Accordingly, the present findings are best categorized as representative of the differences in FI among veteran and non-veteran households. Additional analyses are necessary to understand whether accounting for homelessness among both veterans and non-veterans would affect the nature of the findings reported here.
Conclusion
In conclusion, using a large, nationally representative sample of all American households, these analyses represent an important complement to previous research that is based on highly selected groups of veterans or used a more limited measure of FI. Other study strengths are the use of nutrition programme participation and detailed demographic variables to explore many of the differences between veteran and non-veteran households that are likely linked to FI. While on balance veteran households were at substantially lower risk of FI than comparable civilian households, there is reason for concern among the most recent veterans. These findings support continued monitoring of the well-being of recent veterans and exploration of targeted outreach to ensure their full participation in nutrition programmes and other benefits.
Acknowledgements
Financial support: This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. Conflicts of interest: None. Authorship: D.P.M. helped conceptualize the study, prepared and analysed the data, and led in the writing of the manuscript. M.J.L. helped to conceptualize the study, assisted in the interpretation of results and assisted in the writing of the manuscript. T.B. assisted in the interpretation of results and assisted in the writing of the manuscript. E.D. assisted in the interpretation of results and assisted in the writing of the manuscript. Ethics of human subject participation: All analyses were conducted with de-identified secondary data with no means to link information to individual respondents; therefore the study was considered to be not human subjects research and required no review by an institutional review board.
Supplementary Material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1368980015003067
