Causal modeling of substance use and mental health among male military veterans

ABSTRACT

Objective: In order to understand the intersecting effects of race/ethnicity and military service on substance use and mental health, this study maps the unique differences between military and nonmilitary men in order to provide nuanced information for mental health professionals and behavioral interventionists in order to address mental health and other substance use disorders among military populations.Methods: Using structural equation modeling to examine mental health and substance use among male military veterans, this study sample contains information about 449 male veterans from the National Longitudinal Study of Adolescent to Adult Health (1994–2008): Waves I and IV in-home interviews. Results: Models for substance use and mental health patterns were not similar between veterans and non-veterans and or racial/ethnic groups. Overall, Pre-9/11 veterans, depression and anxiety significantly predicted higher cocaine use and higher prescription drug use. Among Post- 9/11 veterans, depression and PTSD predicted greater use of marijuana, methamphetamine and prescription medications. Conclusions: Study findings highlight opportunities for potential interventions and have implications for the design of substance use programs. The evidence from this study supports the need to develop integrated prevention interventions that build emotional resilience and coping mechanisms of participants with military service

KEYWORDS:

• Mental health
• veterans
• substance-use

Notes

1. SEM allows one to operationalize a conceptual theoretical framework and its validation indices can verify several dimensions simultaneously measuring various aspects of mental health and substance use.

2. s. Men are considered because men are twice as likely to be stressed for an upcoming deployment than women and more likely to cope with stress by drinking (Bray, Fairbank, & Marsden, 1999). most veterans are men, 85% respectively (U.S. Department of Defense,2015).

3. SEM models could be meaningfully tested even if the sample size is small (Hoyle, 1999; Hoyle & Kenny, 1999; Marsh & Hau, 1999), but usually, N = 100–150 is considered the minimum sample size for conducting SEM (Tabachnick & Fidell, 2001).

4. As recommended by Muthén and Muthén (2002), the stability of results was tested by running the analyses multiple times with a new, randomly selected seed number. Second, models were designed in which the number of indicators per factor was greater than the minimum number of indicators required for model identification (Marshall, Panuzio, & Taft, 2005).

5. Regression results indicated that the model was not multicollinear. Absolute values of correlations of the independent variables are <0.3; estimated regression coefficients were positive. All standardized coefficients were small, and the significance of the regression coefficients was consistent with the significant R2.

6. Goodness of fit was evaluated using: (1) root means square error of approximation (RMSEA) incorporates a penalty function for poor model parsimony (Hu & Bentler, 1999). Values <0.06 suggest adequate fit, whereas values >0.10 signify a poorly fitting model and that the model should be rejected (Kline, 2011). (2) The comparative fit index (CFI) and (3) goodness-of-fit index (GFI) represent incremental fit indices contrasting the hypothesized theoretical model to a more restricted nested baseline model (Kline, 2011). Both ranges from zero to one and values >0.9 are indicative of adequate fit (Tucker & Lewis, 1973).

7. Modification index reflects how much the overall model Χ2 decreases or improves if a constrained parameter is freely estimated. Candidate paths involving conspicuous values (modification indices>10) were then examined for model fit improvement and magnitude of the freed estimated coefficients. The decision to explore and keep new paths followed theoretical meaningfulness.

8. Process adheres to the hierarchical principle advocated in the theoretical model.